0:09 All right, welcome everyone to the next live stream here. It's been about a
0:12 month or maybe even more than a month since I have done one and I have been
0:16 waiting a very long time for this. I've been hyping it up past couple of weeks
0:21 and this is going to be a good one. And so I'm going to give a couple of minutes
0:25 just for everyone to get in the stream. While we do that, uh, let me know where
0:29 you're from. I always like to kick off live streams by doing that because I
0:32 know we do have a very global community here on YouTube. So yeah, let me know
0:36 where you're from and I can highlight certain comments as well. I do this in
0:41 all my live streams too. Um where you can see like uh let me do this. Let me
0:46 uh get this here going so I can highlight comments. If I go to like my
0:51 scene for example, I'll show you this here. You can see that I can uh show
0:55 comments. So, as you guys are coming in saying where you're from, I'll highlight
0:59 some here. We got Greece, Brazil, Wyoming. Very, very cool. Hungary, North
1:08 Carolina, Germany. All right, another Germany. Sean asked, "How's the throat?" It's almost better.
1:15 You guys can probably tell that my voice is a little horsearse right now still.
1:19 And if you watched my video on Wednesday, you know that it was kind of
1:22 rough, but pushing through and I'm doing pretty good. So yeah, Japan, we got
1:29 Italy, Finland. It's always so cool to see where all you guys are coming in
1:33 from. So very, very cool. Switzerland, Taiwan, Ontario, Denver, Portugal, Costa
1:38 Rica. I'm not like showing all these on the screen, but yeah. Oh, very very
1:44 cool. All right. Nice. Another Brazil. Sweet. Cool. Sweden. Spain.
1:55 Dallas. Another Spain. Very very cool. Been sick here too. Uh you're talking
1:59 better than me. [laughter] I'm sorry, Joseph. That's unfortunate.
2:04 Yeah, [clears throat] I think it's kind of like a mix. Like I got under the
2:07 weather earlier this week and I'm feeling like 100% now, but it's just the
2:11 fact that I didn't give my voice a chance to rest because I was doing a ton
2:15 of workshops in the Dynamis community earlier this week and I have my YouTube
2:20 video and then um doing some prep for this as well. Um but yeah, it's
2:25 definitely not as bad as it was earlier this week. Like I could barely talk on
2:31 Thanksgiving. That was not fun. So I was with my entire family and then I'm just
2:35 like the silent guy in the corner. Um because it's hard for me to to talk. So
2:42 yeah, but almost better now. Let's see. Uh talk a little bit about
2:45 what's new in Dynamis. I want to go back. Yeah, we will definitely get to
2:50 that. So stay tuned. We'll be talking about what's new in the community
2:53 because of course we got the Black Friday Cyber Monday sale going on right
2:58 now. So, there is also going to be a special discount for Dynamus in this
3:03 live stream only on top of this remote agent coding system that you're looking
3:07 at right here. This is the other big thing that I'm giving away only during
3:12 this live stream. And so, that's kind of like the big hype that I've been
3:17 generating around the stream here is giving away this system. So, this is
3:21 going to be the focus of our stream mirror. So, I'm going to release this as
3:26 a public repository, and I'll actually do that live with you guys in a little
3:30 bit here. And then I'll show you how to set it up. I'll show you kind of how it
3:35 works and why I think it's really um a better system compared to something
3:40 maybe similar like maybe you guys have seen Cloud Code for the web or you've
3:45 seen Codeex Web before. These are other kinds of remote agentic coding systems,
3:49 but I want to talk about why what we have here is even better. [snorts] And
3:54 then we'll get into a live demo and we'll build something together using
3:59 this agentic application that I have built. And then we'll [clears throat]
4:04 have some time for Q&A as well. And so I really just love when we got the chat
4:08 going, everyone's asking a ton of questions and that's always fun too. So
4:12 we'll have a good amount of time for that towards the end of the stream as
4:16 well. And then because my voice is really hoarse right now, I will be
4:20 taking uh sit breaks as I call them in Dynamus all the time like I did just there. So going to try
4:28 to protect my voice as much as I can so it doesn't get any worse than this
4:30 because right now it's pretty manageable. But yeah, that's the agenda
4:34 for today. I'm very very excited. And let's see how many people we got live
4:39 right now. I'm going to refresh. Okay, we got 257 live currently. Pretty
4:44 good. Awesome. Yeah. Well, that's yeah, that's an honor. So, thank you to all
4:48 260 of you who are live right now from all across the globe like we saw. Uh
4:54 where I'm at right now, it uh started snowing about an hour and a half ago.
4:58 And it's very beautiful outside right now. So, trust me, here in Minnesota, we
5:03 get very sick of the snow in January and February, but for the the first couple,
5:08 it's really nice. And so, we had like one larger snow earlier this week, and
5:12 this is like the second one. Um, so yeah, I'm not sick of it yet. So it's
5:16 actually very nice. Like I took my dog out to the bathroom earlier today. Uh,
5:19 just walking around with all the snowflakes. You miss it, but then you
5:24 get sick of it in like four weeks. So yeah, but um anyway, let's get into
5:30 things here. And so the very first thing that I want to do is I want to make this
5:35 repository public and open it up to all of you guys to download this right now.
5:40 And like I said, end of the live stream, this is going away with one small
5:45 exception. I know that there are certain time zones that don't work for this live stream.
5:52 And so if you're in Australia or New Zealand, you're probably watching the
5:57 recording for this event. And so at 4:00 p.m. Central time on Monday, Cyber
6:02 Monday, I'm going to open up this repo again for like just an hour. So 400 pm
6:08 central time, you can use this link. Otherwise though, this live stream is
6:11 the only time that I'm opening this up. So I'm going to go into the settings
6:14 right now and I'm going to make this repo public because right now it is
6:19 private. So I'll change this to public. And there we go. Make this repo public.
6:26 And boom. All right. It's cool. So I'll put a link in the chat, of course.
6:31 So let me go over to the chat. Put this in for you guys. Boom. There we go. And
6:38 um yeah, so someone asked here, I'll go back to the chat. Two commits only.
6:42 Yeah. So the reason this has two commits is because this system is a part of the
6:48 Dynamus Aentic coding course. And so that is where I did all of the work and
6:51 then I ported this over to a repository to make public for today. So the live
6:56 stream is when I'm making this publicly available. But this is a resource that
7:00 I'm going to continue to evolve in the Dynamis community. So, I'll take a
7:05 second to talk about that as well and then we will get to our live demo of
7:10 this and I'll show you guys how to use it and just why like this is the future
7:14 of agentic coding because that's the other thing is I truly believe that
7:19 having a system like this where we are able to kick off our agents where we
7:25 work like Slack or Telegram or GitHub that is the true future of agentic
7:29 coding. And so yeah, I want to talk really quickly about the course because
7:35 this is the big focus of mine and also one of the reasons that I am doing the
7:39 Black Friday sale because I am celebrating the fact that we have the
7:44 Aentic coding course fully complete as of Wednesday of this week. So I
7:48 completed it just in time for our live stream here. So I'm going to go ahead
7:54 and put this in the comments as well. So the two primary links for this live
8:00 stream is obviously this special that we have for only this live stream and then
8:04 our remote agentic coding system that again is only for this live stream. So
8:09 it's a special event that we got today. And so the agentic coding course in the
8:14 Dynamis community, it takes you from beginner all the way to advanced
8:19 building your own system for agentic coding. And so we have 71 lessons in
8:25 total. This is a super valuepacked course and it's 18 hours in content
8:30 overall. A lot of it is optional. So if you just want to go through something
8:34 quicker, you definitely can, but we [clears throat] have a module that
8:37 introduces everything. Then I get into the piv loop, which is the core mental
8:41 model for agentic coding that I've covered a bit on my channel. And then we
8:46 get into global rules and commands. And then we get into the systems for each of
8:50 the different layers of our piv loop. So systems for planning, implementation,
8:56 and validating. And then we get into how we can incorporate coding agents
8:59 directly in GitHub. And then I have the module dedicated to the system that
9:02 we're going to be talking about in the live stream today. And so I go even
9:06 deeper into it here. And then this system is going to be something that
9:10 I'll keep evolving in Dynamis as well, adding in integrations for more
9:15 applications and more coding assistance. And later in our stream, I'll also talk
9:20 about the architecture that we have for this system and how I've made it really
9:25 really easy for us to add in more [clears throat] applications like Slack
9:29 or Notion or Jira and then also adding in more coding assistants like um Open
9:34 Code or Klein, whatever we'd want to incorporate in our coding system. That's
9:38 one of the benefits of this is it's going to work no matter the tool or
9:42 coding assistant that you want to use. And then we've got MCP servers and
9:46 skills and sub agents and parallel agent coding. Like everything that you could
9:49 hope to cover in a course on agentic coding, we have for you. And the
9:56 discount for this live stream only is 45% off the yearly, the annual
10:01 subscription and 25% off the monthly. And this discount is going to go away at
10:05 the end of this stream. And if you're watching the recording, there still is a
10:09 Black Friday sale that goes through Cyber Monday that is almost as good. But
10:14 this is exclusive to the live stream right now. And so I'll call this out a
10:18 couple of times throughout the stream, but the focus that we have here really
10:23 is going to be on the remote agentic coding system. So if you want to really
10:27 dive into using this more and watch it evolve and be a part of a thriving
10:31 community where we also have daily events and weekly workshops and the AI
10:35 agent mastery course as well. Definitely come be a part of Dynamis.
