you2idea@video:~$ watch pTTvyQDK_tk [1:09:37]
// transcript — 2646 segments
0:02 AI isn't just coming for jobs. It's coming for entire departments. Teams
0:06 that once required dozens of people can now be replaced by something entirely
0:11 new. An AI workforce. These are digital teams that run all day and all night
0:15 that never get tired. They never quit. And they never ask for a paycheck. And
0:18 they're showing up in every department these days from marketing to sales to
0:22 operations and creative. And at my agency, Morningside AI, we implement AI
0:26 within some of the world's biggest sports teams and even publicly traded
0:29 companies. And I've been seeing these kinds of AI workforces just starting to
0:32 take off now. So, I wanted to make a video like this just to open the door to
0:36 others to not only survive, but thrive in this new era of work by being able to
0:40 build AI workforces and not be replaced by them. Because while this shift may
0:43 sound scary, and most people are naturally worried about what's going to
0:47 disappear, the real opportunity lies in what's going to be created. Because for
0:50 the first time in history, you don't need millions and millions of dollars in
0:54 funding or a giant company to build a workforce. You can create one yourself
0:57 without writing a single line of code. You can build entire digital teams made
1:01 up of AI agents that keep working while This isn't even far off science fiction.
1:10 Gartner predicts that 15% of all work decisions will be made by AI agents by
1:15 2028. That's up from almost zero today. So, in the next few years, every
1:18 industry will start to feel this shift. That's why businesses will be seeking
1:21 people who can guide them through this transformation. People like you if you
1:24 stay to the end of this video. So, we've got a lot to cover in this video. As you
1:26 can probably tell, we're going to be starting off with foundations, which is
1:29 breaking down what an AI workforce actually is, how they work, and the core
1:33 concepts that you need to understand to be able to build them. And then we'll
1:36 walk through an entire end-to-end AI workforce build of a personal AI
1:39 assistant that can book meetings for you, prepare presentations, take notes,
1:43 and even assign tasks for you as well. And in the third and final chapter, I'll
1:47 reveal exactly how you can monetize this new skill by selling to businesses who
1:50 are starving for their own teams of AI employees. If you want to learn more
1:53 about me and my story, you can see the behind the scenes of building
1:55 Morningside on my second channel, which is in the description below. But without
2:07 So, we're witnessing the birth of a trillion dollar market here. And the
2:10 good news is that anyone can get a slice of it with a little bit of hard work. Of
2:13 course, there's a reason OpenAI's founder, Sam Alman, and his friends are
2:16 betting on this. There's actually a video of him talking about it. I in my
2:20 little like group chat with my like tech CEO friends, there's this there's this
2:23 betting pool for the first year that there's a uh a oneperson billion dollar
2:28 company, which would have been like unimaginable without AI and now will
2:32 happen. The reason this is even possible now is because of these AI workforces.
2:35 They're not just a shiny new tool. They'll soon be as essential to a
2:38 business as having a website. And companies that don't start to adopt them
2:42 will fall behind fast. And companies who can adopt them quick, they're going to
2:45 gain an immediate competitive advantage. So over the next 12 to 24 months, as we
2:49 see demand for these kinds of workforces increase, businesses will be seeking the
2:53 people like you and I to help build them and manage these workforces. Why are
2:56 these AI workforcees so lucrative? And why will every business need one, too?
2:59 Well, an AI workforce doesn't just save money, it changes the way work gets done
3:03 in a company. Instead of salaries and benefits and vacation time, these
3:07 digital workers just keep going for a fraction of the cost. So instead of the
3:10 ups and downs that come when people having good days and bad days or sick
3:14 days, every task gets delivered at the exact same standard that you built the
3:17 workforce to do every time. And when things start scaling up in a business,
3:19 you don't have to go through a long hiring and training process to bring new
3:22 people on. You can scale the business almost instantly. AI workforces are also
3:26 really fast. So what might take a human team hours or days to coordinate can be
3:30 finished in minutes by AI agents working together. And the more they work, the
3:33 better they get. And so each interaction can become fuel for improvement without
3:36 the need for expensive training costs like you have with staff. And I get it.
3:39 All of this probably sounds quite intimidating. Like the idea that whole
3:42 parts of a business can run themselves might feel unsettling or even impossible
3:46 to achieve. But the truth is that this shift is already underway and it's
3:49 happening whether we feel ready for it or not. That's why it's so important to
3:53 lean into this now to understand what AI agents and work forces are and learn how
3:57 to use them for your own advantage. Because the people who figure out how to
3:59 harness it are the ones who will shape the future and not be pushed aside by
4:02 it. Just think about the kinds of tasks that businesses already pour enormous
4:05 amounts of time and money into. answering customer questions, keeping
4:09 records organized, updating schedules, researching information, writing
4:13 reports, producing content for websites and social media. These are the kind of
4:16 repetitive and draining tasks that every business wrestles with. And they're the
4:19 exact kind of work the AI workforces are designed to take care of. Businesses
4:22 that act first will gain an enormous advantage, and the people who learn to
4:25 assemble these now will become the trusted experts in this industry that is
4:29 just beginning. So, it's the perfect time to get into it. And the good news
4:32 is you don't need to be super technical to build them either. There are now
4:34 platforms like the one we're going to go through in the build section of this
4:37 video that allow you to create agents and tools and workforces and give
4:41 instructions like you would a person. You just describe what you want them to
4:44 do in plain language and the system will literally build it for you and you can
4:47 tweak from there for your needs. So your real job is simply learning how to
4:50 connect these pieces together so that they function like a team. It's kind of
4:53 like building with Lego where each block is simple on its own, but when you
4:56 connect them together in the right way, you can create something much more
5:00 complex. And AI workforces work the same way where you're assembling these
5:02 digital workers who have access to tools and even other apps to operate together
5:06 efficiently. By becoming a workforce builder, the opportunities are wide
5:09 open. You can help companies to spot where these systems fit and build custom
5:12 solutions for them or even package and sell readymade agents and full
5:16 workforces direct to companies or in marketplaces. And later in this video,
5:19 I'll show you exactly how and where to start selling them, even if you're
5:22 completely brand new to the space. But first, we need a clear grasp on what
5:25 we're dealing with and why it matters. So, let's understand how AI work forces
5:34 So if a workforce is composed of AI agents, what exactly is an AI agent? So
5:38 you can think of them as like a digital employee. You can give them a task like
5:42 look this up or write this message or organize this list and then they take
5:45 action for you and they do it quickly. They don't get tired and they're always
5:49 ready. It's one worker with one job and a few tools that allow it to do that job
5:52 properly. So it's kind of like a specialized digital employee. But no
5:56 business runs on just one employee. A shop doesn't just survive with one
5:59 cashier. It needs someone to do the inventory, someone to restock the
6:03 shelves, someone to handle the finances. Each person in the company has their
6:06 role and together they keep the place running. And that's exactly what an AI
6:09 workforce is. It's a team of AI agents, each with its own specialty working
6:13 together like a department inside a company. Workforce that we're going to
6:15 be building in this video looks a bit like this. It all kicks off over Slack,
6:19 which is a business messaging platform where you can send a Slack message,
6:22 schedule a meeting with anyone in your company, or even external contacts. The
6:25 workforce then books the event and it can even prepare a presentation for the
6:29 meeting based on company docs and external research. And then finally, it
6:32 can take notes during the call and assign tasks when the meeting ends. So
6:36 together, this workforce acts like a personal assistant ready to help out
6:39 with the literal click of a button. So this is the shift that's happening right
6:42 now. We're moving from these single AI assistants or agents that only handle
6:46 one-off tasks to these collaborative AI workforces that can take care of entire
6:49 business processes. That's not just helpful, that's ultimately going to be
6:52 world changing. And if you can learn how to build and sell them, then that's
6:55 going to be life-changing for you over the next few years. So, now that you
6:58 know what a workforce is, how are they So, you can visualize an AI workforce
7:07 like an org chart from a typical company with human employees where there are
7:11 specific roles, clear reporting structures, and defined workflows
7:14 between the different positions. And the core principles of human and AI
7:17 workforces are the same. A workforce runs on three pillars. You have
7:21 specialization, collaboration, and coordination. So specialization means
7:24 that each agent has one clear job and gets great at it. For example, you have
7:28 a researcher that pulls information, a writer that turns that into a useful
7:32 plan, a designer then prepares a presentation for it, and so on. But we
7:35 don't want our agents to be working in silos. So you need collaboration where
7:38 agents can work together by passing tasks and information between each
7:41 other. Just like a designer might create a logo and then pass it to an animator
7:45 to animate it. And of course, these work forces need coordination. So, it's our
7:48 job as the workforce builders to create a clear and predictable system
7:52 structured to keep everything organized without error. We can even have agents
7:55 in the workforce that function like a project manager who orchestrates
7:58 everything, deciding what runs and when, making sure everyone is communicating
8:01 well and everything is operating smoothly and of course addressing issues
8:05 when they do happen. Just like a real org chart, AI workforces usually take a
8:09 few different shapes with agents who are working sequentially and completing
8:12 tasks in a straightforward line where each agent depends on the last, like an
8:15 assembly line. You can also have agents that run in parallel, completing the
8:19 task at the same time, which can cut down on the overall time it takes for
8:21 work to complete. Like in human organizations, AI workforces can have a
8:25 hierarchy as well, where assignments flow from the top to the bottom. And
8:28 there can be an orchestrator agent who functions like the team leader who
8:31 understands the goals and then breaks a big job into smaller parts, assigning it
8:34 to these different specialist agents and then pulling the results back into a
8:37 final result like a project manager would. But what enables these agents to
8:41 actually work? What gives them the ability to understand, remember, and
8:44 take action? Well, each agent is powered by a few core components that mirror how
8:47 real employees work. First off, they have a clear understanding of how to do
8:51 their job. For a human employee, that would be your job description, but for
8:54 an agent, it's their prompt. These are the instructions that you give to an
8:57 agent to define its role and its responsibility. It's the blueprint for
9:00 how the work needs to be done. But if you just handed a sculptor a sketch of
9:03 what you needed it to sculpt, but you didn't actually give the actual stone,
9:06 they wouldn't be able to do anything. So, we often need to give our agents
9:10 access to resources or materials like a transcript of a call or a customer's
9:14 order or campaign performance data in order for it to do its job. But of
9:17 course, if we only gave the sculptor a block of marble but nothing to carve it
9:20 with, then it would be powerless. Similarly, we can empower agents by
9:23 giving them access to tools that they can use to execute their duties like a
9:27 web scraper to perform research on a company or a document converter that
9:31 turns text into PDFs. These tools can include integrations with external apps
9:35 like CRM like HubSpot or project management platforms like Notion. This
9:38 way, your agents can interface with the external world and sync up with other
9:41 systems in the company. You can also give agents access to a knowledge base,
9:45 which is essentially a database full of useful knowledge that agents can
9:47 reference as needed. This could be anything from internal company reports
9:51 to a list of frequently asked questions that a customer support agent could
9:54 reference in order to come up with standardized responses to customer
9:57 questions. In addition to your agent being able to know things, they can also
10:00 have the ability to remember things, too. So, by giving an agent the power of
10:03 memory, they can recall what's happened in previous steps or even across time.
10:06 And this not only helps an agent to do its job better now, but allows them to
10:10 improve over time if you set it up correctly. So, circling back to
10:13 collaboration and coordination. When it's time to pass work along from one
10:16 agent to the other, there are different patterns that keep things smooth. Often,
10:19 the handoff is clean, like a baton pass, where one agent finishes the task and
10:22 hands the next agent what it needs, and that next agent takes off running. Other
10:26 times, an agent might have to clear a couple hurdles first, like making sure
10:29 the right file is ready before it can hand things over to the next agent who
10:32 would not be able to do the task if they didn't have that file. And other times,
10:35 this coordination can go both ways where an orchestrator agent might request work
10:38 from an agent it oversees. And then that subordinate agent performs a task and
10:41 hands the results back up to its manager. Then that manager might process
10:44 things and hand it off to another agent who finishes things off. So with all
10:46 that out of the way, we are about to build. So let's quickly recap to make
10:56 So, while these work forces can get complex, ultimately building them is
10:59 really about starting with the simplest piece that works. You break down work by
11:03 specialization, creating unique agents that have a prompt that specifies its
11:06 role and responsibilities. Then, by providing it with all of the necessary
11:10 resources, knowledge, and tools, it has what it needs to perform the tasks
11:13 you've told it to do. Then by connecting these specialized agents into a virtual
11:16 org chart and making handoffs between teammates clear and predictable, you've
11:20 built the foundation for collaboration and coordination that can start small
11:23 and continue to scale. With this essential understanding in place, let's
11:26 see this in action and build our first AI workforce, starting with our first
11:29 agent. So to make building your first AI workforce as smooth as possible for you,
11:32 I have organized all of the prompts and instructions for this workforce over in
11:35 my free school community. So if you haven't yet, feel free to pause this
11:38 video now. You can head down to the first link in the description, join my
11:40 school. It'll take a minute or two for you to get accepted. Then you can go to
11:43 the classroom section where you'll be able to access everything. All the
11:46 prompts, tools, links, everything you To build our workforce, we'll be using
11:59 Relevance, who we're happy to be partnering with on this video. Like it
12:03 says, we can build teams of agents that deliver human quality work. We can even
12:08 invent our agent with a simple prompt. For example, we could tell relevance
12:12 that we want an agent to research a person on LinkedIn and Google, then
12:16 click invent, and it will spin one up for us. We'll take a look at that
12:19 feature later when we're inside the platform and use it along with other
12:24 tools to build out our full AI workforce. If you don't already have a
12:28 relevance account, you can sign up to create one, but I'm going to log into my
12:32 existing one through Google. Here we're looking at the relevance marketplace.
12:37 This is an ecosystem of agents, tools, and entire workforces that the builder
12:41 community can use. Like this image generator agent, for example, which
12:46 generates images using GPT models. We could clone this into our project and
12:50 use it for our needs. There are also agents that can be purchased like this
12:55 Gmail task creation agent. As you can see, we could buy this for 99 and use it
13:00 straight away inside of our projects. So, how do we actually start to create
13:04 our own agents? Well, over here on the left, you can see several tabs. There's
13:08 a tab for agents where we can build or find agents we already built or ones
13:13 that we've cloned or purchased. We can give our agents access to tools that
13:16 empower them to perform their responsibilities and give them access to
13:20 knowledge that provides the context for how to do their job well, such as
13:24 information about your company, your clients, your industry, and more.
13:29 Putting that all together, we can form our workforces. And these workforces and
13:33 the agents within them are empowered by integrations through different APIs.
13:37 That's just a fancy way of saying that we can sync up with other apps out on
13:41 the internet and use their functionality. This includes integrations with apps such as Gmail or
13:48 Google Drive, Google Meet, HubSpot, Slack, or Trello. When we're giving our
13:52 agents functionality, often we're giving them the ability to use external
13:56 integrations with other apps. Here you can see all of the agents that I've
14:01 either built, cloned, or purchased. We can store them in folders to categorize
14:05 them. And they are also grouped by the workforce they belong to. So, we're
14:09 about to create our first agent. But before we do that, I want to show you
14:12 this chat feature, which opens up this new window. What you're looking at here
14:17 is essentially like a regular LLM chat window, like chat GPT. But the cool
14:22 thing is you can prompt within here and add your agents or even entire
14:28 workforces and ask them to run tasks for you. So if I select this gamma
14:32 presentation designer agent, I can tell it to build me a presentation on selling
14:38 AI services to small to mediumsiz businesses. And it's able to do that
14:42 straight from this chat window. Then on the left, we have the chat history that
14:46 we can revisit. Later in the video, I'll show you exactly how to set up this
14:50 Gamma Graphic presentation designer, which is super powerful. Before we start
14:54 building our workforce, I want to orient So, now let's head into the workforce
15:05 tab and open up this meeting workforce cuz this is what we're going to be
15:09 building. At the top here, we have our two triggers. This is how we tell the
15:13 workforce to start. It'll start based on a message it receives. We can send our
15:17 workforce messages from within relevance, like through that chat window
15:20 I just showed you. But we can also trigger it from apps that we use every
15:25 day for work, like Slack. So, we'll set up a way to interact with this workforce
15:29 directly through Slack. And what are we telling it to do? Well, we're requesting
15:33 this workforce to book meetings for us. It all starts off with this orchestrator
15:38 agent that understands what we're wanting it to do, whom we're wanting it
15:42 to invite, and what's required for that meeting. This orchestrator agent has a
15:46 few agents that report to it called sub agents. There is one that finds internal
15:51 participants. These are members of your own team or company. We'll give the
15:55 agent a knowledge base to find the correct participant. We have another
15:59 agent that can find external participants. This one will be looking
16:03 within our integrated HubSpot to find leads uh potential clients of ours and
16:08 then perform research with LinkedIn and Google on that lead and the company they
16:12 work for. This ensures we have enough information about who we're meeting with
16:16 to feel prepared going in. And we also have this gamma graphic presentation
16:20 designer which will run anytime we request a presentation to be made. Maybe
16:24 we want to do a pitch to a lead or discuss internal metrics during a
16:29 company call. This agent can build those presentations for us automatically. If
16:33 you're not familiar with Gamma, it's an AI powered tool that turns your ideas or
16:37 documents into beautifully designed presentation ready slides and web pages
16:42 in seconds. And they now have an API which means we can ask Gamma to create
16:47 these kinds of assets for us from from agents within programs like relevance
16:51 and from other workflow platforms like make or N8N. Once the presentation is
16:55 ready, it will let the meeting orchestrator know. Ultimately, the
16:58 orchestrator's job is to book the meeting. So, it calls this meeting
17:02 booker agent, which uses Google Meet to schedule a meeting with all of the
17:06 required participants. And then it sends that meeting link to our notetaker agent
17:10 who is going to attend the meeting, take notes, transcribe the call, identify any
17:16 next steps or action items, and then create tasks within a to-do list app.
17:19 Finally, it will even send an email summary to all of the participants with
17:23 a link to those tasks. So, as you can see, this is a nice well-rounded
17:28 workforce that functions almost like a personal assistant, booking meetings
17:32 with internal or external participants, preparing presentations, and documenting
17:37 calls with action items for follow-up. So, if you're excited to start building
17:41 it, let's head over to the agents interface and build our first agent by
17:45 clicking on the new agent button. Like I mentioned before, we could invent this
17:50 and describe exactly what we want built. And relevance is going to do its best
17:54 job at building that for us. Or we could use a guided setup. We could even import
17:58 an existing agent from a file that someone shared with you. Or we can build
18:01 it from scratch. Since this is our first agent that we're building, I want to
18:05 make sure you understand exactly how it's built. So we're going to build this
18:16 Now we're inside the agent builder and on the left we can see the prompt. This
18:21 is where we create the guidelines for how the agent should function. We can
18:25 give it tools, knowledge, set up how it's triggered, build in some
18:30 escalations, memory, and variables. We'll touch on some of this in a moment.
18:33 And when we add any of this, it'll show up on the right hand panel over here. So
18:37 let's start and give our agent a name. We'll call it Borealis the meeting
18:41 booker. And below here, we're going to write some instructions. But first, I
18:45 want to bring your attention to the model section here. So, here is where we
18:50 select which AI LLM model we're using. As you can see here, we're using a
18:54 performance optimized model where it just picks the best one for us. We could
18:59 optimize by cost or select a specific one with chat, GPT, Claude, Gemini, or
19:04 whatever best suits our needs. Now that we're clear on which model we're using,
19:09 we can start to write our prompt. Now, there's not necessarily a standard way
19:13 of structuring these prompts, but the way that I like to do it is I'd like to
19:17 start out by defining the agents role. So, in this case, I'm telling it you are
19:22 Borealis and specifying that it is a meeting booker agent. Then, I'll just
19:26 bold its name so it's more easily scannable for the user. And I also want
19:30 to get clear on what inputs this agent should be receiving. So in this case,
19:34 it's going to be receiving information for the participant or participants to
19:39 invite to a meeting. And it's also going to be receiving the time zone of the
19:43 participant and I'll tell it that it may also receive an exact date and time to
19:48 book the meeting. So now we told our agent who it is, what information is
19:52 going to receive. Now we need to define its responsibility, instructing it on
19:56 what it's supposed to do and how to do that. So, we're going to tell it your
20:00 purpose is to schedule meetings and invite each participant. If a specific
20:04 meeting time is provided, book the meeting at that time. Otherwise,
20:08 identify a meeting time that logically is most convenient considering the time
20:12 zones of all participants. Now, I'm going to tell it that when it's
20:15 communicating with the user who is actually triggering it, refer to times
20:20 within their time zone. And this word here, time zone, we're actually going to
20:24 convert into something called a variable. You can think of a variable as
20:29 a placeholder that fills in with whatever value is present at that time.
20:33 So, we're going to go to the right hand panel and set up a new variable. It's
20:37 going to be a text variable. And we'll name the variable time zone. This is for
20:41 the agent to be able to reference what this is called. And we'll describe it as
20:45 the time zone of the user who is requesting the meeting. And below here
20:50 and the input is where we actually put the value that we want the placeholder
20:54 to be replaced by. So the actual time zone such as EDT or Pacific Standard
20:59 Time. And here in the green, we set the actual name that the agent is going to
21:03 be using inside of the prompt to refer to this variable. Now over in the
21:07 prompt, if we add these double curly braces around that variable name, then
21:11 that activates the variable. So it's going to be filled in with the value of
21:16 it, which in this case is EDT. Now, once you are clear on the meeting time,
21:20 schedule the meeting using a tool in order to add a tool into this prompt.
