0:02 You've seen these people who are always first on a big story, instantly owning
0:07 the conversation, and you know that's how you build the authority that drives
0:12 real business. But at this point, it's an exhausting race to keep up. So, what
0:18 if you had an AI agent to scout the news for you and a second AI, your personal
0:22 ghostriter, to instantly draft the post in your unique voice? You'd go from
0:27 hunting for news to simply hitting publish. Your audience would quickly
0:32 notice you being the first to the conversation every single time. This is
0:36 the exact system we'll build in this video step by step turning you into a
0:42 top voice in your industry. Let's dive in. If you're new to this channel,
0:46 welcome here. I'm Simon Hardburg and I'm building Founder Stack, a collection of
0:51 five SAS tools specifically built for bootstrap founders. I use social media
0:56 to drive sales and I know better than most people how powerful a personal
1:01 brand can be and how big a difference it can make when you build authority with
1:05 your audience. So I want to have one AI agent that I can chat with and delegate
1:10 this task to. It should have a chat GPT like interface and based on a quick
1:13 description it should be able to delegate all other tasks to other AI
1:18 agents and we should end up with a post that's fully done and ready to publish.
1:22 So, the only times I'm needed is at the very beginning by describing the topic I
1:27 want to post about and at the very end by clicking schedule or publish and the
1:33 post will go out. Let's cook. Openai recently launched their agent builder
1:37 and their agent kit. And this is exactly what we need. So, let's head over to
1:45 Let's go to the dashboard and let's go to the new agent builder.
1:50 here. We'll see a few templates we can use, but in this case, we want to create
1:55 a new one. Now, in order for this to work well, we'll split up this workflow into
2:00 multiple AI agents with each their own responsibility and access to only the
2:06 tools they need to do that one job. So, let's set up our first agent. This
2:12 one is going to be a new scout. Let's give it instructions. The job of this AI agent is to search
2:18 for news related to whatever the user is asking for and present a list of
2:23 headlines that we can pick between. And we also want to add web search as a
2:36 Keep it broad and open and make sure the Perfect. So now we have one AI agent
2:44 that will find a handful of news headlines related to whatever topic the
2:49 user wants to post about. Now let's create another AI agent. We can do that
2:54 by dragging in another AI agent node here. Let's make sure to connect them.
3:03 This one will be our news reporter. Let's give it its instructions.
3:09 Its job is to pick one well suited news headline and further research it and
3:13 create one news report based on that headline. And we also wanted to find a good image
3:20 we can use for the post. Here we can include the output from the
3:38 And because we're putting that here, we can disable the include chat history
3:43 option here. In this way, we'll keep the context nice and clean. And also here,
3:48 we'll add the web search tool. We'll keep it wide open and have a medium
3:54 search context size. And because this task requires a bit more thinking, let's
4:05 There we go. Perfect. We now have two AI agents in place. And just to be safe, let's add a user
4:13 approval step here. We could remove this later, but I think it's nice to have
4:16 this step for a while until we battle And we'll see the AI thinking
4:40 searching for the news [Music] and there we go. It will write one wells
4:53 suited news report, find an image we can use and ask if we want to continue.
5:01 The first part of our Top Voice news agent is ready. We had the AI search
5:06 through various topics and pick one news to move forward with. And it created a
5:11 report and found an image. Now, we need to have another AI agent write a good
5:15 post about this for social media. And this part is actually much trickier than
5:21 you'd think. Please, please do not just dump this whole thing into ChatGpt and
5:25 ask it to write a post for you. It will come out super sloppy. And at this
5:29 point, the internet is so filled up with bad AI slop already. Trust me, people on
5:34 the internet can smell AI riding a thousand km away by now. No one's going
5:39 to read it, and it will make you look super silly and hurt your brand. That
5:43 doesn't mean you can't have AI ride it, though. But you need to lay the
5:47 groundwork. This next AI agent needs to learn about your brand, your specific
5:51 writing style, your tone of voice, and your brand values. It needs to
5:55 understand you and your brand well enough to write in a way that looks and
5:59 feels exactly like you would write it if you had done it manually. And this is
6:03 actually pretty complex. It involves fine-tuning an AI model, creating a
6:07 comprehensive brand brief, and building a quality assessment loop into your
6:12 system. If you want to learn how to do this, I have another video where I go
6:17 through this whole thing step by step. And because this video got so much
6:20 traction, we actually built this whole thing into one of the products we run
6:25 called FeedHive. So we have made this whole thing incredibly streamlined by
6:29 building the whole process into our tool. And we have built a custom MCP
6:33 server that enables you to set this up with other tools like OpenAI's Agent
6:37 Builder. I'll leave a link to the other video in the description below. But for
6:41 now, we're going to continue doing this using FeedHive, and I'll show you just
6:46 how easy this has become. Let's go to feedhive.com and sign up for an account.
6:51 The first thing we want to do is add our There we go. Now, let's go to the brand
7:01 profile page. Here, we start by picking a writing style that most closely fits with how we
7:07 normally write or how we'd like to write. I like this one that's already
7:13 selected, the vertical writing style. I think this one fits best with LinkedIn.