10:41 And so with that, let's get into things here. And so I want to start with a
10:47 quick demonstration and then we will get into how you can set this up for
10:51 yourself. So I'll go through the instructions very quickly. It's actually
10:55 really really straightforward and we can use our claude code or our codec
11:00 subscription. We don't have to use API keys. And I know that's a concern that a
11:03 lot of people have for this kind of custom system. They're like, "Oh, what
11:07 if I have to use my enthropic API key and then I'm paying, you know, hundreds
11:10 of dollars to use this throughout the month." But it's going to be just as
11:14 affordable as if you're using the primary service, like if you're just
11:23 And so I obviously have this up and running on my computer already. And so
11:29 I'll start by giving you a quick demonstration here within Telegram. And
11:33 then I'll show you uh GitHub in a little bit as well. But yeah, that this is like
11:37 the main thing is whatever application we want to work in, we get to use it
11:43 that use the agent directly there. And so I can start with um kind of like a
11:48 command directory here. So I have custom commands built into the system as well.
11:54 So you can really take advantage of the customization that we have here. Like
11:57 you can add in your own commands and your own ways to manage your
12:02 repositories and your agents. And so for example, we have commands that are built
12:06 directly kind of like claude code slashcomands, but we get to load and
12:12 create our own directly for our agent. And then we can manage our code bases
12:16 like we can clone our private or public repositories and we can list them out.
12:21 So for example, I can do /reos here to see which repositories from GitHub I
12:27 currently have in my system. And we add the GitHub token into our system here.
12:33 So it can use the GitHub CLI to manage our repositories and issues and pull
12:37 requests. And so this is very deeply integrated with Git, which I think is
12:42 really important because I think GitHub or another similar platform like GitLab
12:47 is definitely the future of remote agentic coding because it's kind of like
12:51 that orchestration layer when we can create issues for bugs or feature
12:55 requests. We can manage pull requests so that we can review the changes of our
12:59 coding assistant and the workflow that I'll demonstrate for you here in a
13:04 little bit. It has a human in the loop element where we get to review the
13:09 changes that we kick off remotely before it goes into our main branch and gets
13:14 deployed to production because I don't want this to be like a replacement of us
13:18 doing the code review. Now, it's going to be a little bit harder to dig into
13:22 the individual line changes because this is a remote system, but we still can
13:26 create as a part of our system a process for us to review things. And so, right
13:33 here, I'm currently in the um the repository that I'm using for our
13:37 demonstration today. And so, for example, just to show you how this works
13:42 really quickly, I can say summarize the readme. And this is going to be just
13:46 like using claude code or cursor or wind surf wherever where it's going to show
13:51 us the messages in real time as it is taking action for us like it's looking
13:55 for the readme and then it is reading it and then in a second here we'll get the
13:59 summary spit back out to us. And so literally anything that you can do with
14:05 the claude code I can do right here in Telegram. And I can even do it from my
14:10 phone if I want. Like I'll actually show you this right here. I'm going to go
14:15 over to Telegram on my phone and so hands off the keyboard. You're going to
14:18 see a message pop up for me right here, but I'm actually doing it from my phone.
14:23 So, I'll type this out here and I'll say uh summarize the steps to get the agent
14:30 up and running. All right. So, I'll send this in and then there you go. It's
14:34 showing up on my desktop Telegram app as well. And so that's the other beauty of
14:39 this is no matter what platform I'm working in, I am able to kick this off
14:46 from any application. Um, and I'm able to do it from any device as well. And so
14:50 there we go. We have the steps to get our agent up and running. Very, very
14:54 cool. And so I could have viewed this on my phone. I can have an entire
14:58 conversation on my phone. and the workflow that I'm about to show you
15:01 where we make a change to a front-end application. We could do the full end
15:06 toend validation oursel also from our phone. And that's one of the things that
15:09 I've been really thinking about and designing as I'm building the system is
15:14 like how can we include human in the loop when we're not working directly in
15:19 a codebase like operating in an IDE. And that's a tricky thing to solve, but I
15:25 have that at the top of my mind here. And so, uh, someone asked, "How much
15:29 does this, uh, every day cost?" And so, yeah, this is going to cost exactly as
15:34 much as your subscription to, um, Claude Code or Codeex or whatever. Like, it's
15:37 it doesn't cost anything more because it's not using your API. It can use your
15:43 Claude or Codeexcription. And so, yeah, I like basically never hit
15:47 rate limits when I'm using this. And so, I pay for the max plan for Claude Code.
15:50 And I'm just able to use this however much I want. We can even kick off agents
15:55 in parallel. Like I can have one agent that's working in Telegram, another one
15:59 that is working in GitHub, and then I can have another agent that is working
16:03 on a different GitHub repository. I can have even like a dozen agents working at
16:08 the exact same time going through this single remote agent coding system. And
16:12 that's another exciting thing as well is just the idea of parallel execution
16:17 without us having to really set that up ourselves. We can have that as a part of
16:22 our custom system. And when we have these kind of like pre-built solutions
16:26 for us like cloud code for the web or codeex web, they have that idea of
16:30 parallel execution as well. But what these don't have is the ability for us
16:34 to include our own custom commands. And in cloud code for the web, for example,
16:39 we can't use MCP servers, which is a big bummer. If we want to use the playright
16:43 MCP to have it autonomously validate a website, or the superbase MCP, so it can
16:49 manage our database schema, you can't actually set that up here. But when we
16:53 manage things in our own custom solution, uh we can actually like go in
16:58 and add our own MCP servers. We have full control as if we are using the
17:04 cloud code CLI or whatever CLI, but we still can use it from any device, any
17:10 application. Um and so yeah, like this is quickly starting to become my primary
17:16 driver for just making like any changes overall to my code bases. is like when I
17:20 have a new idea for something that I want to try in an app, I'll just like go
17:25 to Telegram or go in GitHub and then I'll send off a request and this will
17:29 sometimes take a while just like it could take a while when you're running
17:33 directly on your computer. So you send off the task and then you just come back
17:36 like you'll get a notification on Telegram or on GitHub like oh your plan
17:40 is ready or oh your implementation is ready and then you just go and check it
17:43 and so you like send off the request and just come back whenever it's done. And I
17:49 will say that like for super super complex implementations when I'm really
17:53 getting into like starting up a project from scratch, I'll probably still work
17:58 in a local development environment. And the reason for that is like I might want
18:02 to look more deeply at the code as it is starting things. But when I have a
18:06 system built that I really trust like my process for planning then implementing
18:11 then validating is really really robust and like I know the kinds of things it
18:15 can handle in one shot. that's when I'll plug it into this system and then I can
18:19 just send off a single request and have a lot of confidence that it's going to
18:23 work but then still also be able to validate things through the uh kind of
18:28 human in the loop elements that I build So yeah, so let me actually show you guys a quick
18:40 demo here. Um, I'll show one that I've already done so I can just kind of like
18:44 scroll quickly through what happened and then we'll do a live demonstration as
18:50 well. So, let me go here to the Dynamus AI agent. So, this AI agent is what I've
18:56 chosen as an example here. Obviously, we need some repository in this live stream
19:02 to test things out. And so, this is the AI agent that I built as a part of the
19:07 AI agent mastery course in Dynamis. So, if you didn't know, not only do we have
19:11 the agentic coding course in Dynamis, but we also have one where I start you
19:15 from like beginning all the way to fully productionizing AI agents and you get
19:20 access to both courses when you join Dynamis today during the live stream.
19:24 So, keep that in mind as well. So, this agent I'm not sharing right now like I
19:29 am the remote agentic coding system, but this is just going to be our demo
19:33 demonstration for today. And so what I have here, it's really like a chat GPT
19:38 style application where I can start a new chat. I can talk to my agent like I
19:42 can send in a message like hello. And this agent has tools for things like rag
19:48 and web search. It has long-term memory. There's a lot of capabilities built into
19:52 this under the hood. And for our demonstration today, I'm going to make
19:55 some changes to the front end of the application. But here's where the human in the loop
20:01 comes in. Not only do I have my production application at chat.dynamus.ai
20:08 and you guys can go to this site right now, but I have login disabled for this
20:12 because it's just for me to demo to you right now. But anyway, we have the
20:15 production version of the site and then we also have the staging version. So,
20:20 this is a really common practice in software development where you will have
20:24 a sort of test environment that is supposed to be the exact same as your
20:28 production one, but it's going to have some extra changes that you are
20:33 currently testing that are about ready to go into your production environment.
20:37 That's why it's called staging because you're staging these new changes as that
20:42 next feature or that next release, whatever you have in your software
20:46 development life cycle. So I have two versions for the website here and I have
20:52 this just deployed in render. So render is my go-to platform when I want to
20:56 really really quickly deploy things to the cloud. So I have my production
21:02 environment that has my AI agent, it has my rag pipeline and it has my front end.