21:24 We're going to go into this right-hand panel and open up this tools modal and
21:28 we'll see we have a bunch of tools to choose from. They have them organized by
21:32 use case and you can also search for a specific tool. So in this case I'm
21:36 searching for a scheduling tool and I see that there's this Google Meet
21:39 scheduling tool that we can add to our prompt here is asking me to fill in the
21:44 missing tool inputs. In this case it needs me to select which Google Meet
21:48 account I want to connect to. Now, I've already connected my account, but if you
21:52 haven't yet, you can click add account and go through this process where
21:56 relevance AI uses pipedream to connect your Google account. I'll select the
22:00 account I already have connected and hit continue. And now here on the right tab
22:04 under tools, you can see I have that Google Meet tool. That means it's ready
22:08 to use within the prompt. So, I can access that by using forward slash. Then
22:13 I'll click tools, select Google Meet, and here is added right into the prompt
22:17 for the agent to use. Finally, I'm going to add some important considerations for
22:21 my agent. Going to tell it to only invite the participants who are
22:25 explicitly mentioned and tell it that we do not want to invite anyone else. And
22:30 just like our agent had an input, I'm also going to give it an output so we
22:33 can clarify what it should actually return when it's complete. So, we'll say
22:37 once you've booked the meeting, notify the user and provide the meeting link.
22:42 And if any errors or issues occur, notify the user. And with that, we
22:46 basically set up our agent. I'm just going to clean things up and add some
22:50 dividers here. If I hit hyphen three times, then it'll create these lines so
22:55 I can create some visual separation between the structure of my prompt. And
22:59 at the very top here, we can give the agent a description. So, of course,
23:04 Borealis schedules meetings and invites participants. We'll make sure to save
23:08 the agent. And this whole time, we've been on the build tab where we've been
23:11 building the agent. But now we want to go into the run tab where we can run the
23:16 agent to see how it's actually working. If we wanted to, we could add a little
23:20 guide to provide instructions for how to use or set up this agent, but this one
23:24 is super simple. So, we will keep that empty. And to run this, we're going to
23:28 tell Borealis to book a meeting with someone. And we'll give it an email for
23:32 whatever time. Let's say Friday at 100 p.m. Then we'll clarify the
23:38 participant's time zone is PDT. Now when we run this agent, we'll see the agent
23:42 working in real time. We can see it booking the meeting and it was a
23:45 success. It's provided us with the details and the link to the meeting
23:49 which we can click and join the meeting when we're ready. Within the timeline,
23:53 we can see the steps that were performed in the background such as using the
23:57 Google Meet tool. And just to prove this worked, I can go to the calendar that it
24:02 added to. And here is that event created for us with our new Borealis agent. So,
24:06 now that we know that it's working like it should, we could go ahead and share
24:10 this. We can make it publicly available, so anyone could run this agent. They
24:14 could embed it somewhere, or we could just share a link to it. We could also
24:18 turn this agent into a chat widget, which we could add to a website
24:21 somewhere. We could change its styling to fit the branding of the site, add a
24:26 starting prompt, such as, "Book me a meeting. Add a message placeholder so
24:30 people know what to actually type into here and how to use it. And set up other
24:35 configurations like allowing file uploads and toggling off the relevance
24:39 branding. Another option would be to make this agent into a clonable template
24:43 so we can share it with other users and they can clone it and adapt it to their
24:47 needs. But of course, we're building out an entire team of agents. So you'll see
24:51 later in the video how to publish an entire workforce out into the world. But
24:55 before we build out our next agent, let's get a better sense for how tools
25:00 work within relevance. Back around the build tab underneath the tool section,
25:04 we can see the tools that our agent has access to. In this case, our agent only
25:08 has access to the Google Meet tool. And what we're looking at here is all of the
25:12 tool inputs, such as the connected Google account and the calendar ID.
25:17 Here, we're letting the agent decide which ID to use, but we can set this
25:21 manually as well. This tool is currently set to run automatically. We could
25:25 require approval for it to run or have the agent decide. And conveniently, we
25:29 can edit the tool whether we created it ourself or we cloned it from the tool
25:34 marketplace. Inside here, we can set up all of the inputs for the tool. We can
25:39 set them as required or optional. Here, this looks pretty familiar. It has all
25:42 of the variables that this tool is making use of. And if we wanted to, we
25:47 could even add additional steps into this tool. Maybe we want to connect it
25:51 to a CRM like HubSpot. So once it schedules a meeting, it creates a note
25:55 within HubSpot relating to the person that it scheduled the meeting with. We
25:59 could go ahead and save the tool or save it as a draft, but because we don't want
26:03 to change it at all, I'm just going to close out of this. So that's the process
26:07 of looking under the hood of an existing tool. But of course, we can create our
26:11 own tools either by default, which means we're creating it from scratch, or we
26:15 could vibe create it, which is a way for us to invent our own tool with a prompt.
26:20 So, let's say we want to build a tool that takes in a transcript of a video or
26:24 a meeting and then produces an engaging post for LinkedIn. Now, once I hit go,
26:29 it's going to build this tool for me. I'm speeding this up for your
26:32 convenience, but as you can see, it walks through all of the steps to invent
26:36 this tool for me in the background. And I configure these inputs such as the
26:41 transcript, style, audience, and focus. So, if I hit run, it's going to generate
26:45 that LinkedIn post for me. And if I pop open the output, I see the actual post
26:50 and just scanning through it is looking pretty good. I could either use this
26:54 tool as is or continue to use the invent feature to refine this. So let's say I
26:58 don't need this post style input. I could go into this prompt and tell it I
27:02 don't need an input for this and just to remove it and it'll go through that
27:06 process and remove it for me. I could continue to make changes in this way, or
27:10 I could even switch to the default builder, which brings me to that view
27:13 that we were looking at before, which is the traditional tool editor, giving me
27:18 hands-on control over how this tool functions. And if I expand this Python
27:21 code, you can see that it did all of this for us in the background. We did
27:26 not have to write a line of Python code. It invented this all for us. Of course,
27:31 we could continue to add on to here. Maybe once the LinkedIn post is
27:35 generated, then we actually post it out onto LinkedIn via API. Now, just like an
27:39 agent has its run tab, the tool has a use tab. So, here we can actually use
27:43 the tool and we can see a log of all the times that we ran that tool to debug or
27:48 just to get a better sense for what With all that understanding in place,
27:57 now let's move on to creating our next agent. We're also going to build this
28:01 one from scratch and we're going to name it Polaris, the participant finder. And
28:05 it's going to find participants from within a knowledge base, which we will
28:09 set up in a second. First, we're going to define its role and tell it that you
28:14 are Polaris, the participant finder agent. Then we'll set up its input and
28:18 tell it that it's going to receive the name or title of an employee from a
28:22 company whose directory you have access to as a knowledge base. And we'll give
28:26 it a responsibility. For each participant, locate their info within
28:30 the company directory. And this is going to be that knowledge base. In order to
28:34 add a knowledge base, we'll go over to the right hand panel and click into
28:38 knowledge. And we can add a knowledge base in a number of ways. So we could
28:42 either sync up our agent to Google Drive or Notion, add an existing knowledge
28:46 base that was already added into our relevance account here, or we can simply
28:52 upload a file. So here I'm adding a CSV file that lives locally on my computer
28:55 and we can choose how to use that knowledge base. We can either add it all
28:59 to the prompt which is good for smaller data sets because when you do this it's
29:03 going to add the entire file and all of its contents directly into the prompt or
29:08 we can allow the agent to search and this is good for larger data sets. Since
29:13 the agent will be able to do a ragbased search on the entire knowledge base, but
29:17 since our company directory is quite small, we're going to add it all
29:21 directly into the prompt. Once it's uploaded, we'll see over in the prompt
29:24 that our knowledge base is added into the prompt here in the brackets. And
29:28 although it doesn't look complete, our agent actually sees it as if it's the
29:32 entire CSV file. Finally, for the output, we'll specify that it must
29:37 return the participants full info and to report any errors or issues that it
29:41 encounters while running. We'll go ahead and save it. Then head to the run tab
29:45 and test it out. So, we'll say find me my head of content. Again, it can take
29:49 in a title or a name and search the knowledge base for that participant. And
29:53 fortunately, we can see that it worked. It found the correct participant, Adam
29:58 Jar, my head of content. Clicking on this task icon over here, we can see all
30:04 of the tasks that our agent has run. We can see a detailed view. We can see if
30:08 there's anything to review, anything that's been escalated or any errors from
30:13 our tasks. We can see the queue if anything is in process, and then the
30:17 list of tasks that have run here. To run a new task, we could either schedule a
30:21 bunch of tasks in bulk, or we could just hit new task. And this allows us to run
30:27 new tasks in this agent. I could even tell it to find me all members of the
30:31 content department because again it has access to that company directory which
30:36 specifies the department that all of these employees work within. So it's
30:40 able to process really dynamic requests and find participants to match those
30:45 requests. Finally, if you wanted to search within the tasks that you ran,
30:49 you could even go and filter for tasks maybe where it's of a certain status. Or
30:53 you could even filter by a search term and pull up only the tasks that match
30:58 that term. Great. So now with all of that context in place and with two
31:02 agents that we've built, we can now head into the workforce tab and start to
31:06 build out our workforce with these new So, we'll go ahead and create a new
31:15 workforce. We'll give it a name. We'll call it meeting setter. And you can
31:18 think of this interface much like a canvas that you can drag different
31:22 elements onto in order to build out your workforce. If I drag on this agent card
31:27 on the right, I have all the agents that I can choose from. So, I'll select the
31:31 meeting booking agent. And as you can see, its prompt is available right here
31:35 in the panel for me to edit as needed. Next, I'll drag on another agent card.
31:39 And for this one, I'm going to be selecting the participant finder. And if
31:44 you remember from earlier, we need a way for these agents to collaborate together
31:49 through a manager or orchestrator agent who can receive meeting requests and
31:52 then delegate the tasks across this workforce. So, let's save our progress
31:57 so far and head out of the workforce so that we can go back to the agents tab
32:01 and create that orchestrator agent, which will build with the invent
32:05 feature. We could give it a very simple prompt here, but I'm actually going to
32:08 be pretty thorough because I want to get this right. So, I'm going to tell it to
32:12 invent an agent called Orion, the orchestrator. And this agent is going to
32:16 receive inbound messages requesting meetings and sometimes with
32:20 presentations. This orchestrator needs to interpret the intent of the message,
32:25 which means the meeting's purpose to require participants and whether a
32:29 presentation is needed. It then will delegate work to sub agents. Polaris to
32:35 find the internal team members, Lania to locate and research external leads.
32:39 We'll build that agent and a couple others soon like Gamma to create the
32:44 presentations. Borealis to schedule the meetings and Nate to attend, transcribe
32:49 and assign follow-ups after those meetings conclude. We need the
32:52 orchestrator to wait for all of the participant data from Polaris and Lania
32:57 before calling Borealis to book the meeting. And then once the meeting is
33:01 booked, we want the orchestrator to notify the user in Slack with a meeting
33:05 link to then continue checking on the gamma agents progress to make sure the
33:10 presentation is ready. And then when the presentation link is ready, send that to
33:14 the user and then of course report any errors or issues clearly back to the
33:18 user. So once we hit start, just like when we invented the tool, it's going to
33:22 go through this process and build out that entire agent for us. Of course,
33:26 I've sped this up for your convenience. We'll just hit accept on the suggestions
33:30 for its name and description. And it's produced this comprehensive prompt for
33:34 us. And just scanning through it, it's looking pretty good, but we're going to
33:38 walk through this step by step soon. So, let's just accept it for now. It's
33:42 asking for a Slack connection because that prompt mentioned it. We'll set that
33:46 up in a moment. So, now we just need to save the agent and we can use it over in
33:50 our workforce. So, let's head back to the workforce tab, open up the meeting
33:54 setter. We'll make some space and then drag an agent card onto the canvas and
33:59 select the orchestrator. Now, if we zoom into the prompt here, we can start
34:03 perfecting it for our needs. I see that its role is a meeting coordination AI.
34:08 I'm going to add the word agent just to be super clear. And it correctly says
34:12 that it receives inbound meeting requests and orchestrates a team of
34:17 specialized sub aents. And the instructions here are that it analyzes
34:21 those meeting requests, coordinates with the sub agents, waits for all the
34:25 participants, and keeps the user informed and handles any errors. So, it
34:29 looks like it got everything that we had requested. It also added in this
34:34 expected input format, but we don't need that. So, I'm going to actually delete
34:39 that out and exchange it for an example input message specifying that these are
34:44 messages from someone within a company wanting to meet with a teammate. the
34:48 full team or people from outside of their company. And then in here, I'll
34:52 just give a few examples of potential messages like scheduling meetings with a
34:57 person with an entire department or someone external from the company to
35:00 pitch them. And I'll add an important note here that for any mentions of
35:05 pitches, proposals, etc. Gamma should prepare a presentation. Now, as we
35:09 scroll down here, we see this sub aent coordination workflow. All these steps
35:13 are good, just like we wanted them to look. I'm just cleaning up the
35:16 formatting here. The only change I would make here is that we want to clarify
35:20 that it is Borealis, the meeting booker, that tells the note
35:26 taker to attend the meeting. It's not the orchestrator himself who does that.
35:31 And that's because Nate, the notetaker agent we will build later, is actually a
35:36 next step after Borealis, the meeting booker, meaning Nate is not a sub agent
35:41 of the orchestrator. We will also remove the Slack tool and just say to use Slack
35:45 because, as you'll see, since we'll be triggering this from Slack, Orion will
35:49 already have access to Slack through that trigger. Now, we'll just clean up
35:53 this formatting so it's easier to read. And because this orchestrator won't have
35:57 access to our notetaker agent, I'll just delete this step number eight, which
36:00 talks about the orchestrator being able to control it, which is not true. And
36:05 here's talking about how it has access to the knowledge bases, but actually its
36:11 sub agents do. So, we'll fix that. And with that, the prompt is ready. But
36:15 you'll notice that there's this warning on this tool. That's because we removed
36:20 the Slack tool from the prompt. So, it's essentially warning us that we have a
36:23 tool that we're not using. So, we'll just delete that out from the tools tab
36:27 and go ahead and save the changes to this agent. Now, we can wire this
36:30 orchestrator up to its sub agent. Starting with Polaris, we'll make sure
36:34 that the connection type is AI connection, which means it's a sub
36:38 agent, and give instructions for how it should be used to find internal meeting
36:43 participants. And then we can set a label here. We'll make sure that it auto runs. We could
36:49 have the approval required or let the agent decide, but we'll keep it on
36:53 automatic. Then we'll wire Orion to Borealis, giving instructions for how to
36:57 use it when the meeting is ready to be booked and set the edge label to book
37:02 meeting. In order to trigger Orion, we need to wire him up with this message
37:05 trigger. And we can test this out from the run tab. We'll make sure to save and
37:10 publish the workforce and then give it a message to book a meeting with my head
37:14 of content for Friday. And in the timeline here, Orion is working. It's using Polaris to find the
37:21 participant. is using Borealis to book the meeting with a Google Meet
37:25 scheduling tool. And voila, it's successfully worked and booked that
37:28 meeting with the right person. So, our workforce is in a really good spot so
37:31 far. If we head back to the build section, we can add another way to
37:35 trigger our workforce from outside of the relevance platform.
37:39 Specifically, we want to be triggering it from Slack. We already have our Slack
37:44 connected here. If you did not already, you could just connect it through here.
37:48 just put in your Slack organization here and go through the connection process.
37:52 And if you're not familiar with Slack, it's really a collaborative
37:55 communication environment for teams to use. So, I'll hit continue and then I'll
37:59 add a keyword here. I'll call it booker. The keyword is really just a word that
38:03 we use whenever we trigger this. So, that relevance knows which workforce we
38:07 want to run. Now, I'm just telling it where it can be triggered from from my
38:11 direct messages and all of these channels. and I'll confirm that I understand that
38:17 I always need to tag at relevance AI within Slack to use this trigger. If I
38:21 wanted to, I could enable specific working hours for this Slack trigger and
38:25 therefore for this workforce, but I don't want to limit it to specific
38:29 hours. Now, I'll just connect up this trigger to Orion, but now be able to
38:33 save the workforce, then head over into my Slack organization and make sure that
38:39 I have the relevance tool installed. So, I'll click on apps and make sure I
38:44 have it installed. If you don't yet have it installed, you'll want to open the
38:48 marketplace. Search for the relevance tool and then install it from here. But
38:52 since it's already available within my Slack, I should be able to go into my DM
38:58 and then say at relevance AI and then book it. That's that keyword that we set
39:02 up and say book me a meeting with my head of content for let's say Monday
39:07 morning and now over in relevance. That should be working. It's going to be
39:11 taking a while, so I'll speed this up for you, but eventually it's going to
39:14 give you a reply back. If you pop that open, it should say something like the
39:19 meeting was successfully scheduled and here is the meeting link.
39:24 Now, if we head back over into relevance and click on the run tab, we can see
39:29 here is that task which we can tell was triggered by myself from Slack. Now, as
39:33 I've been alluding to, sometimes our meetings are going to require us to
39:37 prepare presentations, pitches, proposals, and for that we're going to
39:47 So, let's head over to the relevance marketplace to get started with that. As
39:51 you'll see here, if we search for gamma, there are a couple agents that have
39:55 already been built that we can make use of. We're going to uh select the gamma
39:59 assistant here and go ahead and clone it. and we'll give it a more specific
40:03 name for our usage and call it gamma the presentation preparer. Now, when we save
40:07 it, we'll see that it's asking us to fill in this variable for the gamma API
40:12 key. I've already done that and I'll show you how to do it in a moment. So, I
40:16 can hit continue, then head back into the workforce and put it to work within
40:20 here. So I will drag on a new agent card and select the gamma agent and wire it up as
40:26 a sub agent to the orchestrator and give it instructions for how to use it where
40:31 we'll call gamma whenever a presentation is requested. Now we'll change this
40:34 label to something more informative such as prepare presentation. And now we're
40:39 set up to request presentations to be made directly from within this
40:42 workforce. But before we actually do that, I want to orient you to the Gamma
40:46 platform. We're happy to be a partner of theirs because they provide a powerful
40:51 way to bring your ideas to life where you can use prompts to create
40:55 presentations, branded documents, social media content, and even websites. And
41:01 now with their new API feature, we can automate the creation of this content
41:06 from places like relevance n make or even your own custom web apps. For
41:10 example, you could have a system where whenever a blog post is approved for
41:14 publication, Gamma runs automatically, creating a graphic that is perfectly
41:18 suited for that post. So, the possibilities are really endless here,
41:22 and it can save you and your team a bunch of time. Of course, in our
41:26 workforce, we're specifically interested in creating presentations, and Gamma is
41:30 capable of creating really any kind of presentation from a pitch deck to
41:34 something more personal to an internal team update to a keynote speech or
41:38 whatever kind of slide deck you need created. It's going to do it very fast
41:43 with little or no effort from you. And you can use AI to refine and polish the
41:47 presentation until it's ready to present. As you can see in their video
41:51 here, it all starts with a prompt which you can feed in from the actual Gamma
41:56 interface or programmatically from whatever workflow or app you're using.
42:00 Then Gamma is going to generate the entire presentation for you styled in a
42:05 theme of your choosing or you can create your own theme based on your branding.
42:09 Once it's ready, you can edit the presentation with AI, changing up the
42:15 verbiage or the layout itself. And you can drag and drop different elements
42:18 onto the slides until you're ready to share it publicly and present it out
42:23 into the world. If you don't yet have an account, you'll want to create one now.
42:27 But I'm going to log into my existing one cuz I want to show you some
42:29 presentations I've already generated from the API. Let's open one of these
42:33 up. And you'll see I have all of these slides. The text is based on internal
42:39 documents that I gave it access to. And it generated all of the images and
42:44 design itself based on the prompt. And if I wanted to change things like this
42:48 image here, I could regenerate some new AI images and change what I'm using
42:53 within the presentation. So, not only is it quick to generate, but it's quick to
42:57 iterate as well. And all of my slides here are using the theme Oasis. So, I'm
43:02 excited to show you how to create these from within the relevance platform. Now,
43:05 it's important to note that in order to use the API, you do have to have a pro
43:10 plan within Gamma. But when you consider all the time and energy you'll be saving
43:14 with a tool like this, it provides a lot of value. So, with that pro plan, you'll
43:18 be able to go into the settings here and generate an API key to use within the
43:22 relevance platform or wherever else you might want to make use of Gamma. You'll
43:27 just make sure that the gamma API tool in the agent has access to that API key
43:31 which you can set up within the integrations tab of relevance and add
43:36 that API key inside of here. So now that we're oriented to the power of gamma and
43:41 have it integrated into our agent, let's get clear on how this agent is
43:44 functioning. So under the core instructions, you can see this
43:47 documentation variable. So if I go to the variables tab, we'll see that we're
43:52 feeding our agent. uh all of this uh documentation that explains how to use
43:56 the gamma API and it even includes all of the code for an example request.