7:17 in these days. Then we're going to scroll down to the brand brief section
7:21 and we'll have AI generate this whole thing for us. So, select your socials
7:27 here. Feed will look through all of them to get a sense of what your brand is
7:31 like. I'll also add my personal website here. Awesome. Let's generate it. This might
7:38 take a few minutes, but we'll let it do its thing. Awesome. We can see now that
7:43 our AI put together a very comprehensive and detailed and very accurate brand
7:47 brief based on the information we gave it. Of course, you can edit this at any
7:51 time you want if there's something missing or something you think the AI
7:55 didn't exactly get right. I think this one is spot on though, so I'll take it
8:00 as is. Now, let's quickly test this out. We can go to the compose page here and
8:04 click this button to create a quick AI Awesome. Yeah, this definitely sounds
8:12 like something I would write. And it used the vertical writing style very
8:17 well here. Very cool. Now, let's go to the automation page. And let's create a
8:25 [Music] I'll call this AI news trigger. And I want to publish for LinkedIn and
8:38 Now, what happens when triggered? Well, we can choose to both create and
8:43 immediately publish a post. Though again, for now, I just want to
8:47 see the post myself before going out. So, I'll stick to create draft for now.
8:53 And we'll write this post with AI. So, That's it. Let's save and publish this
9:03 trigger. And this is everything we need from FeedHive. It's now ready to take a
9:09 request, including a prompt, and it will combine that with your brand brief, your
9:13 writing style, everything FeedHive now knows about you, and it will turn the
9:17 whole thing into a post ready post for social media. Now, let's go back to the
9:22 agent builder. We want to add another node after the approve step, and it
9:28 should be an agent node, too. Under Let's call this fhmcp server.
9:46 Now for this part, we need an access token. If we go back to feed hive, we
9:51 can go to the settings page. Then scroll all the way down until we
9:56 see the API key section. and copy it from here. And if we paste it here, we should be
10:12 Awesome. And we can see our trigger appearing under tools. And if we click here, we can quickly see
10:19 that it is indeed the AI news trigger we created just before. And because it's an
10:25 MCP server, OpenAI's agents already know exactly what to do with this tool and
10:29 how to use it to make a post appear in FeedHive. Now, because we added an
10:33 approval node previously in the workflow, we don't have to ask for
10:38 approval to also use this tool. Let's add it here. And that's it.
10:48 [Music] Let's give it a description here. Its job is to create a prompt for feedhive
10:55 and pass that along with the news report And here we need to tell it to use the
11:05 FHMCP server tool. And that's it. Your final workflow should now look something
11:09 like this. All right, time to test it out. Let's go to the preview mode and give this a
11:22 [Music] All right. And we can see the new scout is getting to work.
11:34 All right. So, it's suggesting a news about something with OpenAI making a
11:37 deal with AMD. Sure, let's take it. [Music] Nice. So, the prompt agent is now
11:45 preparing a prompt. Awesome. and it did a breakdown here about what the prompt consists of.
11:52 That's it. Now, let's check out the final result in FeedHive. Let's go to
11:58 posts. Look under the draft section. There it is. Wow. I mean, this is a great news post.
12:06 And we even have an image of Sam Alman here. Really cool. So, this is great. I'll
12:13 just go ahead and schedule this for next [Music] I know a lot of you will probably ask
12:22 why didn't I use NAD for this and I definitely could have. However, I think
12:26 for a workflow that starts with a chat conversation like this, I really think
12:30 this new agent builder from OpenAI is super promising. And if I were to build
12:34 something more complex, I could have had agent builder delegate to NAD first and
12:38 then have NAND delegate to feed high. So, it's not really a question of NADN
12:42 versus agent builder. I actually think they play very well together for more
12:47 complex workflows. Now, the easiest way to use this agent is to simply click
12:50 preview like we just did, but personally, I'd like a nice standalone
12:55 chat interface I can visit in a browser. So, let's just go ahead and publish this
13:04 We can get the workflow ID here. Let's head over to Lovable and ask it to cook
13:08 up an interface real fast. We'll use the OpenAI's chat kit. So, make sure to
13:12 include the documentation for Lovable to look up. If you're not already familiar
13:16 with Lovable, it's a tool that will build software using AI. And it's one of
13:21 my absolute favorite tools. I use it to build internal tools all the time. Now,
13:25 after a few minutes and maybe a little back and forth, you'll have a chat
13:29 interface that looks like this. And now, it's available on a website that we can
13:35 use more easily. There we go. Really a super cool experience working with agent
13:39 builder, agent kit, and feed hive like this. And right now, you can get
13:43 lifetime access to FeedHive as well as four other awesome AI tools through our
13:48 Founders Stack offer. Go to foundersstack.pro and for a single
13:52 onetime purchase, you can set up this entire system for yourself. With this
13:57 deal, we'll give you 50,000 AI credits to spend every month, which is enough
14:01 for hundreds of posts. And if this is not enough, you can always add your own
14:06 OpenAI API key and remove this limit altogether. all included for a single
14:10 one-time purchase. If you don't want to use Vehive for this, that's of course
14:14 totally understandable. In that case, I recommend you watch this video next.
14:18 Here, I'm breaking down the entire flow, including the things FeedHive does, so
14:22 you can use this with any social tool