21:07 And then to keep things simple here for my staging environment, I just have the
21:12 staging version of the front end. And it's just going to point to the
21:16 production agent and rag pipeline because I just want to keep this really
21:20 simple for our demonstration today. We're just going to be working on the
21:24 front end for this application. And then definitely in the agentic coding course
21:27 when I show you this system, we get more detailed and we do like a more complete
21:31 implementation that's like a lot more involved. But I just want to keep
21:35 things, you know, rather simple for our stream today because it's still going to
21:39 show you the power of our custom system here. So, yeah, I've got both of these things
21:46 set up in render. I mean, you could set this up in really any um environment
21:50 that you want like Digital Ocean or Hetner or Fly.io or Netlifi. Like, you
21:55 can just pick any platform. This is just what I have for our demo here. And so
21:59 where human in the loop comes in is when I have my agent through the remote
22:04 coding system make a change to the website, it's going to publish things to
22:09 this staging version first. And so then I can validate things even going on my
22:13 phone and I can check to see like okay I asked it to change the the title of the
22:20 lefthand chat bar here like did it actually do that change properly? Does
22:23 it look good on mobile? like I can validate those things because I can just
22:27 go to the staging environment and then if it looks good then I can go back to
22:31 my system and say like okay this is great now create another pull request in
22:35 GitHub to merge this into main and deploy it and so I have the opportunity
22:39 to validate things and I could also go into GitHub and like look at the diffs
22:43 and the poll requests and things if I want to validate. So, I'm never trying
22:47 to tell you to take yourself fully out of the loop, but if you trust your
22:51 system enough, you can decide to maybe do more of like an artifact check
22:56 instead of a git diff check. And that's super important is like the future of of
23:01 AI coding, I really believe, is going to be where your agent will give artifacts
23:06 for you to review like here's the staging site or here's a screen
23:11 recording of of me interacting with the API, whatever that might be. instead of
23:15 you having to do a line by line code review. And it's still possible if you
23:20 want to, but I think when we have a level of trust in our system, like after
23:24 we've built all of our rules and our commands and everything and like we know
23:27 what it can accomplish, then we can start to like strip away some of the
23:31 complexity of our own code reviews and just look at the artifacts.
23:36 And so let me show you an example of this. And so I have an issue right here
23:41 where and this is like really really simple. We'll do something more
23:44 complicated live together today which I'm excited for as well. But right here
23:49 I just wanted to update the title. So this title used to be just AI chat like
23:56 it was super generic and kind of boring and so I wanted to update this but using
24:01 my remote agentic coding system. So I do nothing locally and so going here I just
24:08 tag my remote agent. So at remote agent command invoke and then I am using my
24:13 prime github command. So it's actually really cool. If I go back into telegram
24:20 here and I do slash commands, we can see all of the registered slashcomands that
24:25 I have for my system. And so we put these in our repository. And then if I
24:31 go to /help again, we have this command to load everything based on a folder. So
24:35 I can point it to my.cloud/comands cloud/commands folder. [cough]
24:39 [clears throat] Excuse me, I got to take a sip break again. So, I can point it to mycloud/comands
24:47 folder or if we're using codecs, we might have like a codeex/comands folder
24:52 like this can be for any AI coding assistant, any path. We can load in the
24:56 commands so that they're stored in the database and ready to be leveraged from
25:01 anywhere. And so that's how I'm able to basically create my custom system is
25:06 like I have a process for priming or planning or validating. And so priming
25:12 is generally how I start really any workflow for building anything with the
25:16 coding assistant. And all priming does is it lays out a set of instructions for
25:21 the coding assistant to look through the codebase at all the core files like the
25:25 readme and the main entry points and things like that and just like establish
25:30 itself in the codebase. And so I like to say that the coding assistant here is
25:34 like catching itself up to speed for what we currently have implemented and
25:38 what might be coming next based on something like a PRD. And so the coding
25:42 assistant goes through and then it outputs something at the end here which
25:46 is just like the summary of the priming that it did. And so now obviously based
25:51 off of all this output, it clearly understands my codebase. So I'm ready to
25:55 go into planning the next feature. So this just sets the stage for our piv
26:00 loop here where we're going to create a plan. We're going to implement it and
26:05 then we are going to validate. And this is something that I cover very
26:09 extensively in the Agentic coding course. So, so the demo that you're
26:12 about to see, I just want to like call out how I explain this in the course.
26:16 The demo that you're about to see is this part right here where we do our
26:22 structured planning. And so, we create a document at the end that is going to
26:26 outline things like our goal, the documentation that we want to reference
26:30 or existing parts of the codebase we need to edit. It has our task list. So
26:35 how it can knock things out task by task our validation strategy so the coding
26:40 assistant can check its own work. We take this structured plan and we send
26:44 that into the coding assistant to implement things. And so we have our
26:47 execute command for that. So at every step of the way we have a command that
26:51 we're going to invoke in our remote system to kick off this part of the
26:56 process. And then we also have validation at the very end. So it can
27:00 run linting and unit tests, maybe even spinning up the website and visiting it
27:05 to verify things before it pushes into staging. Like this is very very
27:09 comprehensive under the hood. And this is all using the commands and global
27:14 rules that we set up when we first begin our project. And in the Dynamis Agenta
27:19 coding course, I show you how to do all of this. And I get into all the details
27:23 to make it really easy to understand like, okay, this part of the process, I
27:27 need to have these things already set up. And then let me go into the
27:30 structure plan. And I get into like how you set up the structure for a plan and
27:33 how we send that off to the coding assistant. And then what our different
27:38 commands look like. So there is a lot of structure for this behind the scenes.
27:42 And so again, if you're interested in like really really getting deep into
27:47 mastering systems for agent coding, definitely like I I'll put a link to
27:51 this in the chat again here. Definitely check out the the dis the special
27:56 discount that we have for today for Black Friday. So I'll put that in again.
28:01 Um because yeah, this this discount here is going away after the live stream. And
28:05 everything that I'm talking about, even everything that I demo today, I get into
28:09 like even more detail in the Aenta coding course. And so just wanted to preface that
28:14 there. Yeah, let's get back into this here. So after I do my priming,
28:20 this is when I use the command invoke part of my remote agent decoding system
28:25 again. And the specific command that I'm using, [cough] excuse [clears throat]
28:31 me, is called plan feature GitHub. And so, by the way, all of these different
28:34 commands that I have, like I was saying earlier, these are really just commands
28:41 like markdown documents that I have in my repository. And so, for example, the
28:46 prime GitHub that I just showed being run through the GitHub issue, it's just
28:52 this right here. And like I said, it's really just a step-by-step process of
28:56 like here is what you do to quickly prime yourself on the codebase and this
29:01 is what I want you to output at the end. And so we ran this. The coding assistant
29:06 read this markdown document or rather my system just automatically loaded it in
29:11 as context and instructions for cloud code and then it outputed the final
29:15 report. And now we're going into planning our feature. So I already ran
29:20 like the slashload commands and everything. So it has access to all of
29:24 these. And so when we plan a new task with our GitHub workflow here, we just
29:30 pass in the arguments here which is like the feature that we want to build or in
29:34 our case it's like just go and handle this issue like I want you to make a
29:44 And so the step by step here is just outlining the planning process for our
29:49 coding assistant. So we understand the feature, then we analyze the codebase to
29:54 see what things we would have to adjust or like what parts of the codebase we'd
29:58 have to edit. We reference external documentation and we research whatever
30:03 we need to and then we create this structured plan. And when you create
30:08 your own system to inject into the remote agent coding system, you can do
30:12 whatever you want as far as your structured plan or what your commands
30:16 look like. That's the beauty of this is I loaded into my own custom commands and
30:20 I could change these at any point. You could have a completely different set of
30:24 commands that you invoke whenever you want to build something. You could use
30:28 no commands at all if you want to do that for whatever reason. Like it's
30:32 totally up to you how you leverage this system in GitHub, Telegram, wherever it
30:37 is. And I'll talk later about how we can incorporate other applications like
30:41 notion and slack as well. And so I outline like the user story and the
30:45 problem statement. Here are the different pieces of context that I want
30:48 to reference like where I'm going to have to edit in the codebase. The
30:52 implementation plan which has my task by task like really really getting in the
30:56 weeds here of everything that we have to account for to implement this feature
31:01 including the validation process as well. Now obviously for what we are
31:07 building here we don't need this comprehensive of a plan just to update
31:13 the front-end title. But I obviously I keep saying obviously, but I built this
31:17 to work with much more complex feature requests. And so yes, it's overkill for
31:22 what I want to do right now, but I'm only doing something this simple for the
31:26 sake of a brief demonstration. Typically, what I send off for a pivot
31:29 loop is going to be something much more complex like adding a new tool to my AI
31:34 agent, for example. And that's one of the things that I do actually showcase
31:38 in the agentic coding course is I use this system to build like a really
31:43 robust web search tool into the primary agent that I build throughout the
31:48 course. And so we do the planning here and it comes back and it gives me the
31:53 final GitHub summary. And so it created a branch and then it also created the
31:58 structured plan within this branch. So it reports to me absolutely everything
32:02 that I did. And it's all thanks to the prompting in the plan command that it
32:07 knows to like give me this information. And so now I can go and validate the
32:10 structure plan if I want. Like I can put myself in the loop as much as I want.
32:14 Obviously in this case it's very confident that I can do a one pass
32:19 implementation success here. And so I can just go on to invoking the next
32:24 command. And so I'm going to do the same thing here. Slashcomand invoke. But now
32:28 this time I'm using the execute. So it's going to take in two arguments here. We
32:33 need the path to our structured plan for the implementation.
32:38 And then we need the feature branch that we are operating on. So it's a very git
32:42 native workflow. Every time we're working on something new, we are doing
32:47 it in a specific feature branch. And so it goes through this implementation in
32:51 this branch based on the plan here. And then it gives me the report at the end.