44:03 Note here that the theme name is Oasis like I showed you in the gamma platform
44:07 and the number of cards in the presentation is 10. These are all things
44:10 that we have control over from the prompt itself. The only thing I'm
44:14 changing here is I'm going to increase the duration of the delay and the amount
44:18 of times we're going to pull for the presentation to make sure that our
44:22 workforce can gain access to it and it doesn't time out too early before the
44:26 presentation is ready. And since we want to give Gamma access to some internal
44:31 documents to build presentations from, we're going to add a knowledge base and
44:34 sync up to Google Drive. We'll just select our connected Google account,
44:39 specify the drive itself, and whatever document we want to give access to. In
44:44 this case, it's information about how my agency provides AI transformation
44:49 partnership to clients. Since that's not a huge file, we will just add it all
44:53 into the prompt itself. In order for the agent to make proper use of this, we
44:57 just need to let it know that when it's researching and preparing for the
45:01 presentation, it should ask itself, does it need access to the AI transformation
45:06 partnership information? If so, it should reference that knowledge base
45:09 which is attached to the bottom of this prompt. So, if I scroll down here, there
45:14 it is. In order to empower even more, we're going to let the agent know that
45:17 as needed, it can perform Google research to search the meeting
45:21 attendees, the company topics, etc. and it can use a tool to scrape the content
45:26 from a website to discover helpful information about the website of the
45:30 person or company that is building a presentation for it. And with all of
45:35 that set up, our agent is empowered to prepare a presentation on a bunch of
45:39 different topics. If anything went wrong, let's say an AWS outage causes
45:44 gamma to not be functioning temporarily, we can set up what is called an
45:47 escalation, which means the agent will notify us via a method of our choosing
45:52 that there are issues that need addressing. So, we can set up our agent
45:57 to notify us via Slack whenever a task has timed out, for example, or there's
46:02 an unreoverable error or it's exhausted, it's retries. In any of these scenarios,
46:06 we can make sure to be notified within our connected Slack account at the
46:10 destination of our choosing. For example, through a direct message to
46:14 myself. We could also choose to be notified via email as well. So, we'll go
46:19 ahead and save this agent and make sure that it's working by heading over to the
46:23 run tab, saving our changes to the workforce, and asking our Gamma agent to
46:28 build a presentation on how Morningside AI can be an AI transformation partner
46:33 to this example company. I'll give it a URL. Now, as soon as I hit go, we can
46:38 see Orion has called the gamma agent. It's doing its research and planning,
46:42 which looks pretty thorough. It's thinking through how to structure
46:46 the presentation and planning how it's going to make use of the gamma API
46:51 itself. As it's calling that API, it looks like it ran into an error first,
46:55 but it keeps trying using delays during those attempts. And of course, I'm
46:59 speeding this up for you. and it eventually found success and it returned
47:03 this message that the presentation is ready including a summary of what it
47:08 covers and a link to the presentation itself. So if we click that open we can
47:13 view what it built for us which is already looking great. If we wanted to
47:19 we could tweak this and then on the day of open the presentation link and
47:22 present it live. So, this is a super quick and efficient way to get custom
47:27 pictures, presentations, and even proposals generated for anyone that
47:30 you're meeting with. Speaking of which, let's head back into relevance and
47:34 create our next agent, which is going to be responsible for locating our leads
47:38 within a connected customer relationship So, let's start building the next agent,
47:49 the lead locator. This agent is going to help us locate external participants so
47:54 that we can book meetings with people from outside of our company. These could
47:57 be people we want to pitch our business to and turn into clients. Our leads are
48:02 going to live in a tool called HubSpot. So, we'll switch over to that. Now, if
48:06 you haven't heard of it yet, HubSpot is a customer relationship management
48:11 platform or a CRM where businesses store all their customer lead information. You
48:15 can think of it as a central database for everyone that you're doing business
48:19 with or people that you want to do business with. You'll see leads
48:23 organized under contacts, companies, deals, tickets, and orders. But we're
48:27 mostly concerned about our contacts here. This is where the lead's
48:32 information will be stored, including their name, email, phone, and their
48:37 company. These fields are actually linked to their full company data, which
48:42 we can view as a list over in this company's tab. I'm walking you through
48:46 this because our lead locator agent will search through HubSpot and retrieve the
48:50 contact information for the leads we want to book meetings with. Now that you
48:53 understand what exactly we're asking for, let's go back to relevance and
48:57 start building out that agent who will be asking for these leads. We'll build
49:01 the agent from scratch. I'm choosing to name this one Lania the lead locator.
49:05 And we're going to give it a short description to locate external
49:10 participants from outside of our company. Okay. and we'll come down here
49:15 to the prompt and define its role, telling it who it is and specifying that
49:20 it's a sub agent of Orion, our orchestrator agent. Now, we'll tell it
49:24 what to expect as an input, which will be information about the lead, like
49:28 their email and the company they work for. And for added context, we'll tell
49:32 it that this is a participant of a future meeting. As for its
49:36 responsibility, we'll instruct it to locate the info with HubSpot. We can
49:41 access tools by typing a forward slash then the kind of tool we're looking for.
49:45 In this case, HubSpot. We have a few options to choose from. So, we'll select
49:49 the tool to retrieve contact details from HubSpot. Since this is an external
49:53 integration, we need to connect to our account so that we can have access to it
49:57 from within this agent. Now, let's move down a little bit here and continue its
50:00 instructions, letting our agent know that we want it to perform research on
50:04 this lead and their company. And then we're going to the next tool. We're
50:07 going to use LinkedIn. will use this LinkedIn tool to search for info from
50:12 their personal and company profiles. For more thorough research, we're also going
50:16 to use Google search. So, we'll add in a Google search tool here as well. This
50:20 research will be used for things like creating personalized pictures and
50:24 eventually proposals. Finally, we'll define the agents output and tell it
50:28 that it must return the lead's complete information and a research summary to
50:32 its boss, agent, Orion. And of course, it should let us know if there are any
50:36 errors or issues with locating the lead. Okay. And that's our prompt. By giving
50:41 it these tools, we've connected it to HubSpot and enabled it to search through
50:45 both LinkedIn and Google. So, it's pretty powerful with only a few lines of
50:49 prompt. I'll make sure to save this agent, then go into the run tab to test
50:53 it out to ensure everything is working. From this run tab, I'll pass in the
50:57 contact info for a potential lead. In this case, someone in my network that I
51:01 added into my HubSpot contacts. This way I can make sure this agent is not only
51:06 finding the correct contact but doing accurate research on them. Once I enter
51:09 the email you can see it started running. If you're ever wondering if the
51:14 agent is actually working on the right hand side here you can see the status
51:18 will update to running and it also shows the tools that it's going to use. So we
51:22 can see it is definitely working. It's looking into HubSpot. Then it's grabbing
51:27 info from LinkedIn profile. Now we can see it's searching Google. Looks like
51:30 the first one failed, but that's okay because it ran multiple searches, which
51:35 is great to see because that means this agent, like all agents you build in
51:40 relevance, is able to adapt on the fly. And once it completes, we can scroll up
51:44 here and see we have the correct contact details from HubSpot. It's grabbed a
51:48 bunch of details from LinkedIn about this uh lead and his company and
51:53 structured it out quite well, too. This is cool because our other agents like
51:56 Gamma will now have access to all of this context as it builds out super
52:01 custom presentations. For good measure, I'll mark this as complete. Now that
52:05 it's built and we've confirmed it works, we're ready to add this agent into our
52:09 workforce. So, we'll head out of the agent and head back to the workforces
52:13 tab and go into our meeting setter and drag another agent block onto the canvas
52:17 and choose Lania. We'll connect Orion down to Lania and make sure it's set as
52:22 an AI connection since this is another sub agent of Orion who will of course
52:26 call her to locate external participants. Now, I'll just spread them
52:29 out a little bit here to clean up the canvas. And as a final step, I'll update
52:33 the label here to find external participants. I'll make sure to save the
52:37 workspace because we're going to head out of here and go find our next and
52:42 final agent, the notetaker agent, who As we've seen, the marketplace
52:52 conveniently has a bunch of existing agents we can repurpose for our needs.
52:56 So, if we search for a notetaker, we'll see a few options. Nate fits our needs
53:01 quite well, so we'll select him and go ahead and clone it. Its current prompt
53:05 is pretty useful, so we don't need to start this one from scratch, and we can
53:09 reuse some of what is already here. But, we do need to change a few things. And
53:14 we'll start by clarifying its role is a notetaker agent who attends meetings, transcribes what
53:20 happens during that meeting and afterwards assigns tasks and sends a
53:24 meeting summary. As for its input, that will be information about the meeting
53:28 the agent must attend received from Borealis. The meeting booker agent will
53:32 tell it that it's responsible for generating summaries of all types of
53:36 meetings from internal team syncs to client calls and project updates or
53:41 strategy sessions. Inside of its instructions, the agent is being told to
53:45 use this send meeting bot tool to record and transcribe the call. And then when
53:49 the agent receives that transcript, it reads it carefully and writes a summary
53:52 of that meeting, including things like key discussion points, decisions made,
53:57 and action items. Then it's able to use this turn transcript into text tool to
54:02 generate a file of the transcript. And we're telling the agent to refer to the
54:06 transcript whenever a user asks questions about that meeting. I'll add a
54:09 bit more to the prompt here so the agent knows to determine the action items from
54:14 that meeting and then create tasks based on those actions items and create tasks
54:19 for them over in our to-do list in Trello. If you're not familiar with it,
54:24 Trello is a visual task management tool. Uh, and it's used to organize personal
54:28 and professional projects. It's kind of like a digital bulletin board with
54:32 sticky notes. We find those boards in this tab where we've got out to-do list
54:36 board here. Opening up that board, you can see that it has lists arranged in
54:41 columns, and these lists get tasks or cards added into them. You can name the
54:45 lists whatever you want or add new ones and then drag cards across them as your
54:50 tasks move through different statuses. Each card can be opened up and contains
54:54 a bunch of properties you can set up. Each one of our tasks that the notetaker
54:59 agent creates for us will be a new card, which is added to this to-do list board.
55:03 So, back in our agents prompt, we're telling it we want to use the Trello
55:07 tool to create a card on a Trello board. Here, we need to create a Trello
55:10 connection, and we're just going to link it through to our Trello account. We
55:15 need to authenticate with Trello. Just scroll down here, and we'll just allow
55:20 that connection, and then click continue and continue again. We scroll back down
55:25 and now we can see that's enabled. And now we can continue. When it creates the
55:30 card, it should fill in the following: the card name, its description, and its
55:35 due date. That's it. Finally, for the output, we're going to tell it to email
55:39 a summary to all participants using their emails they attended the meeting
55:43 with and include a link to the Trello board that we just added tasks to. And
55:48 it will do this using the send Gmail tool from our connected account so it
55:52 knows how to structure the email. We'll give it a format to follow, which I'll
55:57 just paste in here. and we'll include a sample email as well.
56:04 Great. So, we'll save that agent. Lastly, we need to add a trigger. So,
56:07 we're going to add relevance meeting bot and click setup trigger. And that's been
56:11 added in. Noteaker agents are a little different and need to be triggered
56:15 twice. Once to join the meeting, which in this case, the Borealis agent will
56:19 do, and once when the meeting is finished. In order to trigger it for the
56:22 second time, you need to add this meeting bot trigger to the agent. Now we
56:27 can add it to our workforce. So let's head back over there. Inside our
56:32 workforce, I'll drag on a new agent. Tell it to use Nate the note taker. And
56:36 for the handoff type, this one should be next step. This means that instead of
56:41 the orchestrator agent managing this agent, Nate runs automatically once
56:46 Borealis books the meeting and hands him the meeting link. For good measure,
56:49 we'll edit the label here and specify that we're sending a meeting link on
56:53 this handoff. Now, we're good to go. So, Okay, we've built all of the agents and
57:04 now we can see them working together as a workforce. So, let's jump over to
57:08 Slack. This is where we're going to trigger the workforce. Call the relevant
57:12 AI bot. And then our particular trigger also needs this trigger here. So, we're
57:17 going to say booker. I'm going to say book me a meeting with Adam. And we'll put
57:24 Adam's email in here. next Tuesday at 10:00 a.m. Eastern Standard Time and create a presentation
57:30 about how we can be their AI transformation partner and share it as a
57:34 link in the meeting. Okay. So, we're going to click to send that and straight
57:38 away it responds. So, we can look at that and view thread. So, we've got that
57:41 confirmation here that the agent is cooking up a reply. So, let's jump over
57:46 to relevance AI. We will have a look at the run tab and we can see that has
57:50 already started here and it's already running. Lania the lead locator is
57:54 working and then we are getting the LinkedIn details and we're doing the
57:59 Google search and then book the meeting. Okay, now moving on to creating the
58:09 Now that's completed. So let's jump over to Slack again and we'll just look at
58:13 the reply. So it's created how Morningside AI can be this company's AI
58:18 transformation partner exactly as we asked it to do. So, it's gone through
58:23 and created this detailed presentation. Might need a little bit of a review and
58:26 then edit, but that's certainly a good starting point. Okay. So, let's close
58:31 that. Now, let's jump over to Google Calendar. And we can see that the
58:36 appointment here we need to say yes, we are going to attend. So, let's join with
58:40 Google Meet. Inside Google Meet here, we can see Nate wanting to join. So, I'll
58:44 admit him. And there he is taking all of the notes that I need. He will write it
58:49 all down, email it to us, and assign tasks as we will see later. After the
58:53 meeting, let's head back to relevance. We've seen that complete. We can jump
58:58 over to Trello, and we can see we've got a couple of tasks here that have been
59:01 created, and we can jump up to our email. There's a summary with the key
59:05 points, the decisions made, and then action items and next steps. This was
59:09 sent to all participants automatically after the meeting ended. So, there you
59:14 have it. The workforce works flawlessly. Now that our workforce is complete, we
59:23 can submit it to the relevance marketplace and look at how we can start
59:28 to monetize this. So under more actions, we'll click submit to marketplace and
59:31 select some categories that our workforce relates to, such as sales,
59:36 research, and operations. Then we'll add a description for the workforce, telling
59:40 the public that this workforce helps you book meetings with your teammates and
59:44 contacts and generates custom crafted presentations to use during those calls.
59:49 Then we'll decide if we want to submit this as a free or paid workforce. I'm
59:53 going to submit mine as free, but of course you can set yours as paid and
59:56 start to earn some money. You could allow sharing it as a clonable template
60:00 or allow republishing, but I'll leave those off for now. Then on the next
60:04 step, we just select one of our past tasks to be the preview that shows up
60:08 when this is published. Then once we submit for review, it's going to run
60:12 through some automated checks that I'm speeding up for you. And once those
60:16 checks have passed, you can click finish and you should see your builder
60:20 dashboard where the approval status is hopefully pending. If it was
60:24 autorejected, you might have to resubmit it with some changes. Clicking this
60:28 dropown, you'll see all of the agents and tools that this workforce was
60:32 submitted with. And if we click on the submission itself, we'll get a preview
60:36 of how this would show up within the relevance marketplace. As a builder, you
60:41 have your own profile. You can set up a cover photo and add an image of
60:45 yourself. And in order to receive payments on the platform, you'll just
60:48 link a Stripe account. You'll click generate Stripe link. If you don't have
60:52 a Stripe account, you'll want to create one first and then just enter your
60:56 Stripe email address and continue the process from here. And with that, you've
61:00 taken the first step of monetizing your new skills as an AI agent workforce
61:05 builder. There are so many opportunities for making money with these new skills.
61:09 So, join me in the next and final chapter where we talk about the
2:07 So, we're witnessing the birth of a trillion dollar market here. And the
2:10 good news is that anyone can get a slice of it with a little bit of hard work. Of
2:13 course, there's a reason OpenAI's founder, Sam Alman, and his friends are
2:16 betting on this. There's actually a video of him talking about it. I in my
2:20 little like group chat with my like tech CEO friends, there's this there's this
2:23 betting pool for the first year that there's a uh a oneperson billion dollar
2:28 company, which would have been like unimaginable without AI and now will
2:32 happen. The reason this is even possible now is because of these AI workforces.
2:35 They're not just a shiny new tool. They'll soon be as essential to a
2:38 business as having a website. And companies that don't start to adopt them
2:42 will fall behind fast. And companies who can adopt them quick, they're going to
2:45 gain an immediate competitive advantage. So over the next 12 to 24 months, as we
2:49 see demand for these kinds of workforces increase, businesses will be seeking the
2:53 people like you and I to help build them and manage these workforces. Why are
2:56 these AI workforcees so lucrative? And why will every business need one, too?
2:59 Well, an AI workforce doesn't just save money, it changes the way work gets done
3:03 in a company. Instead of salaries and benefits and vacation time, these
3:07 digital workers just keep going for a fraction of the cost. So instead of the
3:10 ups and downs that come when people having good days and bad days or sick
3:14 days, every task gets delivered at the exact same standard that you built the
3:17 workforce to do every time. And when things start scaling up in a business,
3:19 you don't have to go through a long hiring and training process to bring new
3:22 people on. You can scale the business almost instantly. AI workforces are also
3:26 really fast. So what might take a human team hours or days to coordinate can be
3:30 finished in minutes by AI agents working together. And the more they work, the
3:33 better they get. And so each interaction can become fuel for improvement without
3:36 the need for expensive training costs like you have with staff. And I get it.
3:39 All of this probably sounds quite intimidating. Like the idea that whole
3:42 parts of a business can run themselves might feel unsettling or even impossible
3:46 to achieve. But the truth is that this shift is already underway and it's
3:49 happening whether we feel ready for it or not. That's why it's so important to
3:53 lean into this now to understand what AI agents and work forces are and learn how
3:57 to use them for your own advantage. Because the people who figure out how to
3:59 harness it are the ones who will shape the future and not be pushed aside by
4:02 it. Just think about the kinds of tasks that businesses already pour enormous
4:05 amounts of time and money into. answering customer questions, keeping
4:09 records organized, updating schedules, researching information, writing
4:13 reports, producing content for websites and social media. These are the kind of
4:16 repetitive and draining tasks that every business wrestles with. And they're the
4:19 exact kind of work the AI workforces are designed to take care of. Businesses
4:22 that act first will gain an enormous advantage, and the people who learn to
4:25 assemble these now will become the trusted experts in this industry that is
4:29 just beginning. So, it's the perfect time to get into it. And the good news
4:32 is you don't need to be super technical to build them either. There are now
4:34 platforms like the one we're going to go through in the build section of this
4:37 video that allow you to create agents and tools and workforces and give
4:41 instructions like you would a person. You just describe what you want them to
4:44 do in plain language and the system will literally build it for you and you can
4:47 tweak from there for your needs. So your real job is simply learning how to
4:50 connect these pieces together so that they function like a team. It's kind of
4:53 like building with Lego where each block is simple on its own, but when you
4:56 connect them together in the right way, you can create something much more
5:00 complex. And AI workforces work the same way where you're assembling these
5:02 digital workers who have access to tools and even other apps to operate together
5:06 efficiently. By becoming a workforce builder, the opportunities are wide
5:09 open. You can help companies to spot where these systems fit and build custom
5:12 solutions for them or even package and sell readymade agents and full
5:16 workforces direct to companies or in marketplaces. And later in this video,
5:19 I'll show you exactly how and where to start selling them, even if you're
5:22 completely brand new to the space. But first, we need a clear grasp on what
5:25 we're dealing with and why it matters. So, let's understand how AI work forces
5:34 So if a workforce is composed of AI agents, what exactly is an AI agent? So
5:38 you can think of them as like a digital employee. You can give them a task like
5:42 look this up or write this message or organize this list and then they take
5:45 action for you and they do it quickly. They don't get tired and they're always
5:49 ready. It's one worker with one job and a few tools that allow it to do that job
5:52 properly. So it's kind of like a specialized digital employee. But no
5:56 business runs on just one employee. A shop doesn't just survive with one
5:59 cashier. It needs someone to do the inventory, someone to restock the
6:03 shelves, someone to handle the finances. Each person in the company has their
6:06 role and together they keep the place running. And that's exactly what an AI
6:09 workforce is. It's a team of AI agents, each with its own specialty working
6:13 together like a department inside a company. Workforce that we're going to
6:15 be building in this video looks a bit like this. It all kicks off over Slack,
6:19 which is a business messaging platform where you can send a Slack message,
6:22 schedule a meeting with anyone in your company, or even external contacts. The
6:25 workforce then books the event and it can even prepare a presentation for the
6:29 meeting based on company docs and external research. And then finally, it
6:32 can take notes during the call and assign tasks when the meeting ends. So
6:36 together, this workforce acts like a personal assistant ready to help out
6:39 with the literal click of a button. So this is the shift that's happening right
6:42 now. We're moving from these single AI assistants or agents that only handle
6:46 one-off tasks to these collaborative AI workforces that can take care of entire
6:49 business processes. That's not just helpful, that's ultimately going to be
6:52 world changing. And if you can learn how to build and sell them, then that's
6:55 going to be life-changing for you over the next few years. So, now that you
6:58 know what a workforce is, how are they So, you can visualize an AI workforce
7:07 like an org chart from a typical company with human employees where there are
7:11 specific roles, clear reporting structures, and defined workflows
7:14 between the different positions. And the core principles of human and AI
7:17 workforces are the same. A workforce runs on three pillars. You have
7:21 specialization, collaboration, and coordination. So specialization means
7:24 that each agent has one clear job and gets great at it. For example, you have
7:28 a researcher that pulls information, a writer that turns that into a useful
7:32 plan, a designer then prepares a presentation for it, and so on. But we
7:35 don't want our agents to be working in silos. So you need collaboration where
7:38 agents can work together by passing tasks and information between each
7:41 other. Just like a designer might create a logo and then pass it to an animator
7:45 to animate it. And of course, these work forces need coordination. So, it's our
7:48 job as the workforce builders to create a clear and predictable system
7:52 structured to keep everything organized without error. We can even have agents
7:55 in the workforce that function like a project manager who orchestrates
7:58 everything, deciding what runs and when, making sure everyone is communicating
8:01 well and everything is operating smoothly and of course addressing issues
8:05 when they do happen. Just like a real org chart, AI workforces usually take a
8:09 few different shapes with agents who are working sequentially and completing
8:12 tasks in a straightforward line where each agent depends on the last, like an
8:15 assembly line. You can also have agents that run in parallel, completing the
8:19 task at the same time, which can cut down on the overall time it takes for
8:21 work to complete. Like in human organizations, AI workforces can have a
8:25 hierarchy as well, where assignments flow from the top to the bottom. And
8:28 there can be an orchestrator agent who functions like the team leader who
8:31 understands the goals and then breaks a big job into smaller parts, assigning it
8:34 to these different specialist agents and then pulling the results back into a
8:37 final result like a project manager would. But what enables these agents to
8:41 actually work? What gives them the ability to understand, remember, and
8:44 take action? Well, each agent is powered by a few core components that mirror how
8:47 real employees work. First off, they have a clear understanding of how to do
8:51 their job. For a human employee, that would be your job description, but for
8:54 an agent, it's their prompt. These are the instructions that you give to an
8:57 agent to define its role and its responsibility. It's the blueprint for
9:00 how the work needs to be done. But if you just handed a sculptor a sketch of
9:03 what you needed it to sculpt, but you didn't actually give the actual stone,
9:06 they wouldn't be able to do anything. So, we often need to give our agents
9:10 access to resources or materials like a transcript of a call or a customer's
9:14 order or campaign performance data in order for it to do its job. But of
9:17 course, if we only gave the sculptor a block of marble but nothing to carve it
9:20 with, then it would be powerless. Similarly, we can empower agents by
9:23 giving them access to tools that they can use to execute their duties like a
9:27 web scraper to perform research on a company or a document converter that
9:31 turns text into PDFs. These tools can include integrations with external apps
9:35 like CRM like HubSpot or project management platforms like Notion. This
9:38 way, your agents can interface with the external world and sync up with other
9:41 systems in the company. You can also give agents access to a knowledge base,
9:45 which is essentially a database full of useful knowledge that agents can
9:47 reference as needed. This could be anything from internal company reports
9:51 to a list of frequently asked questions that a customer support agent could
9:54 reference in order to come up with standardized responses to customer
9:57 questions. In addition to your agent being able to know things, they can also
10:00 have the ability to remember things, too. So, by giving an agent the power of
10:03 memory, they can recall what's happened in previous steps or even across time.