32:55 [snorts] And not only does it give me the report at the end, it also makes the pull
33:02 request for me. And so we are merging this feature branch with the
33:07 implementation into our staging branch. And so once I accept this merge or this
33:13 pull request, it is automatically through render going to deploy things to
33:22 So without me having to manually deploy anything, I can automatically review
33:27 basically the artifact of what the coding assistant created for me. And so
33:32 I can make sure that the title here looks better. Obviously this is already
33:36 I've went through the whole process here. So it's already in the production
33:39 version of the site as well. But if I was just doing this for the first time,
33:44 this would still say AI chat in my production URL, but then staging would
33:48 now have the new one. So I can validate things here before I go back and tell it
33:53 to merge things into main. And once I'm ready to merge things into main, I could
33:57 like go to Telegram, for example. Let me just reset the conversation. And I could
34:02 literally just say like uh you know, I just made a change to staging. That
34:09 looks good. Please open a PR to merge this into the main branch. And I'll do
34:13 this exact thing in a live demo in a little bit as well. But like that's how
34:17 simple it is. Like once I check off the box of like okay I visited the artifact
34:21 myself like even on my phone then I can merge things into main. And so if we
34:26 want we can review the structure plan. We can review all of the changes that
34:29 were made in the codebase. Like I can click into the pull request here and I
34:34 can look at the files that were changed. Um like in this case the chat sidebar is
34:37 the one that was really changed. These other ones are just because I had my
34:41 staging branch is a little outdated but this is what was really changed. And so
34:44 I can look at this line by line if I want or if I trust things enough I just
34:50 visit the artifact. So again flexibility in the system is what we have here. That
34:55 is the beauty of it. And so that's pretty much the end of like this simple demonstration going
35:01 through something that I already did. But now to like really drive home the
35:06 power of this for you, I do also want to do a live demonstration
35:11 where I don't just like scroll through comments that I already made, but I
35:16 actually go through this with you and we'll build something new and we'll see
35:20 it like show up in just staging. We'll validate it here and then we'll merge it
35:24 and like have it in production as well. So that's going to be really fun. But
35:27 before we do that, I also want to show you how to get this all running for
35:31 yourself. So, this is kind of like more of the introduction here. Now, I want to
35:35 get into the setup in case you really do want to follow along and like leverage
35:39 this because this is something that I am giving away for the live stream.
35:45 So, I'll do that in a second, but I also want to take a sec to um address any
35:52 questions as well. And so, yeah, I'm going to um go back to my full frame here.
36:03 All right. And then let me move this up. Not that one. Let's do this one. Okay,
36:10 there we go. Cool. So, back to full frame. I'm going to take a little bit of
36:13 time just to chat with you guys and answer some questions. Then we'll get
36:16 into the setup, then the live demonstration. Then I'll also talk about
36:20 how we can extend this to work with other apps and coding assistants as
36:23 well. So, like I said, I got a lot of really exciting stuff in store for
36:26 today. So, all right, let me start with the Uh, sorry, I got to like organize things
36:38 on my other monitor for a sec. So, let me go back here. Okay,
36:44 so Austin asked, uh, will we already have access to the course if we're
36:47 already in the Dynamus community? Yes, Austin, 100%. So, the Dynamus Agentic
36:52 Coding course is a part of the community. It is nothing separate. So,
36:56 if you're already in the community, you have access to this right now. But yeah,
37:01 very good question. And uh yeah, thank you Sydney for saying that as well.
37:06 That framework engine is insane. Yeah, it is. I appreciate it.
37:13 Cool. Um let's see. There was a another comment that I wanted to address
37:18 here. [snorts] Well, first Cole's working hard day and night. Looks very tired. Well, yeah, I'm
37:25 just coming off a cold. I'm not actually tired. Like, I'm doing well and I'm I'm
37:29 hyped for this, but yeah, I probably just don't quite look 100%. Um, and and
37:33 probably like the voice kind of is like a placebo where maybe it makes me look
37:38 like I'm tired, too, but I'm not. But yeah, um, Peter said it also has an element of
37:44 memory as there is a Postgress database with three tables to track history.
37:48 Yeah, that's another part of the system that I will get into. Um, let me
37:54 actually open up my scene again. So, I'll get into the architecture for
37:58 this system. And I know that like architecture isn't always as
38:01 interesting, but like I think you guys will actually really appreciate how I
38:05 built this system. And the database tables is one of the things that I'll
38:08 talk about, which is actually very cool because what I've done is I've set it up
38:13 so that you can restart this application like tear it down and spin it back up
38:16 and it still knows all the commands that you have loaded for your different
38:20 repos. You can resume conversations because everything is persisted to a
38:25 Postgress database and you can run the database locally or you can do something
38:29 remote like with Superbase or Neon. And so yeah, the data persistence is a big
38:34 part of what I've architected as well. So yeah, just finally got my membership.
38:39 Happy customer. I appreciate that very much. Thank you and welcome to the
38:46 community. All right. Uh let's see. Is this something I could connect to
38:53 Archon? That is a really good question. So, yes, this system can work with any
38:59 MCP server that you want. Now, that is outside of the scope of this live
39:03 stream, unfortunately, but that is one of the ways that I'm going to be
39:05 evolving this. So, if you join the Dynamis community, um, this remote agent
39:10 coding system is something I'm going to be continuing to work on and
39:17 incorporating MCP servers more and like actually setting up a way for us to like
39:21 in Telegram or in Slack connect our MCPs. That is something that I'm going
39:25 to be building into this system. So, definitely stay tuned for that. And it's
39:28 something that like you could incorporate right now if you want. So, I
39:31 know that I'm saying that it's outside the scope of this live stream, but just
39:35 really quick, like if you were to take this remote agent coding system and you
39:40 were to go into your coding assistant, give it the configuration for an MCP
39:45 server like Playright or Archon or Sneak or whatever, and just tell it like, I
39:51 want you to add this into my Claude Code agents SDK or add it into Codeex SDK.
39:55 It's going to be able to do that and probably knock it out in just a single
39:58 go. So yeah, that is definitely something you could do if you want to
40:02 extend the system for yourself. And that's another thing with the the system
40:07 that we have here is um it's very well architected not only so that like I can
40:11 extend it, but that's so you can as well. And so coding assistants are able
40:16 to work with this codebase really well because I've optimized the entire
40:21 codebase for agents to understand, navigate, and implement new things. And
40:25 that is yet another thing that I cover a lot in the agentic coding course is not
40:30 only how we can build our system for AI coding assistance, but how we can create
40:35 code bases that are optimized for coding agents to work in them. And when you
40:39 combine both of these things, like you have a really robust system for
40:42 planning, implementing, and validating, and you have an AI optimized codebase,
40:47 the the sky the like the ceiling is just like so high for the things that you can
40:53 build um even in like one shot. Not saying that you can build an entire
40:56 application in one shot and like you'll build that muscle to kind of like
41:00 understand what your system can accomplish, but yeah, it's pretty crazy
41:03 the kinds of things that I'm able to build with um really just having to like
41:08 leverage my system in the same way that I always do. Like I go through the
41:10 planning, I go through the implementing and I feel like I'm doing the same thing
41:13 every time, but I'm building like these really cool stuff like just really
41:22 Let's see. Um, I cannot make my fork private. Change repository visibility for
41:28 security reasons. You cannot change the visibility of a fork. Yeah, so when you
41:32 fork an open source repo, it does have to be public, unfortunately. Uh, but the easy solution
41:38 to that is you can just make a private repository and then copy over everything
41:44 from the remote agent coding system. And so yeah, you can basically do a fork.
41:48 It's just an extra step of you have to copy things manually. But yeah, then
41:51 you'll be you'll be good. and it'll be U Joseph said the same thing. Yeah, that
42:00 same solution there. Planning is the key is what I'm hearing.
42:05 Uh yes, 100% planning is the key. And let me actually go back to my scene here
42:10 because I want to talk about that really quickly. So when you see like what you
42:15 see um me doing in the remote agentic coding system, that is how I always use
42:20 AI coding assistance. Now I go through this piv loop process where I do the
42:25 priming, I do the planning and then I go into implementation. But the planning
42:31 like this takes up half of my diagram here and it takes up half for a reason.
42:35 And what I just showed you in the demonstration, usually I do a lot more
42:41 planning because I I start with what I like to call vibe planning. And so at
42:45 this point it is an unstructured conversation where our goal is to simply explore
42:51 ideas, architecture, concepts and tech stack. We want to get on the same page
42:56 with our AI coding assistant what we are building for the next feature for our
43:01 next PIV loop. And it's actually like a nice relief. Like when we first start
43:05 working with a coding assistant on something, we don't have to have this
43:09 insane structure in place. Like we just start with a very casual conversation.
43:13 And then once we're confident that we're on the same page with what we're
43:17 building like we're making this change to the front end or we're doing
43:21 whatever, then we go on to creating our structured plan. So you evolve into the
43:26 structure where you have that document I was talking about earlier. But you don't
43:29 have to start there. It can be very free form at first, but like this process of
43:35 planning is the most important part out of any [snorts] agentic coding system
43:39 because if your plan is bad, it does not matter if you're using the latest and
43:44 greatest opus 4.5. It does not matter if your execute command is really
43:48 wellcraftrafted. If your structure plan is bad and the agent doesn't understand
43:53 what you actually want to build, it is going to completely fall on its face
43:57 here. Doesn't matter what it has for validation either. You need the plan to
44:02 be crystal clear. Not just that it understands what to build, but that it
44:06 also understands or I should say not just so that it understands how to do
44:09 things well and like how to research and how to document, but also that it
44:14 understands your vision for the project and that feature. And so this is your
44:19 place to guide it on exactly what your design decisions are and your vision for
44:23 the project. So, I appreciate you saying that because Um, another question, have you ever
44:36 tried specit and other frameworks that level up the project project context for
44:42 AI coding agents? Very good question. The answer is 100% yes. even to the
44:48 point where I have actually a couple of modules in the course that are dedicated
44:54 to trying out some of these other context engineering strategies. And so
45:00 part of the reason why my voice is right now is because I did three one and a
45:05 halfhour workshops in the Dynamis community earlier this week, Monday,
45:09 Tuesday, and Wednesday. And Wednesday especially is when my voice was like
45:13 really going. But we pushed through it. It was a great workshop. But what I did
45:18 to end off the course is [snorts] I covered the PRP framework and then the
45:22 two that you literally just mentioned, GitHub spec kit and the BMAD method.
45:28 Now, the reason that these are at the end of the course is because the rest of
45:35 the course is all about showing you how to create your own system, right? Like
45:41 the BMAD method and the GitHub spec kit. You can kind of think of these as like
45:47 an opinionated outofthe-box system for context engineering. So they're really
45:51 really powerful and all three of these are very welldesigned. Like I have a I
45:56 have a deep level of respect for all three of these. There are also a lot of
45:59 other good strategies for context engineering um like clawed flow for
46:04 example. So they're all fantastic. But the thing is, when you have the system
46:09 in your own hands and you get to evolve it to work the best for your code bases,
46:13 you are always going to be able to outperform any outofthe-box solution.
46:18 So, while these are fantastic, the way that I position them in my the live
46:23 streams that I did here and at the end of our course is it's like, okay, now
46:27 that you've gone through the course and you understand how to build your own
46:32 system, now I want you to try these, not to replace the system you just built for
46:36 yourself. Like, you already have your own rules. You already have your own
46:39 commands and contexts and everything created, but I want you to go through
46:44 using these so you can learn from them because they are pretty geniusly, if
46:49 that's a word, geniusly implemented. Like the the BMAD method has so many
46:55 really cool um pieces of architecture built into it. And there's just so many
46:59 considerations that you can look at how they do things like how they incorporate
47:04 sub agents for you know like Mary the analyst and Winston the architect and
47:08 things like that John your project manager and you can like grab parts of
47:12 that and you can build it into your own system. So using these as a learning
47:17 opportunity to evolve your own system versus just using them out of the box
47:20 because it's not going to perform as well as what you can do if you're really
47:26 evolving your own system. And so, yeah, I hope that sounds good to
47:29 you. Like, these are great, but you want to build your own thing. And so, yeah,
47:33 like if you really are interested in building your own custom system for
47:38 coding that is like actually optimized for your code, that's when you'd want to
47:42 take advantage of the offer that we got for just the live stream here, the 45%
47:49 off annual and 25% off um the monthly because like this this course is done as
47:53 of Wednesday. And I'm repeating myself a little bit because people come in and
47:56 out of the stream, but like as of Wednesday this week, this full agentic
48:00 coding course is now complete. And I have been working my butt off u on this
48:05 over the last couple of months here. And so yeah, it's it's like like let me just
48:10 tell you guys, it is surreal that this course is done. I started planning this
48:16 all the way in June like thinking through how I'd want to start the course
48:20 with like module two on the piv loop and then getting like okay when do I cover
48:23 MCP servers when do I cover like creating a system that can evolve itself
48:27 because I go over that in module 7 as well and then thinking about like well
48:31 how do I want to position the different context engineering strategies like I
48:34 was just talking about like there's just so much that went into planning the
48:38 structure of this and everything and so yeah it's just like surreal that this is
48:43 available know, but yeah, I appreciate you asking about um these other context engineering
48:50 strategies and yes, you are very very welcome. Jack asked, "How can we access the
48:57 course?" Uh yeah, really good question. Well, here, let me put the link in the
49:01 chat again for the uh special opportunity for the live stream here
49:05 with the the really really big discount. And by the way, like this is in
49:09 celebration of the course being released. And so way back in like July
49:13 or August, like I already knew like I want the course done in time for this
49:19 live stream and just like the whole, you know, Black Friday, Cyber Monday thing.
49:24 Cool. All right, let's go back to the chat here. Um, one moment. There we go.
49:30 Just signed up for the monthly price. Thanks. Awesome. You're welcome. And and
49:34 welcome to the community. Appreciate it a lot. Cool. Uh, Opus 4.5 for planning and
49:41 backend. Gemini 3 for the front end. Yeah, that's something that I've been
49:45 playing around with a lot as well is trying um yeah, Gemini 3 seems to be
49:49 really good at um visual stuff a lot more than the cloud models. Um
49:54 especially because in anti-gravity we have the Google Chrome integration. So
49:58 it's kind of like the Playright MCP server where um in anti-gravity Gemini 3
50:03 can spin up a browser and autonomously navigate your website to validate things
50:08 visually. It's very very powerful. But yeah, like just in general for most
50:12 things, um, Claude Opus 4.5 does seem to be the strongest um, coding model by a
50:18 decent margin. Like it's pretty noticeable and so I'm definitely using
50:26 Cole sounded like an old smoker at the end of the workshops. Yeah, well, I
50:29 probably will the end of this one today, too, Sean. No, that that's funny. Yeah,
50:34 I kind of [clears throat] Yeah, but I'm pushing through it. It's It's good. No,
50:39 luckily I feel 100%. Like I was I had like a massive headache and even a
50:43 stomach ache a little bit on Wednesday. That was not fun. Um but like even on
50:49 Thursday I was like feeling good again. Um quick shout out to the Dynamus AI
50:53 community. Joined only a week ago and I'm already blown away. I've been able
50:55 to attend three workshops already and learn a ton. Worth every penny. Thank
50:59 you very very much. I appreciate that a lot. And yeah, three workshops in the
51:03 first week is that's a lot of time. So thank you for being a part of all that.
51:07 And by the way, all workshops are recorded in the community, too. So, I
51:10 know that like every week we got a ton of stuff going on. So, you can always
51:13 like come back to stuff. And then when you join the community, you also have
51:17 access to everything that has already been done, like all courses, all
51:21 workshops, and other events. And so, it's never like you join and then like
51:24 you only get the stuff going forward Um, I realize I can make anything fast
51:33 and suddenly nothing has value beyond the experience of making it. But that's
51:36 the part I'm removing. Love building my aentic platform but hate actually making
51:41 with it. That's interesting. I guess maybe you could like elaborate a bit
51:44 more of what you mean. I assume you mean that like it's so easy to build anything
51:49 now that it like everything's kind of commoditized. I don't know if that's
51:51 what you mean by what you're saying at first, but that definitely is like an
51:55 interesting thing to consider is like the the moat like you know like the
51:59 protection you used to have around what you build. A lot of times it could be
52:04 just because like that thing was so complicated to build, but now like
52:10 thanks to coding assistance and how fast you can build things, especially with a
52:14 proper system, um that moat goes away. Like when you build something that's
52:18 like very architecturally complex, someone else can still probably build it
52:22 in a month or less. And so now to like really protect your business, it's more
52:25 about like the audience that you have, the reach that you have for what you're
52:29 building, and the vision behind it. like you want people to get behind your
52:32 vision more than you want them to get behind like you know the capabilities of
52:36 your application because that kind of thing is very replicable. So I don't
52:39 know if that was exactly what you're getting at but it is something like
52:42 really interesting to think about um just because like software is a
52:46 commodity now it's at least kind of how a lot of people uh phrase it these days.
52:52 Yeah. I hope uh you'll give us an anti-gravity video. I am planning that. Yes. cuz
52:57 there there's a couple things in anti-gravity that are really fascinating
53:02 to me. The Google Chrome integration that I already mentioned is one of those
53:06 things and it's not something that's like brand new. You have MCP servers to
53:11 do that kind of like visual validation of websites already and even like taking
53:16 screenshots as artifacts for you to review. Um but it it does seem like the
53:20 Google Chrome integration works a lot better than um using an MCP. like it's
53:25 integrated directly in the IDE and the system prompt for Gemini in anti-gravity
53:30 is like optimized to work with it. So, it's really powerful. And then the other
53:35 thing in anti-gravity is the agent um work orders or like the agent manager um
53:39 that you have as like another option to the traditional IDE. So, you can manage
53:44 your coding assistance in parallel um kind of like the remote agent coding
53:48 system, but you're doing it directly in the IDE. It's also very cool. So, yeah,
53:53 definitely do want to cover that. in a video. [snorts] David asked, "Are are you okay? Sound
54:00 different today?" Yeah, so my voice is I'm still recovering from something
54:04 earlier this week. Um, but yeah, luckily though, okay, how long? So, the stream
54:09 has been going on for 55 minutes and my voice is like definitely not ideal, but
54:13 it also is not getting worse. So, I just want to call it out like I'm feeling
54:16 good about that, even though I do wish, of course, that um my voice was
54:23 Cool. Funny to see that you're building similar things to me. I built for myself
54:28 CI/CD for adding MCP tools. Highly recommend real DB. Very, very cool.
54:32 That's neat. Best training online. I appreciate Karen. Thank you very much.
54:40 Uh Jim enjoying the snowy day today. Oh yeah, cuz Jim, you're in MN as well.