10:06 And this not only helps an agent to do its job better now, but allows them to
10:10 improve over time if you set it up correctly. So, circling back to
10:13 collaboration and coordination. When it's time to pass work along from one
10:16 agent to the other, there are different patterns that keep things smooth. Often,
10:19 the handoff is clean, like a baton pass, where one agent finishes the task and
10:22 hands the next agent what it needs, and that next agent takes off running. Other
10:26 times, an agent might have to clear a couple hurdles first, like making sure
10:29 the right file is ready before it can hand things over to the next agent who
10:32 would not be able to do the task if they didn't have that file. And other times,
10:35 this coordination can go both ways where an orchestrator agent might request work
10:38 from an agent it oversees. And then that subordinate agent performs a task and
10:41 hands the results back up to its manager. Then that manager might process
10:44 things and hand it off to another agent who finishes things off. So with all
10:46 that out of the way, we are about to build. So let's quickly recap to make
10:56 So, while these work forces can get complex, ultimately building them is
10:59 really about starting with the simplest piece that works. You break down work by
11:03 specialization, creating unique agents that have a prompt that specifies its
11:06 role and responsibilities. Then, by providing it with all of the necessary
11:10 resources, knowledge, and tools, it has what it needs to perform the tasks
11:13 you've told it to do. Then by connecting these specialized agents into a virtual
11:16 org chart and making handoffs between teammates clear and predictable, you've
11:20 built the foundation for collaboration and coordination that can start small
11:23 and continue to scale. With this essential understanding in place, let's
11:26 see this in action and build our first AI workforce, starting with our first
11:29 agent. So to make building your first AI workforce as smooth as possible for you,
11:32 I have organized all of the prompts and instructions for this workforce over in
11:35 my free school community. So if you haven't yet, feel free to pause this
11:38 video now. You can head down to the first link in the description, join my
11:40 school. It'll take a minute or two for you to get accepted. Then you can go to
11:43 the classroom section where you'll be able to access everything. All the
11:46 prompts, tools, links, everything you To build our workforce, we'll be using
11:59 Relevance, who we're happy to be partnering with on this video. Like it
12:03 says, we can build teams of agents that deliver human quality work. We can even
12:08 invent our agent with a simple prompt. For example, we could tell relevance
12:12 that we want an agent to research a person on LinkedIn and Google, then
12:16 click invent, and it will spin one up for us. We'll take a look at that
12:19 feature later when we're inside the platform and use it along with other
12:24 tools to build out our full AI workforce. If you don't already have a
12:28 relevance account, you can sign up to create one, but I'm going to log into my
12:32 existing one through Google. Here we're looking at the relevance marketplace.
12:37 This is an ecosystem of agents, tools, and entire workforces that the builder
12:41 community can use. Like this image generator agent, for example, which
12:46 generates images using GPT models. We could clone this into our project and
12:50 use it for our needs. There are also agents that can be purchased like this
12:55 Gmail task creation agent. As you can see, we could buy this for 99 and use it
13:00 straight away inside of our projects. So, how do we actually start to create
13:04 our own agents? Well, over here on the left, you can see several tabs. There's
13:08 a tab for agents where we can build or find agents we already built or ones
13:13 that we've cloned or purchased. We can give our agents access to tools that
13:16 empower them to perform their responsibilities and give them access to
13:20 knowledge that provides the context for how to do their job well, such as
13:24 information about your company, your clients, your industry, and more.
13:29 Putting that all together, we can form our workforces. And these workforces and
13:33 the agents within them are empowered by integrations through different APIs.
13:37 That's just a fancy way of saying that we can sync up with other apps out on
13:41 the internet and use their functionality. This includes integrations with apps such as Gmail or
13:48 Google Drive, Google Meet, HubSpot, Slack, or Trello. When we're giving our
13:52 agents functionality, often we're giving them the ability to use external
13:56 integrations with other apps. Here you can see all of the agents that I've
14:01 either built, cloned, or purchased. We can store them in folders to categorize
14:05 them. And they are also grouped by the workforce they belong to. So, we're
14:09 about to create our first agent. But before we do that, I want to show you
14:12 this chat feature, which opens up this new window. What you're looking at here
14:17 is essentially like a regular LLM chat window, like chat GPT. But the cool
14:22 thing is you can prompt within here and add your agents or even entire
14:28 workforces and ask them to run tasks for you. So if I select this gamma
14:32 presentation designer agent, I can tell it to build me a presentation on selling
14:38 AI services to small to mediumsiz businesses. And it's able to do that
14:42 straight from this chat window. Then on the left, we have the chat history that
14:46 we can revisit. Later in the video, I'll show you exactly how to set up this
14:50 Gamma Graphic presentation designer, which is super powerful. Before we start
14:54 building our workforce, I want to orient So, now let's head into the workforce
15:05 tab and open up this meeting workforce cuz this is what we're going to be
15:09 building. At the top here, we have our two triggers. This is how we tell the
15:13 workforce to start. It'll start based on a message it receives. We can send our
15:17 workforce messages from within relevance, like through that chat window
15:20 I just showed you. But we can also trigger it from apps that we use every
15:25 day for work, like Slack. So, we'll set up a way to interact with this workforce
15:29 directly through Slack. And what are we telling it to do? Well, we're requesting
15:33 this workforce to book meetings for us. It all starts off with this orchestrator
15:38 agent that understands what we're wanting it to do, whom we're wanting it
15:42 to invite, and what's required for that meeting. This orchestrator agent has a
15:46 few agents that report to it called sub agents. There is one that finds internal
15:51 participants. These are members of your own team or company. We'll give the
15:55 agent a knowledge base to find the correct participant. We have another
15:59 agent that can find external participants. This one will be looking
16:03 within our integrated HubSpot to find leads uh potential clients of ours and
16:08 then perform research with LinkedIn and Google on that lead and the company they
16:12 work for. This ensures we have enough information about who we're meeting with
16:16 to feel prepared going in. And we also have this gamma graphic presentation
16:20 designer which will run anytime we request a presentation to be made. Maybe
16:24 we want to do a pitch to a lead or discuss internal metrics during a
16:29 company call. This agent can build those presentations for us automatically. If
16:33 you're not familiar with Gamma, it's an AI powered tool that turns your ideas or
16:37 documents into beautifully designed presentation ready slides and web pages
16:42 in seconds. And they now have an API which means we can ask Gamma to create
16:47 these kinds of assets for us from from agents within programs like relevance
16:51 and from other workflow platforms like make or N8N. Once the presentation is
16:55 ready, it will let the meeting orchestrator know. Ultimately, the
16:58 orchestrator's job is to book the meeting. So, it calls this meeting
17:02 booker agent, which uses Google Meet to schedule a meeting with all of the
17:06 required participants. And then it sends that meeting link to our notetaker agent
17:10 who is going to attend the meeting, take notes, transcribe the call, identify any
17:16 next steps or action items, and then create tasks within a to-do list app.
17:19 Finally, it will even send an email summary to all of the participants with
17:23 a link to those tasks. So, as you can see, this is a nice well-rounded
17:28 workforce that functions almost like a personal assistant, booking meetings
17:32 with internal or external participants, preparing presentations, and documenting
17:37 calls with action items for follow-up. So, if you're excited to start building
17:41 it, let's head over to the agents interface and build our first agent by
17:45 clicking on the new agent button. Like I mentioned before, we could invent this
17:50 and describe exactly what we want built. And relevance is going to do its best
17:54 job at building that for us. Or we could use a guided setup. We could even import
17:58 an existing agent from a file that someone shared with you. Or we can build
18:01 it from scratch. Since this is our first agent that we're building, I want to
18:05 make sure you understand exactly how it's built. So we're going to build this
18:16 Now we're inside the agent builder and on the left we can see the prompt. This
18:21 is where we create the guidelines for how the agent should function. We can
18:25 give it tools, knowledge, set up how it's triggered, build in some
18:30 escalations, memory, and variables. We'll touch on some of this in a moment.
18:33 And when we add any of this, it'll show up on the right hand panel over here. So
18:37 let's start and give our agent a name. We'll call it Borealis the meeting
18:41 booker. And below here, we're going to write some instructions. But first, I
18:45 want to bring your attention to the model section here. So, here is where we
18:50 select which AI LLM model we're using. As you can see here, we're using a
18:54 performance optimized model where it just picks the best one for us. We could
18:59 optimize by cost or select a specific one with chat, GPT, Claude, Gemini, or
19:04 whatever best suits our needs. Now that we're clear on which model we're using,
19:09 we can start to write our prompt. Now, there's not necessarily a standard way
19:13 of structuring these prompts, but the way that I like to do it is I'd like to
19:17 start out by defining the agents role. So, in this case, I'm telling it you are
19:22 Borealis and specifying that it is a meeting booker agent. Then, I'll just
19:26 bold its name so it's more easily scannable for the user. And I also want
19:30 to get clear on what inputs this agent should be receiving. So in this case,
19:34 it's going to be receiving information for the participant or participants to
19:39 invite to a meeting. And it's also going to be receiving the time zone of the
19:43 participant and I'll tell it that it may also receive an exact date and time to
19:48 book the meeting. So now we told our agent who it is, what information is
19:52 going to receive. Now we need to define its responsibility, instructing it on
19:56 what it's supposed to do and how to do that. So, we're going to tell it your
20:00 purpose is to schedule meetings and invite each participant. If a specific
20:04 meeting time is provided, book the meeting at that time. Otherwise,
20:08 identify a meeting time that logically is most convenient considering the time
20:12 zones of all participants. Now, I'm going to tell it that when it's
20:15 communicating with the user who is actually triggering it, refer to times
20:20 within their time zone. And this word here, time zone, we're actually going to
20:24 convert into something called a variable. You can think of a variable as
20:29 a placeholder that fills in with whatever value is present at that time.
20:33 So, we're going to go to the right hand panel and set up a new variable. It's
20:37 going to be a text variable. And we'll name the variable time zone. This is for
20:41 the agent to be able to reference what this is called. And we'll describe it as
20:45 the time zone of the user who is requesting the meeting. And below here
20:50 and the input is where we actually put the value that we want the placeholder
20:54 to be replaced by. So the actual time zone such as EDT or Pacific Standard
20:59 Time. And here in the green, we set the actual name that the agent is going to
21:03 be using inside of the prompt to refer to this variable. Now over in the
21:07 prompt, if we add these double curly braces around that variable name, then
21:11 that activates the variable. So it's going to be filled in with the value of
21:16 it, which in this case is EDT. Now, once you are clear on the meeting time,
21:20 schedule the meeting using a tool in order to add a tool into this prompt.
21:24 We're going to go into this right-hand panel and open up this tools modal and
21:28 we'll see we have a bunch of tools to choose from. They have them organized by
21:32 use case and you can also search for a specific tool. So in this case I'm
21:36 searching for a scheduling tool and I see that there's this Google Meet
21:39 scheduling tool that we can add to our prompt here is asking me to fill in the
21:44 missing tool inputs. In this case it needs me to select which Google Meet
21:48 account I want to connect to. Now, I've already connected my account, but if you
21:52 haven't yet, you can click add account and go through this process where
21:56 relevance AI uses pipedream to connect your Google account. I'll select the
22:00 account I already have connected and hit continue. And now here on the right tab
22:04 under tools, you can see I have that Google Meet tool. That means it's ready
22:08 to use within the prompt. So, I can access that by using forward slash. Then
22:13 I'll click tools, select Google Meet, and here is added right into the prompt
22:17 for the agent to use. Finally, I'm going to add some important considerations for
22:21 my agent. Going to tell it to only invite the participants who are
22:25 explicitly mentioned and tell it that we do not want to invite anyone else. And
22:30 just like our agent had an input, I'm also going to give it an output so we
22:33 can clarify what it should actually return when it's complete. So, we'll say
22:37 once you've booked the meeting, notify the user and provide the meeting link.
22:42 And if any errors or issues occur, notify the user. And with that, we
22:46 basically set up our agent. I'm just going to clean things up and add some
22:50 dividers here. If I hit hyphen three times, then it'll create these lines so
22:55 I can create some visual separation between the structure of my prompt. And
22:59 at the very top here, we can give the agent a description. So, of course,
23:04 Borealis schedules meetings and invites participants. We'll make sure to save
23:08 the agent. And this whole time, we've been on the build tab where we've been
23:11 building the agent. But now we want to go into the run tab where we can run the
23:16 agent to see how it's actually working. If we wanted to, we could add a little
23:20 guide to provide instructions for how to use or set up this agent, but this one
23:24 is super simple. So, we will keep that empty. And to run this, we're going to
23:28 tell Borealis to book a meeting with someone. And we'll give it an email for
23:32 whatever time. Let's say Friday at 100 p.m. Then we'll clarify the
23:38 participant's time zone is PDT. Now when we run this agent, we'll see the agent
23:42 working in real time. We can see it booking the meeting and it was a
23:45 success. It's provided us with the details and the link to the meeting
23:49 which we can click and join the meeting when we're ready. Within the timeline,
23:53 we can see the steps that were performed in the background such as using the
23:57 Google Meet tool. And just to prove this worked, I can go to the calendar that it
24:02 added to. And here is that event created for us with our new Borealis agent. So,
24:06 now that we know that it's working like it should, we could go ahead and share
24:10 this. We can make it publicly available, so anyone could run this agent. They
24:14 could embed it somewhere, or we could just share a link to it. We could also
24:18 turn this agent into a chat widget, which we could add to a website
24:21 somewhere. We could change its styling to fit the branding of the site, add a
24:26 starting prompt, such as, "Book me a meeting. Add a message placeholder so
24:30 people know what to actually type into here and how to use it. And set up other
24:35 configurations like allowing file uploads and toggling off the relevance
24:39 branding. Another option would be to make this agent into a clonable template
24:43 so we can share it with other users and they can clone it and adapt it to their
24:47 needs. But of course, we're building out an entire team of agents. So you'll see
24:51 later in the video how to publish an entire workforce out into the world. But
24:55 before we build out our next agent, let's get a better sense for how tools
25:00 work within relevance. Back around the build tab underneath the tool section,
25:04 we can see the tools that our agent has access to. In this case, our agent only
25:08 has access to the Google Meet tool. And what we're looking at here is all of the
25:12 tool inputs, such as the connected Google account and the calendar ID.
25:17 Here, we're letting the agent decide which ID to use, but we can set this
25:21 manually as well. This tool is currently set to run automatically. We could
25:25 require approval for it to run or have the agent decide. And conveniently, we
25:29 can edit the tool whether we created it ourself or we cloned it from the tool
25:34 marketplace. Inside here, we can set up all of the inputs for the tool. We can
25:39 set them as required or optional. Here, this looks pretty familiar. It has all
25:42 of the variables that this tool is making use of. And if we wanted to, we
25:47 could even add additional steps into this tool. Maybe we want to connect it
25:51 to a CRM like HubSpot. So once it schedules a meeting, it creates a note
25:55 within HubSpot relating to the person that it scheduled the meeting with. We
25:59 could go ahead and save the tool or save it as a draft, but because we don't want
26:03 to change it at all, I'm just going to close out of this. So that's the process
26:07 of looking under the hood of an existing tool. But of course, we can create our
26:11 own tools either by default, which means we're creating it from scratch, or we
26:15 could vibe create it, which is a way for us to invent our own tool with a prompt.
26:20 So, let's say we want to build a tool that takes in a transcript of a video or
26:24 a meeting and then produces an engaging post for LinkedIn. Now, once I hit go,
26:29 it's going to build this tool for me. I'm speeding this up for your
26:32 convenience, but as you can see, it walks through all of the steps to invent
26:36 this tool for me in the background. And I configure these inputs such as the
26:41 transcript, style, audience, and focus. So, if I hit run, it's going to generate
26:45 that LinkedIn post for me. And if I pop open the output, I see the actual post
26:50 and just scanning through it is looking pretty good. I could either use this
26:54 tool as is or continue to use the invent feature to refine this. So let's say I
26:58 don't need this post style input. I could go into this prompt and tell it I
27:02 don't need an input for this and just to remove it and it'll go through that
27:06 process and remove it for me. I could continue to make changes in this way, or
27:10 I could even switch to the default builder, which brings me to that view
27:13 that we were looking at before, which is the traditional tool editor, giving me
27:18 hands-on control over how this tool functions. And if I expand this Python
27:21 code, you can see that it did all of this for us in the background. We did
27:26 not have to write a line of Python code. It invented this all for us. Of course,
27:31 we could continue to add on to here. Maybe once the LinkedIn post is
27:35 generated, then we actually post it out onto LinkedIn via API. Now, just like an
27:39 agent has its run tab, the tool has a use tab. So, here we can actually use
27:43 the tool and we can see a log of all the times that we ran that tool to debug or
27:48 just to get a better sense for what With all that understanding in place,
27:57 now let's move on to creating our next agent. We're also going to build this
28:01 one from scratch and we're going to name it Polaris, the participant finder. And
28:05 it's going to find participants from within a knowledge base, which we will
28:09 set up in a second. First, we're going to define its role and tell it that you
28:14 are Polaris, the participant finder agent. Then we'll set up its input and
28:18 tell it that it's going to receive the name or title of an employee from a
28:22 company whose directory you have access to as a knowledge base. And we'll give
28:26 it a responsibility. For each participant, locate their info within
28:30 the company directory. And this is going to be that knowledge base. In order to
28:34 add a knowledge base, we'll go over to the right hand panel and click into
28:38 knowledge. And we can add a knowledge base in a number of ways. So we could
28:42 either sync up our agent to Google Drive or Notion, add an existing knowledge
28:46 base that was already added into our relevance account here, or we can simply
28:52 upload a file. So here I'm adding a CSV file that lives locally on my computer
28:55 and we can choose how to use that knowledge base. We can either add it all
28:59 to the prompt which is good for smaller data sets because when you do this it's
29:03 going to add the entire file and all of its contents directly into the prompt or
29:08 we can allow the agent to search and this is good for larger data sets. Since
29:13 the agent will be able to do a ragbased search on the entire knowledge base, but
29:17 since our company directory is quite small, we're going to add it all
29:21 directly into the prompt. Once it's uploaded, we'll see over in the prompt
29:24 that our knowledge base is added into the prompt here in the brackets. And
29:28 although it doesn't look complete, our agent actually sees it as if it's the
29:32 entire CSV file. Finally, for the output, we'll specify that it must
29:37 return the participants full info and to report any errors or issues that it
29:41 encounters while running. We'll go ahead and save it. Then head to the run tab
29:45 and test it out. So, we'll say find me my head of content. Again, it can take
29:49 in a title or a name and search the knowledge base for that participant. And
29:53 fortunately, we can see that it worked. It found the correct participant, Adam
29:58 Jar, my head of content. Clicking on this task icon over here, we can see all
30:04 of the tasks that our agent has run. We can see a detailed view. We can see if
30:08 there's anything to review, anything that's been escalated or any errors from
30:13 our tasks. We can see the queue if anything is in process, and then the
30:17 list of tasks that have run here. To run a new task, we could either schedule a
30:21 bunch of tasks in bulk, or we could just hit new task. And this allows us to run
30:27 new tasks in this agent. I could even tell it to find me all members of the
30:31 content department because again it has access to that company directory which
30:36 specifies the department that all of these employees work within. So it's
30:40 able to process really dynamic requests and find participants to match those
30:45 requests. Finally, if you wanted to search within the tasks that you ran,
30:49 you could even go and filter for tasks maybe where it's of a certain status. Or
30:53 you could even filter by a search term and pull up only the tasks that match
30:58 that term. Great. So now with all of that context in place and with two
31:02 agents that we've built, we can now head into the workforce tab and start to
31:06 build out our workforce with these new So, we'll go ahead and create a new
31:15 workforce. We'll give it a name. We'll call it meeting setter. And you can
31:18 think of this interface much like a canvas that you can drag different
31:22 elements onto in order to build out your workforce. If I drag on this agent card
31:27 on the right, I have all the agents that I can choose from. So, I'll select the
31:31 meeting booking agent. And as you can see, its prompt is available right here
31:35 in the panel for me to edit as needed. Next, I'll drag on another agent card.
31:39 And for this one, I'm going to be selecting the participant finder. And if
31:44 you remember from earlier, we need a way for these agents to collaborate together
31:49 through a manager or orchestrator agent who can receive meeting requests and
31:52 then delegate the tasks across this workforce. So, let's save our progress
31:57 so far and head out of the workforce so that we can go back to the agents tab
32:01 and create that orchestrator agent, which will build with the invent
32:05 feature. We could give it a very simple prompt here, but I'm actually going to
32:08 be pretty thorough because I want to get this right. So, I'm going to tell it to
32:12 invent an agent called Orion, the orchestrator. And this agent is going to
32:16 receive inbound messages requesting meetings and sometimes with
32:20 presentations. This orchestrator needs to interpret the intent of the message,
32:25 which means the meeting's purpose to require participants and whether a
32:29 presentation is needed. It then will delegate work to sub agents. Polaris to
32:35 find the internal team members, Lania to locate and research external leads.