54:45 Yeah, it's beautiful in Minnesota today. Now, if it looks this way in March, I
54:49 will tell you it's ugly. But when it's uh December and we're or I guess end of
54:53 November and we're just getting snow like for the second time, it it's it's
55:01 Um digital novelty has evaporated. I mean, yeah, that's kind of another way
55:04 to say that um software is a commodity now, but I I don't know if like I'd say
55:10 novelty is totally gone. It's an interesting thing to think about. Like
55:13 we can build anything very quickly, but it still has to start with the human to
55:18 have the idea, at least generally. And so I'm sure there's still going to be
55:21 like a lot of different applications that are released over the next decade
55:26 and beyond that um might be kind of like obvious like someone's like, "Oh, that's
55:29 so easy. I could have built that in a week and I could have made a million
55:33 dollars." Um but it's just the person that first has the idea and then also
55:38 has the reach to make it known to people. Um, so I don't I don't know if
55:41 I'd say that it's totally evaporated, but it's definitely like harder now to
55:46 come up with a totally novel idea that hasn't been built now because the
55:49 people, [cough] [clears throat] excuse me, people have just like built everything now. It
55:54 definitely feels like that. So I get what you're saying. [snorts]
56:00 Um, I'm in curious still the remote uh the stuff is that similar to jewels and
56:06 claude GitHub integration. Yep. So the answer is yes. And so, um, yeah, I think
56:12 I'll explain this and then let me I'll go back to, um, kind of like
56:16 getting you guys set up with the remote agent coding system as well. But yeah,
56:21 so let me go back to my screen here. So I explained this a little bit earlier
56:25 but um what we're building here with the remote agent coding system and what I'm
56:29 giving you guys today only in the live stream this is something that is yeah I
56:35 wouldn't say brand new like you said jewels and claude GitHub we have codeex
56:41 um cloud code for the web we got factory as well but what really is new here is
56:48 the level of control that we have when this is our own system which [snorts] by
56:52 the way Since you guys have the code and I have optimized this codebase to make
56:56 it really easy to build new things, you can also feel free to extend this for
57:01 yourself if you want to integrate it within um you know like notion or linear
57:06 or slack or you want to add in open code as your AI coding assistant and you're
57:10 able to do things like configure MCP servers and build in your own commands
57:15 that you really can't do at least as easily in something that's more out of
57:19 the box. And then the other main benefit is since we can integrate this anywhere,
57:23 we are able to bring our coding agents into the applications where we actually
57:28 do our work. And you can't do that with something like claw code for the web. If
57:32 I'm on my phone, I have to go to this website. I can't just open up the
57:37 Telegram app. Um, and then like there's just a lot less room for customization
57:42 even with the user interface. Like maybe I also kind of want to like customize in
57:45 Slack how I'm managing different conversations in different threads with
57:50 my agents. Like I have the option to customize the front end as well through
57:55 the specific app that I integrate and Excuse me. I'm going to take a sip break
58:07 All right. Oh, cool. Maronei joined. Finally able to join. You're live due to
58:11 time zone. Just want to support here. done all the modules like Sean. Very,
58:15 very cool. [cough and clears throat] Excuse me. I might have a coughing fit
58:19 once in a while, but hopefully I'll be good. But yeah, thank you, Marone. I
58:22 appreciate it a lot. Yeah, there's there's a couple of of you guys that
58:25 have gone through the entire course already, even though I like blitzed it
58:29 over the last like month and a half. So, I appreciate you guys because there's a
58:34 lot there. Like, if you join today, there's going to be a ton of content for
58:38 you to consume for the rest of your life. It feels like there's just so much
58:43 time that I put into giving value here. So, I appreciate that, Maronei.
58:48 All right, cool. So, with that, let's get into the setup a little bit here.
58:53 Excuse [clears throat] me. So, if you guys have been following along [snorts]
58:59 at this point, um you you know like how this works and why this is important,
59:03 like why I truly believe this is the future of agentic coding. But um yeah,
59:08 if you like have the repository cloned or downloaded right now, I just [snorts]
59:11 want to walk through the setup really quickly. Now, I will go kind of quick
59:17 and the biggest reason is it's pretty straightforward. If you go through the
59:20 read me here, I have everything documented very clearly. Um it looks
59:24 like this readme has to get updated though. Hold on, let me edit this. I'm
59:30 sorry guys, I have a typo here because the repository is colam0000.
59:34 So, let me fix that really quick. So, you guys could pull the latest changes
59:38 if you want [snorts] or you could just go and copy um this right here. So, this
59:42 is the URL to clone the GitHub repository. So, you clone the repository
59:47 and then we set up our environment variables. And so, there's a couple of
59:51 things that we need for our application up front. We need our database URL
59:57 because like I was saying when I covered the architecture just a little bit, I'll
60:00 dive into this more later. This is like actually really fascinating. We have a
60:05 few different tables in our database to persist the code bases that we're
60:10 working on in our system, the different conversations that we have going, and
60:14 then also the different sessions. And so we're able to with these sessions, we
60:18 can resume conversations between different threads. Even if we turn on
60:22 the application and spin it back up, we don't lose any information. So we need
60:28 the database. And then we also need our GitHub token because in order for this
60:33 remote agentic coding system to work, it has to be very git native. Like we saw
60:39 in the demonstration earlier, everything that I was doing was based
60:45 around something in GitHub. Even if we're using the application in Telegram
60:50 and we're not using it in like a GitHub issue or a GitHub pull request, it still
60:54 has to use GitHub to for example make the pull request like we see right here.
60:58 So if we want to be able to review changes and manage [snorts] different
61:02 environments like staging and main, we always need the coding assistant to have
61:08 direct access to our repositories through the GitHub CLI. So the GitHub CLI is a core part of this
61:17 application. It's a tool that we always make available to our coding agents that
61:22 run in this system. And I cannot stress enough how critical it is that we have
61:26 access to the GitHub CLI because otherwise like think about what happens
61:30 if we don't have access to GitHub. If we don't have access to GitHub, then the
61:35 agent is going to make a change locally. But because this is a remote system, we
61:40 can't actually see the change that it made. And so, like, if we're talking to
61:44 it on our phone, like Telegram or GitHub or whatever, we literally can't see what
61:49 it did. Um, like it has to use GitHub as a way to share its changes and publish
61:54 things. It's a bit different than when we're working locally. And like this
61:58 really is the key difference. when we're working locally with our coding
62:02 assistant, it's making changes alongside the codebase that we are also looking at
62:06 because it's working side by side with us. Now, with the remote coding system,
62:11 we want to set it up where it feels like the coding assistant is working side by
62:15 side as much as possible, but really it's operating on a different device.
62:19 And so, the local repo is not something we can access besides just asking it to
62:24 summarize things. It's like of course if we don't have access to GitHub, we could
62:28 just tell it like please summarize the changes that you made or I want you to
62:32 like tell me exactly what this front-end component looks like now that you've
62:35 updated it. We can do that, but there's still no way for us to truly dive into
62:40 the code ourselves or a way to publish it. So that's what GitHub is for. So we
62:46 set up these two things and I'll go into the repository here. So I have it cloned
62:51 locally right now. If I go into the env.ample example, we set up our
62:57 database and then we set up our GitHub token. So, you need the value copied
63:03 here just because the GitHub token you use for the GitHub CLI has a different
63:07 environment variable name than the one that's used for the GitHub web hook.
63:12 Excuse [clears throat] me. Sorry about that. I know that that's like a little
63:15 confusing, but you just use the exact same value for both of these. And the
63:21 way that you get access to this token is you go to your GitHub settings and then
63:26 you go to personal access tokens. So I'll click on the top right here. And
63:30 like I said, I'll go through this whole setup pretty quickly, but I just want to
63:33 hit on it quick for those of you who want to like set this up live with me
63:37 right now. And so I'll go to my settings in the top right. And you just scroll
63:41 down to developer settings and then you click on personal access tokens. And
63:45 then typically I'll just make a classic token that has repo access. So this is
63:49 where you go. It'll walk you through. It just is like a couple of steps here to
63:53 create your token. And then you save that uh in these two environment
63:58 variables, the exact same value. And then we get to the point where we
64:02 get to set up our coding assistant and the applications that we want to
64:06 integrate with. And so the rest of these environment variables are going like
64:10 which one you set up. It actually depends on the specific tools that you
64:15 want to use. And so currently the remote agent coding system supports both cloud
64:21 code and codeex. We can use both of these at the same time. We can set up just one. This is
64:27 part of the flexibility that we have here that you're not going to have with
64:30 something else like cloud code for the web. Obviously, you have to use cloud
64:35 code, which is also a bummer, by the way, when Enthropic has an outage. We
64:39 see th that all the time where you'll get outages or even just rate limits.