32:39 We'll build that agent and a couple others soon like Gamma to create the
32:44 presentations. Borealis to schedule the meetings and Nate to attend, transcribe
32:49 and assign follow-ups after those meetings conclude. We need the
32:52 orchestrator to wait for all of the participant data from Polaris and Lania
32:57 before calling Borealis to book the meeting. And then once the meeting is
33:01 booked, we want the orchestrator to notify the user in Slack with a meeting
33:05 link to then continue checking on the gamma agents progress to make sure the
33:10 presentation is ready. And then when the presentation link is ready, send that to
33:14 the user and then of course report any errors or issues clearly back to the
33:18 user. So once we hit start, just like when we invented the tool, it's going to
33:22 go through this process and build out that entire agent for us. Of course,
33:26 I've sped this up for your convenience. We'll just hit accept on the suggestions
33:30 for its name and description. And it's produced this comprehensive prompt for
33:34 us. And just scanning through it, it's looking pretty good, but we're going to
33:38 walk through this step by step soon. So, let's just accept it for now. It's
33:42 asking for a Slack connection because that prompt mentioned it. We'll set that
33:46 up in a moment. So, now we just need to save the agent and we can use it over in
33:50 our workforce. So, let's head back to the workforce tab, open up the meeting
33:54 setter. We'll make some space and then drag an agent card onto the canvas and
33:59 select the orchestrator. Now, if we zoom into the prompt here, we can start
34:03 perfecting it for our needs. I see that its role is a meeting coordination AI.
34:08 I'm going to add the word agent just to be super clear. And it correctly says
34:12 that it receives inbound meeting requests and orchestrates a team of
34:17 specialized sub aents. And the instructions here are that it analyzes
34:21 those meeting requests, coordinates with the sub agents, waits for all the
34:25 participants, and keeps the user informed and handles any errors. So, it
34:29 looks like it got everything that we had requested. It also added in this
34:34 expected input format, but we don't need that. So, I'm going to actually delete
34:39 that out and exchange it for an example input message specifying that these are
34:44 messages from someone within a company wanting to meet with a teammate. the
34:48 full team or people from outside of their company. And then in here, I'll
34:52 just give a few examples of potential messages like scheduling meetings with a
34:57 person with an entire department or someone external from the company to
35:00 pitch them. And I'll add an important note here that for any mentions of
35:05 pitches, proposals, etc. Gamma should prepare a presentation. Now, as we
35:09 scroll down here, we see this sub aent coordination workflow. All these steps
35:13 are good, just like we wanted them to look. I'm just cleaning up the
35:16 formatting here. The only change I would make here is that we want to clarify
35:20 that it is Borealis, the meeting booker, that tells the note
35:26 taker to attend the meeting. It's not the orchestrator himself who does that.
35:31 And that's because Nate, the notetaker agent we will build later, is actually a
35:36 next step after Borealis, the meeting booker, meaning Nate is not a sub agent
35:41 of the orchestrator. We will also remove the Slack tool and just say to use Slack
35:45 because, as you'll see, since we'll be triggering this from Slack, Orion will
35:49 already have access to Slack through that trigger. Now, we'll just clean up
35:53 this formatting so it's easier to read. And because this orchestrator won't have
35:57 access to our notetaker agent, I'll just delete this step number eight, which
36:00 talks about the orchestrator being able to control it, which is not true. And
36:05 here's talking about how it has access to the knowledge bases, but actually its
36:11 sub agents do. So, we'll fix that. And with that, the prompt is ready. But
36:15 you'll notice that there's this warning on this tool. That's because we removed
36:20 the Slack tool from the prompt. So, it's essentially warning us that we have a
36:23 tool that we're not using. So, we'll just delete that out from the tools tab
36:27 and go ahead and save the changes to this agent. Now, we can wire this
36:30 orchestrator up to its sub agent. Starting with Polaris, we'll make sure
36:34 that the connection type is AI connection, which means it's a sub
36:38 agent, and give instructions for how it should be used to find internal meeting
36:43 participants. And then we can set a label here. We'll make sure that it auto runs. We could
36:49 have the approval required or let the agent decide, but we'll keep it on
36:53 automatic. Then we'll wire Orion to Borealis, giving instructions for how to
36:57 use it when the meeting is ready to be booked and set the edge label to book
37:02 meeting. In order to trigger Orion, we need to wire him up with this message
37:05 trigger. And we can test this out from the run tab. We'll make sure to save and
37:10 publish the workforce and then give it a message to book a meeting with my head
37:14 of content for Friday. And in the timeline here, Orion is working. It's using Polaris to find the
37:21 participant. is using Borealis to book the meeting with a Google Meet
37:25 scheduling tool. And voila, it's successfully worked and booked that
37:28 meeting with the right person. So, our workforce is in a really good spot so
37:31 far. If we head back to the build section, we can add another way to
37:35 trigger our workforce from outside of the relevance platform.
37:39 Specifically, we want to be triggering it from Slack. We already have our Slack
37:44 connected here. If you did not already, you could just connect it through here.
37:48 just put in your Slack organization here and go through the connection process.
37:52 And if you're not familiar with Slack, it's really a collaborative
37:55 communication environment for teams to use. So, I'll hit continue and then I'll
37:59 add a keyword here. I'll call it booker. The keyword is really just a word that
38:03 we use whenever we trigger this. So, that relevance knows which workforce we
38:07 want to run. Now, I'm just telling it where it can be triggered from from my
38:11 direct messages and all of these channels. and I'll confirm that I understand that
38:17 I always need to tag at relevance AI within Slack to use this trigger. If I
38:21 wanted to, I could enable specific working hours for this Slack trigger and
38:25 therefore for this workforce, but I don't want to limit it to specific
38:29 hours. Now, I'll just connect up this trigger to Orion, but now be able to
38:33 save the workforce, then head over into my Slack organization and make sure that
38:39 I have the relevance tool installed. So, I'll click on apps and make sure I
38:44 have it installed. If you don't yet have it installed, you'll want to open the
38:48 marketplace. Search for the relevance tool and then install it from here. But
38:52 since it's already available within my Slack, I should be able to go into my DM
38:58 and then say at relevance AI and then book it. That's that keyword that we set
39:02 up and say book me a meeting with my head of content for let's say Monday
39:07 morning and now over in relevance. That should be working. It's going to be
39:11 taking a while, so I'll speed this up for you, but eventually it's going to
39:14 give you a reply back. If you pop that open, it should say something like the
39:19 meeting was successfully scheduled and here is the meeting link.
39:24 Now, if we head back over into relevance and click on the run tab, we can see
39:29 here is that task which we can tell was triggered by myself from Slack. Now, as
39:33 I've been alluding to, sometimes our meetings are going to require us to
39:37 prepare presentations, pitches, proposals, and for that we're going to
39:47 So, let's head over to the relevance marketplace to get started with that. As
39:51 you'll see here, if we search for gamma, there are a couple agents that have
39:55 already been built that we can make use of. We're going to uh select the gamma
39:59 assistant here and go ahead and clone it. and we'll give it a more specific
40:03 name for our usage and call it gamma the presentation preparer. Now, when we save
40:07 it, we'll see that it's asking us to fill in this variable for the gamma API
40:12 key. I've already done that and I'll show you how to do it in a moment. So, I
40:16 can hit continue, then head back into the workforce and put it to work within
40:20 here. So I will drag on a new agent card and select the gamma agent and wire it up as
40:26 a sub agent to the orchestrator and give it instructions for how to use it where
40:31 we'll call gamma whenever a presentation is requested. Now we'll change this
40:34 label to something more informative such as prepare presentation. And now we're
40:39 set up to request presentations to be made directly from within this
40:42 workforce. But before we actually do that, I want to orient you to the Gamma
40:46 platform. We're happy to be a partner of theirs because they provide a powerful
40:51 way to bring your ideas to life where you can use prompts to create
40:55 presentations, branded documents, social media content, and even websites. And
41:01 now with their new API feature, we can automate the creation of this content
41:06 from places like relevance n make or even your own custom web apps. For
41:10 example, you could have a system where whenever a blog post is approved for
41:14 publication, Gamma runs automatically, creating a graphic that is perfectly
41:18 suited for that post. So, the possibilities are really endless here,
41:22 and it can save you and your team a bunch of time. Of course, in our
41:26 workforce, we're specifically interested in creating presentations, and Gamma is
41:30 capable of creating really any kind of presentation from a pitch deck to
41:34 something more personal to an internal team update to a keynote speech or
41:38 whatever kind of slide deck you need created. It's going to do it very fast
41:43 with little or no effort from you. And you can use AI to refine and polish the
41:47 presentation until it's ready to present. As you can see in their video
41:51 here, it all starts with a prompt which you can feed in from the actual Gamma
41:56 interface or programmatically from whatever workflow or app you're using.
42:00 Then Gamma is going to generate the entire presentation for you styled in a
42:05 theme of your choosing or you can create your own theme based on your branding.
42:09 Once it's ready, you can edit the presentation with AI, changing up the
42:15 verbiage or the layout itself. And you can drag and drop different elements
42:18 onto the slides until you're ready to share it publicly and present it out
42:23 into the world. If you don't yet have an account, you'll want to create one now.
42:27 But I'm going to log into my existing one cuz I want to show you some
42:29 presentations I've already generated from the API. Let's open one of these
42:33 up. And you'll see I have all of these slides. The text is based on internal
42:39 documents that I gave it access to. And it generated all of the images and
42:44 design itself based on the prompt. And if I wanted to change things like this
42:48 image here, I could regenerate some new AI images and change what I'm using
42:53 within the presentation. So, not only is it quick to generate, but it's quick to
42:57 iterate as well. And all of my slides here are using the theme Oasis. So, I'm
43:02 excited to show you how to create these from within the relevance platform. Now,
43:05 it's important to note that in order to use the API, you do have to have a pro
43:10 plan within Gamma. But when you consider all the time and energy you'll be saving
43:14 with a tool like this, it provides a lot of value. So, with that pro plan, you'll
43:18 be able to go into the settings here and generate an API key to use within the
43:22 relevance platform or wherever else you might want to make use of Gamma. You'll
43:27 just make sure that the gamma API tool in the agent has access to that API key
43:31 which you can set up within the integrations tab of relevance and add
43:36 that API key inside of here. So now that we're oriented to the power of gamma and
43:41 have it integrated into our agent, let's get clear on how this agent is
43:44 functioning. So under the core instructions, you can see this
43:47 documentation variable. So if I go to the variables tab, we'll see that we're
43:52 feeding our agent. uh all of this uh documentation that explains how to use
43:56 the gamma API and it even includes all of the code for an example request.
44:03 Note here that the theme name is Oasis like I showed you in the gamma platform
44:07 and the number of cards in the presentation is 10. These are all things
44:10 that we have control over from the prompt itself. The only thing I'm
44:14 changing here is I'm going to increase the duration of the delay and the amount
44:18 of times we're going to pull for the presentation to make sure that our
44:22 workforce can gain access to it and it doesn't time out too early before the
44:26 presentation is ready. And since we want to give Gamma access to some internal
44:31 documents to build presentations from, we're going to add a knowledge base and
44:34 sync up to Google Drive. We'll just select our connected Google account,
44:39 specify the drive itself, and whatever document we want to give access to. In
44:44 this case, it's information about how my agency provides AI transformation
44:49 partnership to clients. Since that's not a huge file, we will just add it all
44:53 into the prompt itself. In order for the agent to make proper use of this, we
44:57 just need to let it know that when it's researching and preparing for the
45:01 presentation, it should ask itself, does it need access to the AI transformation
45:06 partnership information? If so, it should reference that knowledge base
45:09 which is attached to the bottom of this prompt. So, if I scroll down here, there
45:14 it is. In order to empower even more, we're going to let the agent know that
45:17 as needed, it can perform Google research to search the meeting
45:21 attendees, the company topics, etc. and it can use a tool to scrape the content
45:26 from a website to discover helpful information about the website of the
45:30 person or company that is building a presentation for it. And with all of
45:35 that set up, our agent is empowered to prepare a presentation on a bunch of
45:39 different topics. If anything went wrong, let's say an AWS outage causes
45:44 gamma to not be functioning temporarily, we can set up what is called an
45:47 escalation, which means the agent will notify us via a method of our choosing
45:52 that there are issues that need addressing. So, we can set up our agent
45:57 to notify us via Slack whenever a task has timed out, for example, or there's
46:02 an unreoverable error or it's exhausted, it's retries. In any of these scenarios,
46:06 we can make sure to be notified within our connected Slack account at the
46:10 destination of our choosing. For example, through a direct message to
46:14 myself. We could also choose to be notified via email as well. So, we'll go
46:19 ahead and save this agent and make sure that it's working by heading over to the
46:23 run tab, saving our changes to the workforce, and asking our Gamma agent to
46:28 build a presentation on how Morningside AI can be an AI transformation partner
46:33 to this example company. I'll give it a URL. Now, as soon as I hit go, we can
46:38 see Orion has called the gamma agent. It's doing its research and planning,
46:42 which looks pretty thorough. It's thinking through how to structure
46:46 the presentation and planning how it's going to make use of the gamma API
46:51 itself. As it's calling that API, it looks like it ran into an error first,
46:55 but it keeps trying using delays during those attempts. And of course, I'm
46:59 speeding this up for you. and it eventually found success and it returned
47:03 this message that the presentation is ready including a summary of what it
47:08 covers and a link to the presentation itself. So if we click that open we can
47:13 view what it built for us which is already looking great. If we wanted to
47:19 we could tweak this and then on the day of open the presentation link and
47:22 present it live. So, this is a super quick and efficient way to get custom
47:27 pictures, presentations, and even proposals generated for anyone that
47:30 you're meeting with. Speaking of which, let's head back into relevance and
47:34 create our next agent, which is going to be responsible for locating our leads
47:38 within a connected customer relationship So, let's start building the next agent,
47:49 the lead locator. This agent is going to help us locate external participants so
47:54 that we can book meetings with people from outside of our company. These could
47:57 be people we want to pitch our business to and turn into clients. Our leads are
48:02 going to live in a tool called HubSpot. So, we'll switch over to that. Now, if
48:06 you haven't heard of it yet, HubSpot is a customer relationship management
48:11 platform or a CRM where businesses store all their customer lead information. You
48:15 can think of it as a central database for everyone that you're doing business
48:19 with or people that you want to do business with. You'll see leads
48:23 organized under contacts, companies, deals, tickets, and orders. But we're
48:27 mostly concerned about our contacts here. This is where the lead's
48:32 information will be stored, including their name, email, phone, and their
48:37 company. These fields are actually linked to their full company data, which
48:42 we can view as a list over in this company's tab. I'm walking you through
48:46 this because our lead locator agent will search through HubSpot and retrieve the
48:50 contact information for the leads we want to book meetings with. Now that you
48:53 understand what exactly we're asking for, let's go back to relevance and
48:57 start building out that agent who will be asking for these leads. We'll build
49:01 the agent from scratch. I'm choosing to name this one Lania the lead locator.
49:05 And we're going to give it a short description to locate external
49:10 participants from outside of our company. Okay. and we'll come down here
49:15 to the prompt and define its role, telling it who it is and specifying that
49:20 it's a sub agent of Orion, our orchestrator agent. Now, we'll tell it
49:24 what to expect as an input, which will be information about the lead, like
49:28 their email and the company they work for. And for added context, we'll tell
49:32 it that this is a participant of a future meeting. As for its
49:36 responsibility, we'll instruct it to locate the info with HubSpot. We can
49:41 access tools by typing a forward slash then the kind of tool we're looking for.
49:45 In this case, HubSpot. We have a few options to choose from. So, we'll select
49:49 the tool to retrieve contact details from HubSpot. Since this is an external
49:53 integration, we need to connect to our account so that we can have access to it
49:57 from within this agent. Now, let's move down a little bit here and continue its
50:00 instructions, letting our agent know that we want it to perform research on
50:04 this lead and their company. And then we're going to the next tool. We're
50:07 going to use LinkedIn. will use this LinkedIn tool to search for info from
50:12 their personal and company profiles. For more thorough research, we're also going
50:16 to use Google search. So, we'll add in a Google search tool here as well. This
50:20 research will be used for things like creating personalized pictures and
50:24 eventually proposals. Finally, we'll define the agents output and tell it
50:28 that it must return the lead's complete information and a research summary to
50:32 its boss, agent, Orion. And of course, it should let us know if there are any
50:36 errors or issues with locating the lead. Okay. And that's our prompt. By giving
50:41 it these tools, we've connected it to HubSpot and enabled it to search through
50:45 both LinkedIn and Google. So, it's pretty powerful with only a few lines of
50:49 prompt. I'll make sure to save this agent, then go into the run tab to test
50:53 it out to ensure everything is working. From this run tab, I'll pass in the
50:57 contact info for a potential lead. In this case, someone in my network that I
51:01 added into my HubSpot contacts. This way I can make sure this agent is not only
51:06 finding the correct contact but doing accurate research on them. Once I enter
51:09 the email you can see it started running. If you're ever wondering if the
51:14 agent is actually working on the right hand side here you can see the status
51:18 will update to running and it also shows the tools that it's going to use. So we
51:22 can see it is definitely working. It's looking into HubSpot. Then it's grabbing
51:27 info from LinkedIn profile. Now we can see it's searching Google. Looks like
51:30 the first one failed, but that's okay because it ran multiple searches, which
51:35 is great to see because that means this agent, like all agents you build in
51:40 relevance, is able to adapt on the fly. And once it completes, we can scroll up
51:44 here and see we have the correct contact details from HubSpot. It's grabbed a
51:48 bunch of details from LinkedIn about this uh lead and his company and
51:53 structured it out quite well, too. This is cool because our other agents like
51:56 Gamma will now have access to all of this context as it builds out super
52:01 custom presentations. For good measure, I'll mark this as complete. Now that
52:05 it's built and we've confirmed it works, we're ready to add this agent into our
52:09 workforce. So, we'll head out of the agent and head back to the workforces
52:13 tab and go into our meeting setter and drag another agent block onto the canvas
52:17 and choose Lania. We'll connect Orion down to Lania and make sure it's set as
52:22 an AI connection since this is another sub agent of Orion who will of course
52:26 call her to locate external participants. Now, I'll just spread them
52:29 out a little bit here to clean up the canvas. And as a final step, I'll update
52:33 the label here to find external participants. I'll make sure to save the
52:37 workspace because we're going to head out of here and go find our next and
52:42 final agent, the notetaker agent, who As we've seen, the marketplace
52:52 conveniently has a bunch of existing agents we can repurpose for our needs.
52:56 So, if we search for a notetaker, we'll see a few options. Nate fits our needs
53:01 quite well, so we'll select him and go ahead and clone it. Its current prompt
53:05 is pretty useful, so we don't need to start this one from scratch, and we can
53:09 reuse some of what is already here. But, we do need to change a few things. And
53:14 we'll start by clarifying its role is a notetaker agent who attends meetings, transcribes what
53:20 happens during that meeting and afterwards assigns tasks and sends a
53:24 meeting summary. As for its input, that will be information about the meeting
53:28 the agent must attend received from Borealis. The meeting booker agent will
53:32 tell it that it's responsible for generating summaries of all types of
53:36 meetings from internal team syncs to client calls and project updates or
53:41 strategy sessions. Inside of its instructions, the agent is being told to
53:45 use this send meeting bot tool to record and transcribe the call. And then when
53:49 the agent receives that transcript, it reads it carefully and writes a summary
53:52 of that meeting, including things like key discussion points, decisions made,
53:57 and action items. Then it's able to use this turn transcript into text tool to
54:02 generate a file of the transcript. And we're telling the agent to refer to the
54:06 transcript whenever a user asks questions about that meeting. I'll add a
54:09 bit more to the prompt here so the agent knows to determine the action items from
54:14 that meeting and then create tasks based on those actions items and create tasks
54:19 for them over in our to-do list in Trello. If you're not familiar with it,
54:24 Trello is a visual task management tool. Uh, and it's used to organize personal
54:28 and professional projects. It's kind of like a digital bulletin board with
54:32 sticky notes. We find those boards in this tab where we've got out to-do list
54:36 board here. Opening up that board, you can see that it has lists arranged in
54:41 columns, and these lists get tasks or cards added into them. You can name the
54:45 lists whatever you want or add new ones and then drag cards across them as your
54:50 tasks move through different statuses. Each card can be opened up and contains
54:54 a bunch of properties you can set up. Each one of our tasks that the notetaker
54:59 agent creates for us will be a new card, which is added to this to-do list board.
55:03 So, back in our agents prompt, we're telling it we want to use the Trello
55:07 tool to create a card on a Trello board. Here, we need to create a Trello
55:10 connection, and we're just going to link it through to our Trello account. We
55:15 need to authenticate with Trello. Just scroll down here, and we'll just allow
55:20 that connection, and then click continue and continue again. We scroll back down
55:25 and now we can see that's enabled. And now we can continue. When it creates the
55:30 card, it should fill in the following: the card name, its description, and its
55:35 due date. That's it. Finally, for the output, we're going to tell it to email
55:39 a summary to all participants using their emails they attended the meeting
55:43 with and include a link to the Trello board that we just added tasks to. And
55:48 it will do this using the send Gmail tool from our connected account so it
55:52 knows how to structure the email. We'll give it a format to follow, which I'll
55:57 just paste in here. and we'll include a sample email as well.