64:42 And so, another benefit of this system is whenever there's an issue with one
64:46 provider, you can just switch over to the next one. Like even over the past
64:49 couple of weeks, there have been outages with Enthropic. And so, I just go to my
64:53 system and like, all right, let's change it up here. The default AI assistant is
64:57 now going to be CodeEx. And then boom, I'm off to the races. just continuing my
65:01 work, but now it's using codecs instead of cloud code. And so, yeah, as I
65:06 continue to expand this system to add in more coding assistance, it's just adds
65:11 even more flexibility. And like I said, I've optimized the codebase in a way
65:15 where with just like a single piv loop like plan, implement, validate, I can
65:20 add in open code or as far as our application integrations, I can add in
65:25 Slack. Um, so yeah, you just pick your coding assistant and then I have like a
65:29 little like chevron here that you can expand that is going to give you the
65:33 instructions to get your subscription key. And so this is not your anthropic
65:38 API key. You can use that if you want, but is going to be much much more
65:43 cost-effective if you use your subscription token. And a lot of people
65:48 don't know this. For some reason, even some YouTubers that I've seen, they have
65:52 this misconception that when you use Claude code programmatically, like I'm
65:57 doing here with the Claude agents SDK, they think that you have to use your
66:01 anthropic API key, which is a bummer because you end up paying like dozens of
66:06 dollars just for a couple of features that you implement. So, you'd much rather just pay like 20,
66:13 $100 or $200 a month. Those are like the three tiers they have or whatever. And
66:17 then you sure you have your rate limits, but otherwise like for each time you go
66:21 through your um remote coding system, you're not paying anything extra like
66:25 you're just playing paying that flat fee for the subscription. So you can um
66:31 within your local machine once you have the claude code CLI installed, let me
66:36 zoom in a little bit here. Then you can run this claude uh setup-token command
66:41 and that's going to give you a long living token so that a remote agent
66:46 coding system can use our claude code subscription and then it's a very
66:50 similar thing for codeex. So you do your codeex login and then when you do that
66:56 I'll actually like show my folder here. So this is in my local computer. It
67:01 creates this o.json file. Now obviously I'm not going to open this right now
67:04 because it has my secrets. That's also why I'm not running the um cloud setup
67:08 token right now either. Like I don't want to, you know, share my secrets with
67:12 you guys. But when you click into this file, it's going to give you these four
67:15 values. And that's all you have to set up in the env.ample right here. So if
67:21 you want to use codec, you set these four. If you want to use claw code, you
67:25 set just this one. Or you can use both. Like you can set up both and then just
67:28 swap between these two by changing the default assistant. And then I also have
67:33 it set up where it actually autodetects if you clone a repository in the remote
67:38 agentic coding system and it has a cloud folder. It will automatically use claude
67:43 for all conversations in that repository. And then if you have a
67:48 codeex folder you know codeex like this then it'll automatically use codeex when
67:53 you start new conversations in this repository. So it's flexible in that
67:57 way. The one limitation is you can't swap between coding assistants in the
68:02 middle of one conversation. And it's actually very interesting. The reason
68:08 for that is because these different SDKs like the Cloud Asian SDK and the Codeex
68:14 SDK, they store and manage the conversation history on their side. So
68:18 there's no way for us to access the conversation to then share it with the
68:23 different coding assistant. So once you start a conversation like what I showed
68:29 you in Telegram or GitHub, like once you start it, you're locked into that
68:33 specific coding assistant, [snorts] but I don't really think that's a problem.
68:37 Like obviously you're going to pick the one that you prefer and you're generally
68:41 going to stick with that. So that is the setup for our coding
68:48 And then the last thing you have to do is just set up your what I call platform
68:53 adapter. [snorts] This is our application where we are interacting with the coding assistant.
69:00 And so for Telegram, they make it really easy to set up a Telegram bot. All you
69:06 have to do is message the bot father. I'll actually show you this really
69:10 quickly. So we go into here and I go into contacts. I search for atbotfather.
69:17 And then you uh click into bot father here. And by the way, Telegram is
69:20 totally free and it's like really really easy to install. I click on start. And
69:24 then the first command at the top here is newbot. So I just do slash newbot.
69:30 And then it's going to ask me a couple of questions like what I want to name
69:33 it. I won't go through this whole process right now. Um but once you go
69:38 through that, it'll give you an API key. And so you just like put that in as the
69:43 Telegrambot API key in your. And that's it. Like literally the
69:46 telegram setup will take you like 5 minutes and then you can instantly start
69:50 using the remote agentic coding system in telegram. And so telegram isn't
69:55 something that I use outside of for this application. Like I installed it on my
70:00 phone literally just to talk to Claude remotely but um I I did that because
70:05 it's just so easy to install and incorporate. [cough] And then when I add in like Slack later
70:13 for example, it'll be um a little bit more complex like still very doable.
70:16 There's going to be a little bit more setup because you have to go and create
70:20 an app in Slack and set up ooth and stuff. It's a little bit more
70:23 complicated. And so I just love how simple Telegram is. And so yeah, even if you know nothing,
70:31 like I see um Parn in the chat said no, he knows nothing about setting up
70:35 Telegram. like don't worry it is like it is just so easy once you in install
70:40 Telegram which um I don't know the exact URL for telegram yeah telegram.org you
70:44 just go to telegram.org org and you install it in like five minutes. It's
70:48 really, really easy. And then for GitHub, it's a little bit more in-depth.
70:52 I'm going to not get into the weeds of this right now because I did outline
70:57 things very clearly in the readme, but you do have to set up what is called a
71:02 GitHub web hook. And so when I'm in GitHub here and I make a comment like,
71:07 okay, remote agent, I want you to use the prime command. The way that this is
71:14 this command or this request is received by my coding agent, my remote system, it
71:21 is leveraging a GitHub web hook. And so I have this repository set up where
71:25 every time it receives a comment on an issue or a pull request, it is going to
71:31 send a request to this API that I have built into the remote agentic coding
71:36 system. And so then obviously we'll just filter out any comments or pull requests
71:41 that don't start with at remote- agent. Right? So like that's kind of the
71:43 gateway there is like if it starts with this prefix then I'm going to send this
71:50 off to the agent in my system. And it's the same kind of thing with any other
71:53 application that I'd integrate with like Slack for example. It would be we would
71:57 just do like at remote- agent in Slack and that triggers the web hook there as
72:01 well. And so web hooks is the way for us to basically send events from our
72:07 repositories into our system. And so I go through the setup here. We're going
72:11 to use the GitHub token that we already set up in the core configuration.
72:16 And then we have to create a secret. And so this is the command that you would
72:20 run on like Linux or Windows. Like I'll just show you on Windows really quick.
72:25 If I run this command here, I'll just go ahead and paste this in. It just gives
72:29 you this value. So you copy this and then you just go in to your environment
72:33 variables and just set that as your secret right here. Like it's that easy
72:36 to create your secret. And then you'll use this exact same secret when you
72:41 create the web hook as well. So you go into your repository settings, you add a
72:46 web hook, and then I walk you through the different values that you need to
72:51 set here. And then you um just need to yeah, like I said, set the web hook
72:56 secret. And then you can also change the streaming mode for GitHub. So one thing
73:00 that you might have noticed when I go into uh Telegram and I say like
73:05 summarize the read me because this is more of a real-time chat application I
73:11 actually get like every single time it does something like replies with some
73:15 text or uses a tool it is communicating with me in real time what it's doing.
73:20 But it was a little bit different in SL in in GitHub because in GitHub it only
73:25 sent back the final message that the coding assistant gave like the summary
73:30 of what it did for example or like in this case the summary of the prime and I
73:35 did that specifically because like GitHub is not a real-time chat
73:40 application. I don't want it to spam my issue here with a ton of comments for
73:44 like okay now I'm going to look at this file and look at this file and then look
73:48 at this file and that's what it does in telegram because it's more of a chat
73:51 application but here I just want it to give me the final thing at the end and
73:56 so we we actually like configuring that here and you can swap between them so if
74:00 you really do want your coding assistant to spam your issues here with all the
74:04 little updates that it's doing and the actions that it's taking you can
74:08 definitely do that if you want so So again, just a another example of the
74:13 customization that I allow for here. And then yeah, I just have some
74:17 documentation for how you can mention like kick off the agent in your issues
74:22 or pull requests. And so for any repository that you have the web hook
74:27 set up, like if I go to my settings in the Dynamis agent and I go to web hooks,
74:32 you can see that I have this web hook already set up. I can click on edit and
74:36 you can see like I have my payload URL and I have the different events that I
74:40 want to trigger like issue comments and issues and then pull requests as well
74:45 which this also includes pull request comments and so you have granular
74:49 control over like when we take any event in the GitHub repo and send it off to
74:54 the API that we have built into the remote agent coding system.
74:59 So yeah, that's pretty much the setup here. Let me scroll back down to the
75:04 bottom of the readme here because once you have everything configured in your
75:07 environment variables all you need is docker desktop like like that's like the
75:10 main requirement here and and the database of course but now you can just
75:15 start the containers and you can either use a remote Postgress like with
75:21 superbase or neon or you can use the local Postgress that we have built right
75:24 into the application. So if you don't want to go and pay for superbase or set
75:29 up a neon project or whatever and both of those have free tiers by the way so
75:32 it is very viable but if you don't want to like set up something else you can
75:36 run the postgress as a part of the application and so we have that like
75:41 built into the docker compose here and so you can pick either one like I
75:45 generally just like have a remote database in the cloud so that way I can
75:49 deploy to different devices and they can reference the same database and so I
75:55 typically go with the external opt here because the other beautiful thing with
76:00 that and just the fact that I have database persistence set up for you with
76:04 this system is you could have it running on your desktop one day and then your
76:08 laptop the next day, but you still have all the commands loaded for your code
76:12 bases and you have the different conversations that you can resume. And
76:16 so that's a really nice um thing just with the database persistence here. And
76:20 then I've got a list of all the commands that you can run as well. And so this is
76:26 your command center here to manage the status and the different commands that
76:30 are loaded into your repos. And you can clone different repositories. And so if
76:34 you're starting completely from scratch, this is the first command that you run.