56:04 Great. So, we'll save that agent. Lastly, we need to add a trigger. So,
56:07 we're going to add relevance meeting bot and click setup trigger. And that's been
56:11 added in. Noteaker agents are a little different and need to be triggered
56:15 twice. Once to join the meeting, which in this case, the Borealis agent will
56:19 do, and once when the meeting is finished. In order to trigger it for the
56:22 second time, you need to add this meeting bot trigger to the agent. Now we
56:27 can add it to our workforce. So let's head back over there. Inside our
56:32 workforce, I'll drag on a new agent. Tell it to use Nate the note taker. And
56:36 for the handoff type, this one should be next step. This means that instead of
56:41 the orchestrator agent managing this agent, Nate runs automatically once
56:46 Borealis books the meeting and hands him the meeting link. For good measure,
56:49 we'll edit the label here and specify that we're sending a meeting link on
56:53 this handoff. Now, we're good to go. So, Okay, we've built all of the agents and
57:04 now we can see them working together as a workforce. So, let's jump over to
57:08 Slack. This is where we're going to trigger the workforce. Call the relevant
57:12 AI bot. And then our particular trigger also needs this trigger here. So, we're
57:17 going to say booker. I'm going to say book me a meeting with Adam. And we'll put
57:24 Adam's email in here. next Tuesday at 10:00 a.m. Eastern Standard Time and create a presentation
57:30 about how we can be their AI transformation partner and share it as a
57:34 link in the meeting. Okay. So, we're going to click to send that and straight
57:38 away it responds. So, we can look at that and view thread. So, we've got that
57:41 confirmation here that the agent is cooking up a reply. So, let's jump over
57:46 to relevance AI. We will have a look at the run tab and we can see that has
57:50 already started here and it's already running. Lania the lead locator is
57:54 working and then we are getting the LinkedIn details and we're doing the
57:59 Google search and then book the meeting. Okay, now moving on to creating the
58:09 Now that's completed. So let's jump over to Slack again and we'll just look at
58:13 the reply. So it's created how Morningside AI can be this company's AI
58:18 transformation partner exactly as we asked it to do. So, it's gone through
58:23 and created this detailed presentation. Might need a little bit of a review and
58:26 then edit, but that's certainly a good starting point. Okay. So, let's close
58:31 that. Now, let's jump over to Google Calendar. And we can see that the
58:36 appointment here we need to say yes, we are going to attend. So, let's join with
58:40 Google Meet. Inside Google Meet here, we can see Nate wanting to join. So, I'll
58:44 admit him. And there he is taking all of the notes that I need. He will write it
58:49 all down, email it to us, and assign tasks as we will see later. After the
58:53 meeting, let's head back to relevance. We've seen that complete. We can jump
58:58 over to Trello, and we can see we've got a couple of tasks here that have been
59:01 created, and we can jump up to our email. There's a summary with the key
59:05 points, the decisions made, and then action items and next steps. This was
59:09 sent to all participants automatically after the meeting ended. So, there you
5:34 So if a workforce is composed of AI agents, what exactly is an AI agent? So
5:38 you can think of them as like a digital employee. You can give them a task like
5:42 look this up or write this message or organize this list and then they take
5:45 action for you and they do it quickly. They don't get tired and they're always
5:49 ready. It's one worker with one job and a few tools that allow it to do that job
5:52 properly. So it's kind of like a specialized digital employee. But no
5:56 business runs on just one employee. A shop doesn't just survive with one
5:59 cashier. It needs someone to do the inventory, someone to restock the
6:03 shelves, someone to handle the finances. Each person in the company has their
6:06 role and together they keep the place running. And that's exactly what an AI
6:09 workforce is. It's a team of AI agents, each with its own specialty working
6:13 together like a department inside a company. Workforce that we're going to
6:15 be building in this video looks a bit like this. It all kicks off over Slack,
6:19 which is a business messaging platform where you can send a Slack message,
6:22 schedule a meeting with anyone in your company, or even external contacts. The
6:25 workforce then books the event and it can even prepare a presentation for the
6:29 meeting based on company docs and external research. And then finally, it
6:32 can take notes during the call and assign tasks when the meeting ends. So
6:36 together, this workforce acts like a personal assistant ready to help out
6:39 with the literal click of a button. So this is the shift that's happening right
6:42 now. We're moving from these single AI assistants or agents that only handle
6:46 one-off tasks to these collaborative AI workforces that can take care of entire
6:49 business processes. That's not just helpful, that's ultimately going to be
6:52 world changing. And if you can learn how to build and sell them, then that's
6:55 going to be life-changing for you over the next few years. So, now that you
6:58 know what a workforce is, how are they So, you can visualize an AI workforce
7:07 like an org chart from a typical company with human employees where there are
7:11 specific roles, clear reporting structures, and defined workflows
7:14 between the different positions. And the core principles of human and AI
7:17 workforces are the same. A workforce runs on three pillars. You have
7:21 specialization, collaboration, and coordination. So specialization means
7:24 that each agent has one clear job and gets great at it. For example, you have
7:28 a researcher that pulls information, a writer that turns that into a useful
7:32 plan, a designer then prepares a presentation for it, and so on. But we
7:35 don't want our agents to be working in silos. So you need collaboration where
7:38 agents can work together by passing tasks and information between each
7:41 other. Just like a designer might create a logo and then pass it to an animator
7:45 to animate it. And of course, these work forces need coordination. So, it's our
7:48 job as the workforce builders to create a clear and predictable system
7:52 structured to keep everything organized without error. We can even have agents
7:55 in the workforce that function like a project manager who orchestrates
7:58 everything, deciding what runs and when, making sure everyone is communicating
8:01 well and everything is operating smoothly and of course addressing issues
8:05 when they do happen. Just like a real org chart, AI workforces usually take a
8:09 few different shapes with agents who are working sequentially and completing
8:12 tasks in a straightforward line where each agent depends on the last, like an
8:15 assembly line. You can also have agents that run in parallel, completing the
8:19 task at the same time, which can cut down on the overall time it takes for
8:21 work to complete. Like in human organizations, AI workforces can have a
8:25 hierarchy as well, where assignments flow from the top to the bottom. And
8:28 there can be an orchestrator agent who functions like the team leader who
8:31 understands the goals and then breaks a big job into smaller parts, assigning it
8:34 to these different specialist agents and then pulling the results back into a
8:37 final result like a project manager would. But what enables these agents to
8:41 actually work? What gives them the ability to understand, remember, and
8:44 take action? Well, each agent is powered by a few core components that mirror how
8:47 real employees work. First off, they have a clear understanding of how to do
8:51 their job. For a human employee, that would be your job description, but for
8:54 an agent, it's their prompt. These are the instructions that you give to an
8:57 agent to define its role and its responsibility. It's the blueprint for
9:00 how the work needs to be done. But if you just handed a sculptor a sketch of
9:03 what you needed it to sculpt, but you didn't actually give the actual stone,
9:06 they wouldn't be able to do anything. So, we often need to give our agents
9:10 access to resources or materials like a transcript of a call or a customer's
9:14 order or campaign performance data in order for it to do its job. But of
9:17 course, if we only gave the sculptor a block of marble but nothing to carve it
9:20 with, then it would be powerless. Similarly, we can empower agents by
9:23 giving them access to tools that they can use to execute their duties like a
9:27 web scraper to perform research on a company or a document converter that
9:31 turns text into PDFs. These tools can include integrations with external apps
9:35 like CRM like HubSpot or project management platforms like Notion. This
9:38 way, your agents can interface with the external world and sync up with other
9:41 systems in the company. You can also give agents access to a knowledge base,
9:45 which is essentially a database full of useful knowledge that agents can
9:47 reference as needed. This could be anything from internal company reports
9:51 to a list of frequently asked questions that a customer support agent could
9:54 reference in order to come up with standardized responses to customer
9:57 questions. In addition to your agent being able to know things, they can also
10:00 have the ability to remember things, too. So, by giving an agent the power of
10:03 memory, they can recall what's happened in previous steps or even across time.
10:06 And this not only helps an agent to do its job better now, but allows them to
10:10 improve over time if you set it up correctly. So, circling back to
10:13 collaboration and coordination. When it's time to pass work along from one
10:16 agent to the other, there are different patterns that keep things smooth. Often,
10:19 the handoff is clean, like a baton pass, where one agent finishes the task and
10:22 hands the next agent what it needs, and that next agent takes off running. Other
10:26 times, an agent might have to clear a couple hurdles first, like making sure
10:29 the right file is ready before it can hand things over to the next agent who
10:32 would not be able to do the task if they didn't have that file. And other times,
10:35 this coordination can go both ways where an orchestrator agent might request work
10:38 from an agent it oversees. And then that subordinate agent performs a task and
10:41 hands the results back up to its manager. Then that manager might process
10:44 things and hand it off to another agent who finishes things off. So with all
10:46 that out of the way, we are about to build. So let's quickly recap to make
10:56 So, while these work forces can get complex, ultimately building them is
10:59 really about starting with the simplest piece that works. You break down work by
11:03 specialization, creating unique agents that have a prompt that specifies its
11:06 role and responsibilities. Then, by providing it with all of the necessary
11:10 resources, knowledge, and tools, it has what it needs to perform the tasks
11:13 you've told it to do. Then by connecting these specialized agents into a virtual
11:16 org chart and making handoffs between teammates clear and predictable, you've
11:20 built the foundation for collaboration and coordination that can start small
11:23 and continue to scale. With this essential understanding in place, let's
11:26 see this in action and build our first AI workforce, starting with our first
11:29 agent. So to make building your first AI workforce as smooth as possible for you,
11:32 I have organized all of the prompts and instructions for this workforce over in
11:35 my free school community. So if you haven't yet, feel free to pause this
11:38 video now. You can head down to the first link in the description, join my
11:40 school. It'll take a minute or two for you to get accepted. Then you can go to
11:43 the classroom section where you'll be able to access everything. All the
11:46 prompts, tools, links, everything you To build our workforce, we'll be using
11:59 Relevance, who we're happy to be partnering with on this video. Like it
12:03 says, we can build teams of agents that deliver human quality work. We can even
12:08 invent our agent with a simple prompt. For example, we could tell relevance
12:12 that we want an agent to research a person on LinkedIn and Google, then
12:16 click invent, and it will spin one up for us. We'll take a look at that
12:19 feature later when we're inside the platform and use it along with other
12:24 tools to build out our full AI workforce. If you don't already have a
12:28 relevance account, you can sign up to create one, but I'm going to log into my
12:32 existing one through Google. Here we're looking at the relevance marketplace.
12:37 This is an ecosystem of agents, tools, and entire workforces that the builder
12:41 community can use. Like this image generator agent, for example, which
12:46 generates images using GPT models. We could clone this into our project and
12:50 use it for our needs. There are also agents that can be purchased like this
12:55 Gmail task creation agent. As you can see, we could buy this for 99 and use it
13:00 straight away inside of our projects. So, how do we actually start to create
13:04 our own agents? Well, over here on the left, you can see several tabs. There's
13:08 a tab for agents where we can build or find agents we already built or ones
13:13 that we've cloned or purchased. We can give our agents access to tools that
13:16 empower them to perform their responsibilities and give them access to
13:20 knowledge that provides the context for how to do their job well, such as
13:24 information about your company, your clients, your industry, and more.
13:29 Putting that all together, we can form our workforces. And these workforces and
13:33 the agents within them are empowered by integrations through different APIs.
13:37 That's just a fancy way of saying that we can sync up with other apps out on
13:41 the internet and use their functionality. This includes integrations with apps such as Gmail or
13:48 Google Drive, Google Meet, HubSpot, Slack, or Trello. When we're giving our
13:52 agents functionality, often we're giving them the ability to use external
13:56 integrations with other apps. Here you can see all of the agents that I've
14:01 either built, cloned, or purchased. We can store them in folders to categorize
14:05 them. And they are also grouped by the workforce they belong to. So, we're
14:09 about to create our first agent. But before we do that, I want to show you
14:12 this chat feature, which opens up this new window. What you're looking at here
14:17 is essentially like a regular LLM chat window, like chat GPT. But the cool
14:22 thing is you can prompt within here and add your agents or even entire
14:28 workforces and ask them to run tasks for you. So if I select this gamma
14:32 presentation designer agent, I can tell it to build me a presentation on selling
14:38 AI services to small to mediumsiz businesses. And it's able to do that
14:42 straight from this chat window. Then on the left, we have the chat history that
14:46 we can revisit. Later in the video, I'll show you exactly how to set up this
14:50 Gamma Graphic presentation designer, which is super powerful. Before we start
14:54 building our workforce, I want to orient So, now let's head into the workforce
15:05 tab and open up this meeting workforce cuz this is what we're going to be
15:09 building. At the top here, we have our two triggers. This is how we tell the
15:13 workforce to start. It'll start based on a message it receives. We can send our
15:17 workforce messages from within relevance, like through that chat window
15:20 I just showed you. But we can also trigger it from apps that we use every
15:25 day for work, like Slack. So, we'll set up a way to interact with this workforce
15:29 directly through Slack. And what are we telling it to do? Well, we're requesting
15:33 this workforce to book meetings for us. It all starts off with this orchestrator
15:38 agent that understands what we're wanting it to do, whom we're wanting it
15:42 to invite, and what's required for that meeting. This orchestrator agent has a
15:46 few agents that report to it called sub agents. There is one that finds internal
15:51 participants. These are members of your own team or company. We'll give the
15:55 agent a knowledge base to find the correct participant. We have another
15:59 agent that can find external participants. This one will be looking
16:03 within our integrated HubSpot to find leads uh potential clients of ours and
16:08 then perform research with LinkedIn and Google on that lead and the company they
16:12 work for. This ensures we have enough information about who we're meeting with
16:16 to feel prepared going in. And we also have this gamma graphic presentation
16:20 designer which will run anytime we request a presentation to be made. Maybe
16:24 we want to do a pitch to a lead or discuss internal metrics during a
16:29 company call. This agent can build those presentations for us automatically. If
16:33 you're not familiar with Gamma, it's an AI powered tool that turns your ideas or
16:37 documents into beautifully designed presentation ready slides and web pages
16:42 in seconds. And they now have an API which means we can ask Gamma to create
16:47 these kinds of assets for us from from agents within programs like relevance
16:51 and from other workflow platforms like make or N8N. Once the presentation is
16:55 ready, it will let the meeting orchestrator know. Ultimately, the
16:58 orchestrator's job is to book the meeting. So, it calls this meeting
17:02 booker agent, which uses Google Meet to schedule a meeting with all of the
17:06 required participants. And then it sends that meeting link to our notetaker agent
17:10 who is going to attend the meeting, take notes, transcribe the call, identify any
17:16 next steps or action items, and then create tasks within a to-do list app.
17:19 Finally, it will even send an email summary to all of the participants with
17:23 a link to those tasks. So, as you can see, this is a nice well-rounded
17:28 workforce that functions almost like a personal assistant, booking meetings
17:32 with internal or external participants, preparing presentations, and documenting
17:37 calls with action items for follow-up. So, if you're excited to start building
17:41 it, let's head over to the agents interface and build our first agent by
17:45 clicking on the new agent button. Like I mentioned before, we could invent this
17:50 and describe exactly what we want built. And relevance is going to do its best
17:54 job at building that for us. Or we could use a guided setup. We could even import
17:58 an existing agent from a file that someone shared with you. Or we can build
18:01 it from scratch. Since this is our first agent that we're building, I want to
18:05 make sure you understand exactly how it's built. So we're going to build this
18:16 Now we're inside the agent builder and on the left we can see the prompt. This
18:21 is where we create the guidelines for how the agent should function. We can
18:25 give it tools, knowledge, set up how it's triggered, build in some
18:30 escalations, memory, and variables. We'll touch on some of this in a moment.
18:33 And when we add any of this, it'll show up on the right hand panel over here. So
18:37 let's start and give our agent a name. We'll call it Borealis the meeting
18:41 booker. And below here, we're going to write some instructions. But first, I
18:45 want to bring your attention to the model section here. So, here is where we
18:50 select which AI LLM model we're using. As you can see here, we're using a
18:54 performance optimized model where it just picks the best one for us. We could
18:59 optimize by cost or select a specific one with chat, GPT, Claude, Gemini, or
19:04 whatever best suits our needs. Now that we're clear on which model we're using,
19:09 we can start to write our prompt. Now, there's not necessarily a standard way
19:13 of structuring these prompts, but the way that I like to do it is I'd like to
19:17 start out by defining the agents role. So, in this case, I'm telling it you are
19:22 Borealis and specifying that it is a meeting booker agent. Then, I'll just
19:26 bold its name so it's more easily scannable for the user. And I also want
19:30 to get clear on what inputs this agent should be receiving. So in this case,
19:34 it's going to be receiving information for the participant or participants to
19:39 invite to a meeting. And it's also going to be receiving the time zone of the
19:43 participant and I'll tell it that it may also receive an exact date and time to
19:48 book the meeting. So now we told our agent who it is, what information is
19:52 going to receive. Now we need to define its responsibility, instructing it on
19:56 what it's supposed to do and how to do that. So, we're going to tell it your
20:00 purpose is to schedule meetings and invite each participant. If a specific
20:04 meeting time is provided, book the meeting at that time. Otherwise,
20:08 identify a meeting time that logically is most convenient considering the time
20:12 zones of all participants. Now, I'm going to tell it that when it's
20:15 communicating with the user who is actually triggering it, refer to times
20:20 within their time zone. And this word here, time zone, we're actually going to
20:24 convert into something called a variable. You can think of a variable as
20:29 a placeholder that fills in with whatever value is present at that time.
20:33 So, we're going to go to the right hand panel and set up a new variable. It's
20:37 going to be a text variable. And we'll name the variable time zone. This is for
20:41 the agent to be able to reference what this is called. And we'll describe it as
20:45 the time zone of the user who is requesting the meeting. And below here
20:50 and the input is where we actually put the value that we want the placeholder
20:54 to be replaced by. So the actual time zone such as EDT or Pacific Standard
20:59 Time. And here in the green, we set the actual name that the agent is going to
21:03 be using inside of the prompt to refer to this variable. Now over in the
21:07 prompt, if we add these double curly braces around that variable name, then
21:11 that activates the variable. So it's going to be filled in with the value of
21:16 it, which in this case is EDT. Now, once you are clear on the meeting time,
21:20 schedule the meeting using a tool in order to add a tool into this prompt.
21:24 We're going to go into this right-hand panel and open up this tools modal and
21:28 we'll see we have a bunch of tools to choose from. They have them organized by
21:32 use case and you can also search for a specific tool. So in this case I'm
21:36 searching for a scheduling tool and I see that there's this Google Meet
21:39 scheduling tool that we can add to our prompt here is asking me to fill in the
21:44 missing tool inputs. In this case it needs me to select which Google Meet
21:48 account I want to connect to. Now, I've already connected my account, but if you
21:52 haven't yet, you can click add account and go through this process where
21:56 relevance AI uses pipedream to connect your Google account. I'll select the
22:00 account I already have connected and hit continue. And now here on the right tab
22:04 under tools, you can see I have that Google Meet tool. That means it's ready
22:08 to use within the prompt. So, I can access that by using forward slash. Then
22:13 I'll click tools, select Google Meet, and here is added right into the prompt
22:17 for the agent to use. Finally, I'm going to add some important considerations for
22:21 my agent. Going to tell it to only invite the participants who are
22:25 explicitly mentioned and tell it that we do not want to invite anyone else. And
22:30 just like our agent had an input, I'm also going to give it an output so we
22:33 can clarify what it should actually return when it's complete. So, we'll say
22:37 once you've booked the meeting, notify the user and provide the meeting link.
22:42 And if any errors or issues occur, notify the user. And with that, we
22:46 basically set up our agent. I'm just going to clean things up and add some
22:50 dividers here. If I hit hyphen three times, then it'll create these lines so
22:55 I can create some visual separation between the structure of my prompt. And
22:59 at the very top here, we can give the agent a description. So, of course,
23:04 Borealis schedules meetings and invites participants. We'll make sure to save
23:08 the agent. And this whole time, we've been on the build tab where we've been
23:11 building the agent. But now we want to go into the run tab where we can run the
23:16 agent to see how it's actually working. If we wanted to, we could add a little
23:20 guide to provide instructions for how to use or set up this agent, but this one
23:24 is super simple. So, we will keep that empty. And to run this, we're going to
23:28 tell Borealis to book a meeting with someone. And we'll give it an email for
23:32 whatever time. Let's say Friday at 100 p.m. Then we'll clarify the
23:38 participant's time zone is PDT. Now when we run this agent, we'll see the agent
23:42 working in real time. We can see it booking the meeting and it was a
23:45 success. It's provided us with the details and the link to the meeting
23:49 which we can click and join the meeting when we're ready. Within the timeline,
23:53 we can see the steps that were performed in the background such as using the
23:57 Google Meet tool. And just to prove this worked, I can go to the calendar that it
24:02 added to. And here is that event created for us with our new Borealis agent. So,
24:06 now that we know that it's working like it should, we could go ahead and share
24:10 this. We can make it publicly available, so anyone could run this agent. They
24:14 could embed it somewhere, or we could just share a link to it. We could also
24:18 turn this agent into a chat widget, which we could add to a website
24:21 somewhere. We could change its styling to fit the branding of the site, add a
24:26 starting prompt, such as, "Book me a meeting. Add a message placeholder so
24:30 people know what to actually type into here and how to use it. And set up other
24:35 configurations like allowing file uploads and toggling off the relevance
24:39 branding. Another option would be to make this agent into a clonable template
24:43 so we can share it with other users and they can clone it and adapt it to their
24:47 needs. But of course, we're building out an entire team of agents. So you'll see
24:51 later in the video how to publish an entire workforce out into the world. But
24:55 before we build out our next agent, let's get a better sense for how tools
25:00 work within relevance. Back around the build tab underneath the tool section,
25:04 we can see the tools that our agent has access to. In this case, our agent only
25:08 has access to the Google Meet tool. And what we're looking at here is all of the
25:12 tool inputs, such as the connected Google account and the calendar ID.
25:17 Here, we're letting the agent decide which ID to use, but we can set this
25:21 manually as well. This tool is currently set to run automatically. We could
25:25 require approval for it to run or have the agent decide. And conveniently, we
25:29 can edit the tool whether we created it ourself or we cloned it from the tool
25:34 marketplace. Inside here, we can set up all of the inputs for the tool. We can
25:39 set them as required or optional. Here, this looks pretty familiar. It has all
25:42 of the variables that this tool is making use of. And if we wanted to, we
25:47 could even add additional steps into this tool. Maybe we want to connect it
25:51 to a CRM like HubSpot. So once it schedules a meeting, it creates a note
25:55 within HubSpot relating to the person that it scheduled the meeting with. We
25:59 could go ahead and save the tool or save it as a draft, but because we don't want
26:03 to change it at all, I'm just going to close out of this. So that's the process
26:07 of looking under the hood of an existing tool. But of course, we can create our
26:11 own tools either by default, which means we're creating it from scratch, or we
26:15 could vibe create it, which is a way for us to invent our own tool with a prompt.
26:20 So, let's say we want to build a tool that takes in a transcript of a video or
26:24 a meeting and then produces an engaging post for LinkedIn. Now, once I hit go,
26:29 it's going to build this tool for me. I'm speeding this up for your
26:32 convenience, but as you can see, it walks through all of the steps to invent
26:36 this tool for me in the background. And I configure these inputs such as the
26:41 transcript, style, audience, and focus. So, if I hit run, it's going to generate
26:45 that LinkedIn post for me. And if I pop open the output, I see the actual post
26:50 and just scanning through it is looking pretty good. I could either use this
26:54 tool as is or continue to use the invent feature to refine this. So let's say I
26:58 don't need this post style input. I could go into this prompt and tell it I
27:02 don't need an input for this and just to remove it and it'll go through that
27:06 process and remove it for me. I could continue to make changes in this way, or
27:10 I could even switch to the default builder, which brings me to that view
27:13 that we were looking at before, which is the traditional tool editor, giving me
27:18 hands-on control over how this tool functions. And if I expand this Python
27:21 code, you can see that it did all of this for us in the background. We did
27:26 not have to write a line of Python code. It invented this all for us. Of course,
27:31 we could continue to add on to here. Maybe once the LinkedIn post is
27:35 generated, then we actually post it out onto LinkedIn via API. Now, just like an
27:39 agent has its run tab, the tool has a use tab. So, here we can actually use
27:43 the tool and we can see a log of all the times that we ran that tool to debug or
27:48 just to get a better sense for what With all that understanding in place,
27:57 now let's move on to creating our next agent. We're also going to build this
28:01 one from scratch and we're going to name it Polaris, the participant finder. And
28:05 it's going to find participants from within a knowledge base, which we will
28:09 set up in a second. First, we're going to define its role and tell it that you
28:14 are Polaris, the participant finder agent. Then we'll set up its input and
28:18 tell it that it's going to receive the name or title of an employee from a
28:22 company whose directory you have access to as a knowledge base. And we'll give
28:26 it a responsibility. For each participant, locate their info within
28:30 the company directory. And this is going to be that knowledge base. In order to
28:34 add a knowledge base, we'll go over to the right hand panel and click into
28:38 knowledge. And we can add a knowledge base in a number of ways. So we could
28:42 either sync up our agent to Google Drive or Notion, add an existing knowledge
28:46 base that was already added into our relevance account here, or we can simply
28:52 upload a file. So here I'm adding a CSV file that lives locally on my computer
28:55 and we can choose how to use that knowledge base. We can either add it all
28:59 to the prompt which is good for smaller data sets because when you do this it's
29:03 going to add the entire file and all of its contents directly into the prompt or
29:08 we can allow the agent to search and this is good for larger data sets. Since
29:13 the agent will be able to do a ragbased search on the entire knowledge base, but
29:17 since our company directory is quite small, we're going to add it all
29:21 directly into the prompt. Once it's uploaded, we'll see over in the prompt
29:24 that our knowledge base is added into the prompt here in the brackets. And
29:28 although it doesn't look complete, our agent actually sees it as if it's the
29:32 entire CSV file. Finally, for the output, we'll specify that it must
29:37 return the participants full info and to report any errors or issues that it
29:41 encounters while running. We'll go ahead and save it. Then head to the run tab
29:45 and test it out. So, we'll say find me my head of content. Again, it can take
29:49 in a title or a name and search the knowledge base for that participant. And
29:53 fortunately, we can see that it worked. It found the correct participant, Adam
29:58 Jar, my head of content. Clicking on this task icon over here, we can see all
30:04 of the tasks that our agent has run. We can see a detailed view. We can see if
30:08 there's anything to review, anything that's been escalated or any errors from
30:13 our tasks. We can see the queue if anything is in process, and then the
30:17 list of tasks that have run here. To run a new task, we could either schedule a
30:21 bunch of tasks in bulk, or we could just hit new task. And this allows us to run
30:27 new tasks in this agent. I could even tell it to find me all members of the
30:31 content department because again it has access to that company directory which
30:36 specifies the department that all of these employees work within. So it's
30:40 able to process really dynamic requests and find participants to match those
30:45 requests. Finally, if you wanted to search within the tasks that you ran,
30:49 you could even go and filter for tasks maybe where it's of a certain status. Or
30:53 you could even filter by a search term and pull up only the tasks that match
30:58 that term. Great. So now with all of that context in place and with two
31:02 agents that we've built, we can now head into the workforce tab and start to
31:06 build out our workforce with these new So, we'll go ahead and create a new
31:15 workforce. We'll give it a name. We'll call it meeting setter. And you can
31:18 think of this interface much like a canvas that you can drag different
31:22 elements onto in order to build out your workforce. If I drag on this agent card
31:27 on the right, I have all the agents that I can choose from. So, I'll select the
31:31 meeting booking agent. And as you can see, its prompt is available right here
31:35 in the panel for me to edit as needed. Next, I'll drag on another agent card.
31:39 And for this one, I'm going to be selecting the participant finder. And if
31:44 you remember from earlier, we need a way for these agents to collaborate together
31:49 through a manager or orchestrator agent who can receive meeting requests and
31:52 then delegate the tasks across this workforce. So, let's save our progress
31:57 so far and head out of the workforce so that we can go back to the agents tab
32:01 and create that orchestrator agent, which will build with the invent
32:05 feature. We could give it a very simple prompt here, but I'm actually going to
32:08 be pretty thorough because I want to get this right. So, I'm going to tell it to
32:12 invent an agent called Orion, the orchestrator. And this agent is going to
32:16 receive inbound messages requesting meetings and sometimes with
32:20 presentations. This orchestrator needs to interpret the intent of the message,
32:25 which means the meeting's purpose to require participants and whether a
32:29 presentation is needed. It then will delegate work to sub agents. Polaris to
32:35 find the internal team members, Lania to locate and research external leads.
32:39 We'll build that agent and a couple others soon like Gamma to create the
32:44 presentations. Borealis to schedule the meetings and Nate to attend, transcribe
32:49 and assign follow-ups after those meetings conclude. We need the
32:52 orchestrator to wait for all of the participant data from Polaris and Lania
32:57 before calling Borealis to book the meeting. And then once the meeting is
33:01 booked, we want the orchestrator to notify the user in Slack with a meeting
33:05 link to then continue checking on the gamma agents progress to make sure the
33:10 presentation is ready. And then when the presentation link is ready, send that to
33:14 the user and then of course report any errors or issues clearly back to the
33:18 user. So once we hit start, just like when we invented the tool, it's going to
33:22 go through this process and build out that entire agent for us. Of course,
33:26 I've sped this up for your convenience. We'll just hit accept on the suggestions
33:30 for its name and description. And it's produced this comprehensive prompt for
33:34 us. And just scanning through it, it's looking pretty good, but we're going to
33:38 walk through this step by step soon. So, let's just accept it for now. It's
33:42 asking for a Slack connection because that prompt mentioned it. We'll set that
33:46 up in a moment. So, now we just need to save the agent and we can use it over in
33:50 our workforce. So, let's head back to the workforce tab, open up the meeting
33:54 setter. We'll make some space and then drag an agent card onto the canvas and
33:59 select the orchestrator. Now, if we zoom into the prompt here, we can start
34:03 perfecting it for our needs. I see that its role is a meeting coordination AI.
34:08 I'm going to add the word agent just to be super clear. And it correctly says
34:12 that it receives inbound meeting requests and orchestrates a team of
34:17 specialized sub aents. And the instructions here are that it analyzes
34:21 those meeting requests, coordinates with the sub agents, waits for all the
34:25 participants, and keeps the user informed and handles any errors. So, it
34:29 looks like it got everything that we had requested. It also added in this
34:34 expected input format, but we don't need that. So, I'm going to actually delete
34:39 that out and exchange it for an example input message specifying that these are
34:44 messages from someone within a company wanting to meet with a teammate. the
34:48 full team or people from outside of their company. And then in here, I'll
34:52 just give a few examples of potential messages like scheduling meetings with a
34:57 person with an entire department or someone external from the company to
35:00 pitch them. And I'll add an important note here that for any mentions of
35:05 pitches, proposals, etc. Gamma should prepare a presentation. Now, as we
35:09 scroll down here, we see this sub aent coordination workflow. All these steps
35:13 are good, just like we wanted them to look. I'm just cleaning up the
35:16 formatting here. The only change I would make here is that we want to clarify
35:20 that it is Borealis, the meeting booker, that tells the note
35:26 taker to attend the meeting. It's not the orchestrator himself who does that.
35:31 And that's because Nate, the notetaker agent we will build later, is actually a
35:36 next step after Borealis, the meeting booker, meaning Nate is not a sub agent
35:41 of the orchestrator. We will also remove the Slack tool and just say to use Slack
35:45 because, as you'll see, since we'll be triggering this from Slack, Orion will
35:49 already have access to Slack through that trigger. Now, we'll just clean up
35:53 this formatting so it's easier to read. And because this orchestrator won't have
35:57 access to our notetaker agent, I'll just delete this step number eight, which
36:00 talks about the orchestrator being able to control it, which is not true. And
36:05 here's talking about how it has access to the knowledge bases, but actually its
36:11 sub agents do. So, we'll fix that. And with that, the prompt is ready. But
36:15 you'll notice that there's this warning on this tool. That's because we removed
36:20 the Slack tool from the prompt. So, it's essentially warning us that we have a
36:23 tool that we're not using. So, we'll just delete that out from the tools tab
36:27 and go ahead and save the changes to this agent. Now, we can wire this
36:30 orchestrator up to its sub agent. Starting with Polaris, we'll make sure
36:34 that the connection type is AI connection, which means it's a sub
36:38 agent, and give instructions for how it should be used to find internal meeting
36:43 participants. And then we can set a label here. We'll make sure that it auto runs. We could
36:49 have the approval required or let the agent decide, but we'll keep it on
36:53 automatic. Then we'll wire Orion to Borealis, giving instructions for how to
36:57 use it when the meeting is ready to be booked and set the edge label to book
37:02 meeting. In order to trigger Orion, we need to wire him up with this message
37:05 trigger. And we can test this out from the run tab. We'll make sure to save and
37:10 publish the workforce and then give it a message to book a meeting with my head
37:14 of content for Friday. And in the timeline here, Orion is working. It's using Polaris to find the
37:21 participant. is using Borealis to book the meeting with a Google Meet
37:25 scheduling tool. And voila, it's successfully worked and booked that
37:28 meeting with the right person. So, our workforce is in a really good spot so
37:31 far. If we head back to the build section, we can add another way to
37:35 trigger our workforce from outside of the relevance platform.
37:39 Specifically, we want to be triggering it from Slack. We already have our Slack
37:44 connected here. If you did not already, you could just connect it through here.
37:48 just put in your Slack organization here and go through the connection process.
37:52 And if you're not familiar with Slack, it's really a collaborative
37:55 communication environment for teams to use. So, I'll hit continue and then I'll
37:59 add a keyword here. I'll call it booker. The keyword is really just a word that
38:03 we use whenever we trigger this. So, that relevance knows which workforce we
38:07 want to run. Now, I'm just telling it where it can be triggered from from my
38:11 direct messages and all of these channels. and I'll confirm that I understand that
38:17 I always need to tag at relevance AI within Slack to use this trigger. If I
38:21 wanted to, I could enable specific working hours for this Slack trigger and
38:25 therefore for this workforce, but I don't want to limit it to specific
38:29 hours. Now, I'll just connect up this trigger to Orion, but now be able to
38:33 save the workforce, then head over into my Slack organization and make sure that
38:39 I have the relevance tool installed. So, I'll click on apps and make sure I
38:44 have it installed. If you don't yet have it installed, you'll want to open the
38:48 marketplace. Search for the relevance tool and then install it from here. But
38:52 since it's already available within my Slack, I should be able to go into my DM
38:58 and then say at relevance AI and then book it. That's that keyword that we set
39:02 up and say book me a meeting with my head of content for let's say Monday
39:07 morning and now over in relevance. That should be working. It's going to be
39:11 taking a while, so I'll speed this up for you, but eventually it's going to
39:14 give you a reply back. If you pop that open, it should say something like the
39:19 meeting was successfully scheduled and here is the meeting link.
39:24 Now, if we head back over into relevance and click on the run tab, we can see
39:29 here is that task which we can tell was triggered by myself from Slack. Now, as
39:33 I've been alluding to, sometimes our meetings are going to require us to
39:37 prepare presentations, pitches, proposals, and for that we're going to
39:47 So, let's head over to the relevance marketplace to get started with that. As
39:51 you'll see here, if we search for gamma, there are a couple agents that have
39:55 already been built that we can make use of. We're going to uh select the gamma
39:59 assistant here and go ahead and clone it. and we'll give it a more specific
40:03 name for our usage and call it gamma the presentation preparer. Now, when we save
40:07 it, we'll see that it's asking us to fill in this variable for the gamma API
40:12 key. I've already done that and I'll show you how to do it in a moment. So, I
40:16 can hit continue, then head back into the workforce and put it to work within
40:20 here. So I will drag on a new agent card and select the gamma agent and wire it up as
40:26 a sub agent to the orchestrator and give it instructions for how to use it where
40:31 we'll call gamma whenever a presentation is requested. Now we'll change this
40:34 label to something more informative such as prepare presentation. And now we're
40:39 set up to request presentations to be made directly from within this
40:42 workforce. But before we actually do that, I want to orient you to the Gamma
40:46 platform. We're happy to be a partner of theirs because they provide a powerful
40:51 way to bring your ideas to life where you can use prompts to create
40:55 presentations, branded documents, social media content, and even websites. And
41:01 now with their new API feature, we can automate the creation of this content
41:06 from places like relevance n make or even your own custom web apps. For
41:10 example, you could have a system where whenever a blog post is approved for
41:14 publication, Gamma runs automatically, creating a graphic that is perfectly
41:18 suited for that post. So, the possibilities are really endless here,
41:22 and it can save you and your team a bunch of time. Of course, in our
41:26 workforce, we're specifically interested in creating presentations, and Gamma is
41:30 capable of creating really any kind of presentation from a pitch deck to
41:34 something more personal to an internal team update to a keynote speech or
41:38 whatever kind of slide deck you need created. It's going to do it very fast
41:43 with little or no effort from you. And you can use AI to refine and polish the
41:47 presentation until it's ready to present. As you can see in their video
41:51 here, it all starts with a prompt which you can feed in from the actual Gamma
41:56 interface or programmatically from whatever workflow or app you're using.
42:00 Then Gamma is going to generate the entire presentation for you styled in a
42:05 theme of your choosing or you can create your own theme based on your branding.
42:09 Once it's ready, you can edit the presentation with AI, changing up the
42:15 verbiage or the layout itself. And you can drag and drop different elements
42:18 onto the slides until you're ready to share it publicly and present it out
42:23 into the world. If you don't yet have an account, you'll want to create one now.
42:27 But I'm going to log into my existing one cuz I want to show you some
42:29 presentations I've already generated from the API. Let's open one of these
42:33 up. And you'll see I have all of these slides. The text is based on internal
42:39 documents that I gave it access to. And it generated all of the images and
42:44 design itself based on the prompt. And if I wanted to change things like this
42:48 image here, I could regenerate some new AI images and change what I'm using
42:53 within the presentation. So, not only is it quick to generate, but it's quick to
42:57 iterate as well. And all of my slides here are using the theme Oasis. So, I'm
43:02 excited to show you how to create these from within the relevance platform. Now,
43:05 it's important to note that in order to use the API, you do have to have a pro
43:10 plan within Gamma. But when you consider all the time and energy you'll be saving
43:14 with a tool like this, it provides a lot of value. So, with that pro plan, you'll
43:18 be able to go into the settings here and generate an API key to use within the
43:22 relevance platform or wherever else you might want to make use of Gamma. You'll
43:27 just make sure that the gamma API tool in the agent has access to that API key
43:31 which you can set up within the integrations tab of relevance and add
43:36 that API key inside of here. So now that we're oriented to the power of gamma and
43:41 have it integrated into our agent, let's get clear on how this agent is
43:44 functioning. So under the core instructions, you can see this
43:47 documentation variable. So if I go to the variables tab, we'll see that we're
43:52 feeding our agent. uh all of this uh documentation that explains how to use
43:56 the gamma API and it even includes all of the code for an example request.
44:03 Note here that the theme name is Oasis like I showed you in the gamma platform
44:07 and the number of cards in the presentation is 10. These are all things
44:10 that we have control over from the prompt itself. The only thing I'm
44:14 changing here is I'm going to increase the duration of the delay and the amount
44:18 of times we're going to pull for the presentation to make sure that our
44:22 workforce can gain access to it and it doesn't time out too early before the
44:26 presentation is ready. And since we want to give Gamma access to some internal
44:31 documents to build presentations from, we're going to add a knowledge base and
44:34 sync up to Google Drive. We'll just select our connected Google account,
44:39 specify the drive itself, and whatever document we want to give access to. In
44:44 this case, it's information about how my agency provides AI transformation
44:49 partnership to clients. Since that's not a huge file, we will just add it all
44:53 into the prompt itself. In order for the agent to make proper use of this, we
44:57 just need to let it know that when it's researching and preparing for the
45:01 presentation, it should ask itself, does it need access to the AI transformation
45:06 partnership information? If so, it should reference that knowledge base
45:09 which is attached to the bottom of this prompt. So, if I scroll down here, there
45:14 it is. In order to empower even more, we're going to let the agent know that
45:17 as needed, it can perform Google research to search the meeting
45:21 attendees, the company topics, etc. and it can use a tool to scrape the content
45:26 from a website to discover helpful information about the website of the
45:30 person or company that is building a presentation for it. And with all of
45:35 that set up, our agent is empowered to prepare a presentation on a bunch of
45:39 different topics. If anything went wrong, let's say an AWS outage causes
45:44 gamma to not be functioning temporarily, we can set up what is called an
45:47 escalation, which means the agent will notify us via a method of our choosing
45:52 that there are issues that need addressing. So, we can set up our agent
45:57 to notify us via Slack whenever a task has timed out, for example, or there's
46:02 an unreoverable error or it's exhausted, it's retries. In any of these scenarios,
46:06 we can make sure to be notified within our connected Slack account at the
46:10 destination of our choosing. For example, through a direct message to
46:14 myself. We could also choose to be notified via email as well. So, we'll go
46:19 ahead and save this agent and make sure that it's working by heading over to the
46:23 run tab, saving our changes to the workforce, and asking our Gamma agent to
46:28 build a presentation on how Morningside AI can be an AI transformation partner
46:33 to this example company. I'll give it a URL. Now, as soon as I hit go, we can
46:38 see Orion has called the gamma agent. It's doing its research and planning,
46:42 which looks pretty thorough. It's thinking through how to structure
46:46 the presentation and planning how it's going to make use of the gamma API
46:51 itself. As it's calling that API, it looks like it ran into an error first,
46:55 but it keeps trying using delays during those attempts. And of course, I'm
46:59 speeding this up for you. and it eventually found success and it returned
47:03 this message that the presentation is ready including a summary of what it
47:08 covers and a link to the presentation itself. So if we click that open we can
47:13 view what it built for us which is already looking great. If we wanted to
47:19 we could tweak this and then on the day of open the presentation link and
47:22 present it live. So, this is a super quick and efficient way to get custom
47:27 pictures, presentations, and even proposals generated for anyone that
47:30 you're meeting with. Speaking of which, let's head back into relevance and
47:34 create our next agent, which is going to be responsible for locating our leads
47:38 within a connected customer relationship So, let's start building the next agent,
47:49 the lead locator. This agent is going to help us locate external participants so
47:54 that we can book meetings with people from outside of our company. These could
47:57 be people we want to pitch our business to and turn into clients. Our leads are
48:02 going to live in a tool called HubSpot. So, we'll switch over to that. Now, if
48:06 you haven't heard of it yet, HubSpot is a customer relationship management
48:11 platform or a CRM where businesses store all their customer lead information. You
48:15 can think of it as a central database for everyone that you're doing business
48:19 with or people that you want to do business with. You'll see leads
48:23 organized under contacts, companies, deals, tickets, and orders. But we're
48:27 mostly concerned about our contacts here. This is where the lead's
48:32 information will be stored, including their name, email, phone, and their
48:37 company. These fields are actually linked to their full company data, which
48:42 we can view as a list over in this company's tab. I'm walking you through
48:46 this because our lead locator agent will search through HubSpot and retrieve the
48:50 contact information for the leads we want to book meetings with. Now that you
48:53 understand what exactly we're asking for, let's go back to relevance and
48:57 start building out that agent who will be asking for these leads. We'll build
49:01 the agent from scratch. I'm choosing to name this one Lania the lead locator.
49:05 And we're going to give it a short description to locate external
49:10 participants from outside of our company. Okay. and we'll come down here
49:15 to the prompt and define its role, telling it who it is and specifying that
49:20 it's a sub agent of Orion, our orchestrator agent. Now, we'll tell it
49:24 what to expect as an input, which will be information about the lead, like
49:28 their email and the company they work for. And for added context, we'll tell
49:32 it that this is a participant of a future meeting. As for its
49:36 responsibility, we'll instruct it to locate the info with HubSpot. We can
49:41 access tools by typing a forward slash then the kind of tool we're looking for.
49:45 In this case, HubSpot. We have a few options to choose from. So, we'll select
49:49 the tool to retrieve contact details from HubSpot. Since this is an external
49:53 integration, we need to connect to our account so that we can have access to it
49:57 from within this agent. Now, let's move down a little bit here and continue its
50:00 instructions, letting our agent know that we want it to perform research on
50:04 this lead and their company. And then we're going to the next tool. We're
50:07 going to use LinkedIn. will use this LinkedIn tool to search for info from
50:12 their personal and company profiles. For more thorough research, we're also going
50:16 to use Google search. So, we'll add in a Google search tool here as well. This
50:20 research will be used for things like creating personalized pictures and
50:24 eventually proposals. Finally, we'll define the agents output and tell it
50:28 that it must return the lead's complete information and a research summary to
50:32 its boss, agent, Orion. And of course, it should let us know if there are any
50:36 errors or issues with locating the lead. Okay. And that's our prompt. By giving
50:41 it these tools, we've connected it to HubSpot and enabled it to search through
50:45 both LinkedIn and Google. So, it's pretty powerful with only a few lines of
50:49 prompt. I'll make sure to save this agent, then go into the run tab to test
50:53 it out to ensure everything is working. From this run tab, I'll pass in the
50:57 contact info for a potential lead. In this case, someone in my network that I
51:01 added into my HubSpot contacts. This way I can make sure this agent is not only
51:06 finding the correct contact but doing accurate research on them. Once I enter
51:09 the email you can see it started running. If you're ever wondering if the
51:14 agent is actually working on the right hand side here you can see the status
51:18 will update to running and it also shows the tools that it's going to use. So we
51:22 can see it is definitely working. It's looking into HubSpot. Then it's grabbing
51:27 info from LinkedIn profile. Now we can see it's searching Google. Looks like
51:30 the first one failed, but that's okay because it ran multiple searches, which
51:35 is great to see because that means this agent, like all agents you build in
51:40 relevance, is able to adapt on the fly. And once it completes, we can scroll up
51:44 here and see we have the correct contact details from HubSpot. It's grabbed a
51:48 bunch of details from LinkedIn about this uh lead and his company and
51:53 structured it out quite well, too. This is cool because our other agents like
51:56 Gamma will now have access to all of this context as it builds out super
52:01 custom presentations. For good measure, I'll mark this as complete. Now that
52:05 it's built and we've confirmed it works, we're ready to add this agent into our
52:09 workforce. So, we'll head out of the agent and head back to the workforces
52:13 tab and go into our meeting setter and drag another agent block onto the canvas
52:17 and choose Lania. We'll connect Orion down to Lania and make sure it's set as
52:22 an AI connection since this is another sub agent of Orion who will of course
52:26 call her to locate external participants. Now, I'll just spread them
52:29 out a little bit here to clean up the canvas. And as a final step, I'll update
52:33 the label here to find external participants. I'll make sure to save the
52:37 workspace because we're going to head out of here and go find our next and
52:42 final agent, the notetaker agent, who As we've seen, the marketplace
52:52 conveniently has a bunch of existing agents we can repurpose for our needs.
52:56 So, if we search for a notetaker, we'll see a few options. Nate fits our needs
53:01 quite well, so we'll select him and go ahead and clone it. Its current prompt
53:05 is pretty useful, so we don't need to start this one from scratch, and we can
53:09 reuse some of what is already here. But, we do need to change a few things. And
53:14 we'll start by clarifying its role is a notetaker agent who attends meetings, transcribes what
53:20 happens during that meeting and afterwards assigns tasks and sends a
53:24 meeting summary. As for its input, that will be information about the meeting
53:28 the agent must attend received from Borealis. The meeting booker agent will
53:32 tell it that it's responsible for generating summaries of all types of
53:36 meetings from internal team syncs to client calls and project updates or
53:41 strategy sessions. Inside of its instructions, the agent is being told to
53:45 use this send meeting bot tool to record and transcribe the call. And then when
53:49 the agent receives that transcript, it reads it carefully and writes a summary
53:52 of that meeting, including things like key discussion points, decisions made,
53:57 and action items. Then it's able to use this turn transcript into text tool to
54:02 generate a file of the transcript. And we're telling the agent to refer to the
54:06 transcript whenever a user asks questions about that meeting. I'll add a
54:09 bit more to the prompt here so the agent knows to determine the action items from
54:14 that meeting and then create tasks based on those actions items and create tasks
54:19 for them over in our to-do list in Trello. If you're not familiar with it,
54:24 Trello is a visual task management tool. Uh, and it's used to organize personal
54:28 and professional projects. It's kind of like a digital bulletin board with
54:32 sticky notes. We find those boards in this tab where we've got out to-do list
54:36 board here. Opening up that board, you can see that it has lists arranged in
54:41 columns, and these lists get tasks or cards added into them. You can name the
54:45 lists whatever you want or add new ones and then drag cards across them as your
54:50 tasks move through different statuses. Each card can be opened up and contains
54:54 a bunch of properties you can set up. Each one of our tasks that the notetaker
54:59 agent creates for us will be a new card, which is added to this to-do list board.
55:03 So, back in our agents prompt, we're telling it we want to use the Trello
55:07 tool to create a card on a Trello board. Here, we need to create a Trello
55:10 connection, and we're just going to link it through to our Trello account. We
55:15 need to authenticate with Trello. Just scroll down here, and we'll just allow
55:20 that connection, and then click continue and continue again. We scroll back down
55:25 and now we can see that's enabled. And now we can continue. When it creates the
55:30 card, it should fill in the following: the card name, its description, and its
55:35 due date. That's it. Finally, for the output, we're going to tell it to email
55:39 a summary to all participants using their emails they attended the meeting
55:43 with and include a link to the Trello board that we just added tasks to. And
55:48 it will do this using the send Gmail tool from our connected account so it
55:52 knows how to structure the email. We'll give it a format to follow, which I'll
55:57 just paste in here. and we'll include a sample email as well.
56:04 Great. So, we'll save that agent. Lastly, we need to add a trigger. So,
56:07 we're going to add relevance meeting bot and click setup trigger. And that's been
56:11 added in. Noteaker agents are a little different and need to be triggered
56:15 twice. Once to join the meeting, which in this case, the Borealis agent will
56:19 do, and once when the meeting is finished. In order to trigger it for the
56:22 second time, you need to add this meeting bot trigger to the agent. Now we
56:27 can add it to our workforce. So let's head back over there. Inside our
56:32 workforce, I'll drag on a new agent. Tell it to use Nate the note taker. And
56:36 for the handoff type, this one should be next step. This means that instead of
56:41 the orchestrator agent managing this agent, Nate runs automatically once
56:46 Borealis books the meeting and hands him the meeting link. For good measure,
56:49 we'll edit the label here and specify that we're sending a meeting link on
56:53 this handoff. Now, we're good to go. So, Okay, we've built all of the agents and
57:04 now we can see them working together as a workforce. So, let's jump over to
57:08 Slack. This is where we're going to trigger the workforce. Call the relevant
57:12 AI bot. And then our particular trigger also needs this trigger here. So, we're
57:17 going to say booker. I'm going to say book me a meeting with Adam. And we'll put
57:24 Adam's email in here. next Tuesday at 10:00 a.m. Eastern Standard Time and create a presentation
57:30 about how we can be their AI transformation partner and share it as a
57:34 link in the meeting. Okay. So, we're going to click to send that and straight
57:38 away it responds. So, we can look at that and view thread. So, we've got that
57:41 confirmation here that the agent is cooking up a reply. So, let's jump over
57:46 to relevance AI. We will have a look at the run tab and we can see that has
57:50 already started here and it's already running. Lania the lead locator is
57:54 working and then we are getting the LinkedIn details and we're doing the
57:59 Google search and then book the meeting. Okay, now moving on to creating the
58:09 Now that's completed. So let's jump over to Slack again and we'll just look at
58:13 the reply. So it's created how Morningside AI can be this company's AI
58:18 transformation partner exactly as we asked it to do. So, it's gone through
58:23 and created this detailed presentation. Might need a little bit of a review and
58:26 then edit, but that's certainly a good starting point. Okay. So, let's close
58:31 that. Now, let's jump over to Google Calendar. And we can see that the
58:36 appointment here we need to say yes, we are going to attend. So, let's join with
58:40 Google Meet. Inside Google Meet here, we can see Nate wanting to join. So, I'll
58:44 admit him. And there he is taking all of the notes that I need. He will write it
58:49 all down, email it to us, and assign tasks as we will see later. After the
58:53 meeting, let's head back to relevance. We've seen that complete. We can jump
58:58 over to Trello, and we can see we've got a couple of tasks here that have been
59:01 created, and we can jump up to our email. There's a summary with the key
59:05 points, the decisions made, and then action items and next steps. This was
59:09 sent to all participants automatically after the meeting ended. So, there you
59:14 have it. The workforce works flawlessly. Now that our workforce is complete, we
59:23 can submit it to the relevance marketplace and look at how we can start
59:28 to monetize this. So under more actions, we'll click submit to marketplace and
59:31 select some categories that our workforce relates to, such as sales,
59:36 research, and operations. Then we'll add a description for the workforce, telling
59:40 the public that this workforce helps you book meetings with your teammates and
59:44 contacts and generates custom crafted presentations to use during those calls.
59:49 Then we'll decide if we want to submit this as a free or paid workforce. I'm
59:53 going to submit mine as free, but of course you can set yours as paid and
59:56 start to earn some money. You could allow sharing it as a clonable template
60:00 or allow republishing, but I'll leave those off for now. Then on the next
60:04 step, we just select one of our past tasks to be the preview that shows up
60:08 when this is published. Then once we submit for review, it's going to run
60:12 through some automated checks that I'm speeding up for you. And once those
60:16 checks have passed, you can click finish and you should see your builder
60:20 dashboard where the approval status is hopefully pending. If it was
60:24 autorejected, you might have to resubmit it with some changes. Clicking this
60:28 dropown, you'll see all of the agents and tools that this workforce was
60:32 submitted with. And if we click on the submission itself, we'll get a preview
60:36 of how this would show up within the relevance marketplace. As a builder, you
60:41 have your own profile. You can set up a cover photo and add an image of
60:45 yourself. And in order to receive payments on the platform, you'll just
60:48 link a Stripe account. You'll click generate Stripe link. If you don't have
60:52 a Stripe account, you'll want to create one first and then just enter your
60:56 Stripe email address and continue the process from here. And with that, you've
61:00 taken the first step of monetizing your new skills as an AI agent workforce
61:05 builder. There are so many opportunities for making money with these new skills.
61:09 So, join me in the next and final chapter where we talk about the
61:17 So, we covered a lot of ground on this course. You now understand the massive
61:20 AI workforce opportunity, a market that's predicted to reach 52.6 billion
61:25 in 2030, and my businesses are desperate for these kinds of solutions. More
61:27 importantly, you've learned how to design and deploy complete AI workforces
61:31 that operate around the clock, automating, delegating, and executing
61:35 work that once would require entire teams. But here's the truth. Knowledge
61:38 without action is worthless. The knowledge gap that you've built through
61:41 this course is your new competitive edge. It's your path to financial
61:44 freedom and building the life of your dreams. Because what you now possess
61:47 isn't just a technical skill. It's a business superpower. And the ability to
61:50 spot inefficiencies and design intelligent systems that replace manual
61:55 processes with scalable AI workforces is exactly what businesses these days are
61:58 dying to pay for. In this monetization section, we'll turn that skill set into
62:01 income by building a scalable monetization engine around your
62:04 expertise. So whether you start as a solo freelancer or launch your own AI
62:09 agency, this is the path forward. The AI workforce monetization model. Your
62:12 journey to monetization follows a simple progression. Diagnose, design, deliver,
62:17 and scale. Step one is diagnosing the opportunity. So, every engagement with a
62:20 business begins with what I call an AI workforce audit. This is a strategic
62:24 diagnostic where you step into a business and uncover where complete AI
62:27 workforces can replace their manual workflows. You'll map out how they
62:31 operate, expose inefficiencies, and show what's possible with a coordinated set
62:34 of agents. The result is a road map that highlights exactly where automation can
62:38 deliver a measurable ROI. This instantly positions you as a trusted adviser to
62:41 the business, and they will pay you thousands of dollars for this clarity
62:45 and this audit before you even begin implementation and building their
62:48 workforce. So, I've documented the whole process of how me and my team at
62:51 Morningside AI do these kinds of AI audits. And I've boiled that down to a
62:54 free resource that you'll be able to get on my free school community. That's
62:57 called how to perform your first $10,000 AI audit, which is going to be available
63:01 for free with the other resources that you found already, which goes step by
63:03 step through how to perform your first AI audit. And that's available with the
63:07 other resources in the classroom on school. Step two is design and
63:09 implement. So, once you've done the audit and the opportunity is mapped,
63:12 it's time to start building. And this is where your technical and creative skills
63:15 have to kind of merge to turn that insight into impact for the business.
63:19 Here you're not just selling features, you are delivering a transformation for
63:22 them in their specific workflows. So a single chatbot might be worth a few
63:26 thousand, but a full AI workforce that replaces an entire role or even an
63:30 entire department justifies much higher value projects and ongoing retainers for
63:34 you as well. Step three is manage and scale. So deploying the AI workforce is
63:37 really just the beginning and the real opportunity lies in the management and
63:41 optimization. the maintaining, improving, and expanding of these
63:44 systems over time. And this is where the recurring revenue or monthly revenue can
63:48 begin for you and your business. So you can oversee multiple client workforces,
63:51 offer ongoing analytics, and evolve their systems as needs change. So as
63:55 your demand grows, you can scale by hiring other builders underneath you,
63:58 extending your capacity to deliver while multiplying your impact and income as
64:01 the founder. This is typically known as an AI automation agency where you have
64:05 yourself running the business, but other developers and automation experts
64:08 underneath you and actually doing the work for you. So, that's the model, but
64:11 how do you put this stuff into motion? It of course starts with landing your
64:15 first clients. And even if that means doing your first few projects for free
64:18 in order to build proof, experience, and confidence. So, who are you actually
64:21 selling to? Well, the businesses that need the AI workforces the most are
64:24 typically service companies or small businesses drowning in manual processes.
64:28 You've got contractors buried in coordination, agencies juggling client
64:33 workflows, consultants overwhelmed by admin. They need 24/7 operations, but
64:37 can't afford full teams. More often than not, they rely on repeatable multi-step
64:40 workflows, which are the exact kind of things that AI workforces are designed
64:44 to automate. So, your ideal clients are businesses with clear and repeatable
64:47 processes that currently require multiple people to execute. Step one in
64:50 getting your first clients is warm outreach. So, you of course want to
64:53 start with people who already know and trust you, and those are found in your
64:56 existing network. So, go through your contacts, your emails, your LinkedIn,
64:59 your social connections. And the key here is that you're not selling to them
65:01 directly. You're asking them for referrals, which essentially lowers the
65:05 pressure and expands your reach. A simple message could be, "Hey name, I'm
65:08 building AI workforce systems that automate operations like customer
65:11 support, scheduling, and project management, and I'm offering a few
65:14 complimentary workforce assessments in order to build a few case studies. Do
65:17 you know anyone who might find this valuable right now?" So, this approach
65:20 works great because it's personal, it's low risk, and it opens doors naturally.
65:23 It helps you to refine your pitch and build your first stories that prove your
65:26 expertise. Inside my free school community, you'll find the full landing
65:29 your first client in 30 days guide, which is complete with the exact
65:32 tracking template that's used in my AAA accelerator program. And this exact
65:35 template and system has helped thousands and thousands of people to book their
65:39 first AI clients. And I mean, my free community itself is a gold mine for
65:42 getting clients. With over 250,000 members, many of them being business
65:46 owners and decision makers by engaging authentically and sharing insights and
65:49 just offering value. It can lead directly to client opportunities within
65:52 the community. Step two is cold outreach. So once you've landed your
65:55 first few clients, it's time to expand systematically with my cold email
65:58 testing framework, which is a process that basically replaces the guesswork
66:01 with data. My full guide on how to do this is also available for free in the
66:04 school community in the monetization section of this course. But the strategy
66:07 roughly goes like this. So you're going to choose four service niches. For
66:11 example, you've got HVAC, you've got law firms, you've got real estate agencies
66:14 or consultants. Then you're going to write one clear results driven email per
66:18 niche. For example, I help HVAC businesses replace their entire
66:21 scheduling and dispatch process with an AI workforce that runs 24/7 for less
66:25 than the cost of one employee. Then you send 500 emails per niche and track the
66:29 open rates, replies, and calls booked. And then after 30 days, you identify
66:32 which niche and message has performed the best and then you double down on
66:35 that. And yes, you can even build your own AI workforce to automate this entire
66:38 process. Within weeks, this framework turns client acquisition into a
66:41 predictable and measurable system that you can scale. The authority flywheel.
66:44 So once you're landing clients and getting results, your next step is to
66:48 turn that execution and results into exposure for yourself. So every build
66:52 and every win and every insight is a story worth sharing. So documenting your
66:55 journey on LinkedIn, YouTube or Instagram and mirror those same kinds of
66:59 posts that you put there inside my school community by posting these
67:02 updates and sharing lessons and showing your results in motion. It creates a
67:05 self-reinforcing loop because you're letting all of these people see you and
67:08 building credibility. You're showing proof, not just promises. It creates
67:12 feedback loops which help you to refine your message. It opens doors and others
67:15 start referring clients your way because you're making noise and letting people
67:18 know you exist. Community engagement and content creation are two sides of the
67:21 same coin. When it comes to creating content, you don't need perfection. You
67:25 just need consistent authenticity. So, document what's real. The builds that
67:28 you're doing, your before and after results. Your growth is an agency. In my
67:31 opinion, if you're trying to get businesses, the best platforms to go
67:35 onto are LinkedIn and YouTube. So, over time, the cycle starts to compound where
67:39 the results you get with your business, they start creating stories that are
67:42 worth sharing. Those stories that you share attract new clients and then new
67:45 clients create more results which fuel your reputation and you get better and
67:48 better. And that's essentially the authority flywheel. It's a system where
67:52 your work markets itself and your presence compounds into authority. So
67:55 now your acquisition engine runs on three cylinders. You have warm outreach
67:58 initially to build that early proof. Your cold outreach which allows you to
68:01 scale more predictably. Then community and content to establish a lasting
68:04 authority. So as mentioned the step-by-step guides on putting these
68:07 exact strategies into action, the warm outreach and cold email are included
68:10 with the other resources in the monetization section on my school
68:13 community. So getting the support you need. If you're feeling intimidated or
68:16 overwhelmed about this new party going down, it's natural to feel that way, of
68:18 course, because you're building something entirely new. And that's why I
68:21 created my free school community. It's at this point the largest AI business
68:25 community in the world with over a quarter million members who are walking
68:28 the same path with you. So, inside you'll find all of my best resources,
68:31 plus weekly live Q&A with me where you can ask me questions directly about
68:35 pricing and implementation, client acquisition, or anything else that
68:37 you're working through. You'll also be able to connect with thousands of other
68:40 builders and business owners who are sharing their wins and solving
68:42 challenges and just generally growing together as a community. Just recently,
68:45 we had an in-person event in the Sydney Harour, which was incredible. So,
68:48 there's also options to come out in person to hang out with me and the team.
68:50 But, if you're ready to dive into this and get results as fast as possible, my
68:54 AAA accelerator program is there for those who are serious about building an
68:57 AI business, and they want one-on-one support every step of the way. That's
69:00 really what the program is made for. So, me and my team will work with you
69:03 directly to implement these systems and get you up and running as fast as
69:05 possible. So, we've covered a lot in this video. You now understand how to
69:08 design and build multi- aent workflows that can replace entire departments for
69:12 businesses, not just their tasks. The knowledge gap you need to make money is
69:15 now there. The market demand is proven for these kinds of systems, and the
69:17 tools are ready. Plus, you've got all the resources you need to take action
69:20 inside the free school community. So, so the point of this video is that the AI
69:23 workforce revolution is here. The question isn't whether it's happening,
69:26 it's whether you'll be the one to lead it. So, your next move determines
69:29 everything. You can either be replaced by this stuff or be the one out there
69:32 helping businesses to adopt it. Join the community, get the support you need, and
$

How to Build & Sell AI Agent Workforces as a Beginner | FULL COURSE

@LiamOttley 1:09:37 17 chapters
[AI agents and automation][marketing and growth hacking][developer tools and coding][solo founder and bootstrapping][productivity and workflows]
// chapters
// description

📚 Get all the resources and prompts for this course in my Skool Classroom: https://bit.ly/48GLQOk Build your own AI Workforces with Relevance: https://bit.ly/liamottley-relevance Turn your ideas into presentations with Gamma: https://bit.ly/liamottley-gamma 📈 Become a Wildly Profitable AI Entrepreneur: https://bit.ly/47m5GMl 🤝 Ready to transform your business with AI? Let's talk: https://bit.ly/4qypPYD AI isn’t just replacing jobs — it’s creating AI workforces that replace entire departments w

now: 0:00
// related videos (5)
Two Ways To Come Up With Business Ideas
[community building and audience growth][market research and validation][revenue model and pricing strategy]
→
Episode 2 - Building a SaaS to $1k MRR
[community building and audience growth][market research and validation][revenue model and pricing strategy]
→
Building a SaaS to $1,000 MRR - episode 3
[community building and audience growth][market research and validation][revenue model and pricing strategy]
→
Someone Will Get Really Rich Doing This
[community building and audience growth][market research and validation][revenue model and pricing strategy]
→
Episode 1 - Building a SaaS to $1k MRR
[community building and audience growth][market research and validation][revenue model and pricing strategy]
→
// actions
← back to terminal open on youtube → @LiamOttley channel →
// tags
[AI agents and automation][marketing and growth hacking][developer tools and coding][solo founder and bootstrapping][productivity and workflows]