76:39 Like you'd go into Telegram here and I'll just do a reset quick. You go into
76:44 Telegram here, you do uh slashclone and then you just grab the URL of your
76:47 repository. Like if I wanted to clone the um the Dynamus agent that I'm using
76:52 for my demo here, you just you copy this and you go in paste it. Boom. I'm not
76:55 going to send this in because I've already run this. But like this is how
76:58 you get your coding assistant like in this remote coding system set up for a
77:04 codebase for the first time. And then from there you can just send
77:07 off any requests like you would typically do with you know cloud code or
77:11 codeex or whatever. Like everything operates exactly the same. It's just the
77:15 way that you work with commands is a little bit different. So when you want
77:18 to use your slash commands for validating or priming or whatever like
77:21 I've been showing you that that's the only part where you have to like load
77:25 them into your system and then you can use them with the command invoke. And
77:31 I'm probably planning on evolving this to make it easier to use this kind of
77:37 like command system as well. Um the the one downside to this like I just want to
77:40 acknowledge like the system isn't perfect yet. I'm working hard to evolve
77:45 it as well. One of the downsides is you have to remember the specific commands
77:50 that you have loaded. Like sure I can just do like slash commands to see what
77:55 I already have and then I can do you know like command-invoke
77:59 and then I can do u you know system review for example I can invoke this
78:03 command. But one of the things that we have in claude code for example that's
78:07 really convenient is we have the autocomplete like I'll just show you
78:10 really quickly and I think we have this in like codecs and stuff as well. Um but
78:18 if I go to my agent here and I'll just like I just want to show
78:22 what I mean really quick. I'll go into claude and then I'll do like slash um
78:27 yeah like for example system review, right? like it autocompletes for me so I
78:30 don't have to know exactly what the full name of the command is
78:34 [cough and clears throat] excuse me but versus like our custom system here if we
78:39 forget the name we have to do slashcomands first and then we know like
78:44 okay it's dash or slashcomand-invoke and then system review so maybe having like
78:48 something I've thought about building into this is um a process where you
78:53 could say like I want to use my GitHub prime command right like I want to use
78:58 my GitHub prime command in and it'll like search the registered commands and
79:02 it'll like be intelligent enough to figure out like okay this is the one
79:06 that they meant to use. They just forgot the exact name of it. So like let me go
79:11 and invoke it now. So I'm thinking like really in the weeds like even like the
79:14 little things that just make this so convenient and powerful for you. I I'm
79:18 thinking about all those things. And so like what you have right now that is
79:23 like given away for the live stream and only the live stream. It's still a
79:27 really powerful application, but there are like those little things that I want
79:31 to keep evolving. And so like that is what I have in the the AI agent um or
79:36 the agentic coding course in Dynamis. So just calling that out again like we have
79:42 a a module that is dedicated to this system and using it for like more
79:45 complex things than I'll cover in the demo here. And then um also like this
79:52 system is not just going to be in the course one and done. like I'm going to
79:56 keep evolving it as a primary resource for our community here. And so like if
80:02 you want to stay on the cutting edge and like you really believe like me that the
80:07 future of coding really is like using these remote coding systems then yeah
80:11 definitely come be a part of Dynamus. I'd love to have you. And um yeah, for
80:15 those of you who are who have just come to the stream and you haven't uh heard
80:19 already, like we got a special offer for Dynamus for just this live stream, 45%
80:25 off the annual, 25% off the monthly. This is going away right after the live
80:31 stream is done. And so take advantage of this now. Love to have you in the
80:34 community. And if you're watching the recording, um like just know that
80:38 there's still a great sale going through Cyber Monday as well, but it's not as
80:42 good as this. And so definitely take advantage of that. So I'm gonna put this
80:48 um in the chat really quick as well. Boom. There we go. All right. So with
80:54 that, uh what I want to do with you now, so now that you have you know how to set
80:58 it up and like how it works as a whole, we can do a quick demo here. And so I'm
81:03 going to go back to my codebase here and I want to show you what it looks like in real time. We're
81:09 going to kick off the prime and then we're going to go through a plan. We're
81:12 going to think about like what we'd want to actually implement as something new
81:16 in our front end here and we'll see it go to staging so we can validate it
81:19 first and then we'll tell it to like make that pull request into main so we
81:24 can deploy the changes to production after we have validated the artifact
81:28 essentially that the coding assistant gives us. So that is that is the full
81:32 demonstration that we're going to to go through here. I'm actually really
81:35 excited for this. And um things might not go perfectly because we're doing a
81:39 live demo with AI coding assistance. That's always the risk that we have. But
81:43 yeah, I mean like I trust my system. If anything messes up, it'll probably just
81:47 be me forgetting to uh prompt it properly because it's always important
81:50 for you to prompt properly even when you do have a system. [clears throat and cough]
81:57 Excuse me. So uh let me go and create a new issue here. So I'll go to issues,
82:03 create a new one. And so now you can think of this like if you've been in
82:07 software engineering for a while or at all, you're probably familiar with this
82:11 process where when there's a bug in the code or you just have like a new feature
82:15 that you want to build into a codebase. Typically you'll start by going into
82:19 your remote repository and you will create an issue and I can
82:25 even like say like labels. I want to make this as an enhancement. So I want
82:27 to do something new. I'm not working on a bug fix, but I could obviously label
82:32 that as well. But yeah, right here I'm working on an enhancement. And so now we
82:36 can think about like, okay, what's something cool that we could build into
82:39 our front end here? And for the sake of this demonstration, I want to do
82:43 something that is only affecting the front end of our application. This is
82:48 the the Dynamus AI agent that I show you how to build completely end to end in
82:52 the Dynamus AI agent mastery course, which you also get access to immediately
82:56 when you join the community. you get the whole AI agent mastery course and the
83:01 agentic coding course and both of those are very very value packed. But um yeah,
83:05 I did think about ahead of time like what are a couple of cool features that
83:09 we could build into the front end just as a live demo here. And um so yeah, one thing that I thought
83:19 about doing is it'd be really cool if we could export our conversations
83:23 like uh I was doing like some testing with the agent earlier just as I was
83:27 prepping for our workshop here like testing the web search and the the rag
83:31 tools and things like that. And so like what if I want to export this
83:34 conversation as a markdown document for example like right now in our user
83:39 interface. There's no way to do that. But maybe it'd be cool if I, you know,
83:43 like for example had a a button down here where I could click on something to
83:46 like export and it would like download the conversation as a markdown. I think
83:50 that'd be really cool. So let's go ahead and do that. So add markdown export to
83:58 the front end. And so like for the issue description here, we'll just keep it
84:00 really simple for the demo. Like currently there is no way to export a
84:07 conversation in the UI. We should have a button next to the file attachment. So
84:11 just kind of giving a little bit of a a spec here. We should have a button next
84:16 to the file attachment to download the conversation as a well formatted
84:22 markdown document. Now typically for an issue you also would provide quite a bit
84:27 more of a description especially if it's something a bit more advanced. this is
84:30 definitely going to do for our demonstration. So, we start by creating
84:34 an issue and then we could go in Telegram and we could tell it to like if
84:38 I do a reset here, we could say like um go look at the latest GitHub issue and
84:45 uh let's start planning, right? Like we could do something in Telegram as well.
84:48 So, this is going to work no matter the application, but for this demonstration,
84:54 I'm going to chat with the agent um directly in the issue. And so like for
85:01 example, I can do slashhelp. Uh well, hold on. I have to do at remote-
85:06 agent/help. And so in just a second here, it'll respond very quickly like we saw with
85:12 Telegram when we do a slashhelp. So there we go. Here are our available
85:16 commands. Um great. So now I want to do at remote agent and then I'm going to
85:21 invoke our first command here. And if you remember from earlier in the stream,
85:24 the command that I always start with is priming. So I will do command invoke
85:28 prime GitHub. So before we go and implement this feature, I wanted to
85:32 understand what our codebase currently looks like and where maybe you would
85:37 have to go to edit things to add this feature into our front end. So I'll
85:42 start by kicking off our prime GitHub command. [clears throat] Excuse me. Now, this is
85:49 going to take a while because when we prime, we are looking through our
85:53 codebase pretty extensively to make sure we know everywhere that we have to touch
85:58 for this feature implementation. And so, what you're what we're looking at right
86:02 here is the Docker container logs. And so, because it's not a real time thing
86:06 like Intelligram, we don't get to see all the actions it's taking. I mean,
86:11 obviously like this is a lot of spam for our GitHub issue, but we can look in the
86:15 logs and just see like really quickly like here are the tool calls that Claude
86:20 is deciding to do. Like bash is really just like most of the commands that it's
86:24 running to like list directories and things like that. And then whenever it's
86:27 reading an individual file, we see the read here. And there's a lot of other
86:31 different kinds of tool calls it can make like to-do write when it's managing
86:35 its task list. We can see all this in the logs, which I mean typically when
86:38 you're just coding things remotely, you're not looking at the logs. I'm just
86:41 showing you this as a part of the demonstration if we want to like peer
86:45 into what's happening under the hood before it finishes and and gives the
86:50 final output as the comment here in the GitHub issue. And so while we wait for that to go, um
86:57 I can just like kind of throw up some more questions here. So, that's also
87:01 kind of the fun in this live demo is um while we're waiting for the coding
87:05 assistant, it's an opportunity for me to just like go through some more Q&A,
87:08 which is always a really fun part of the live stream. Like I always love just
87:12 like going back to the chat and like as much as I love like really diving into
87:15 the setup and the architecture that we'll get into in a little bit, it's
87:17 always a bummer where like I kind of see on my other monitor there's a ton of
87:20 questions coming in but I can't get to them. So, I will definitely do that now.
87:28 Um, yeah. So, robot Glock, uh, need a shorter alias for command invoke. Yeah,
87:33 I actually do agree. Like maybe like - CI or something or like just slash
87:39 invoke. Um, yeah, like I I I should probably do that. So, I appreciate that
87:43 suggestion. And by the way, you could add a shorter alias in probably like
87:48 five minutes with an AI coding assistant. So again, this code base is
87:51 optimized in a way where it's going to be really easy for you to build your own
87:55 things into this if you want. And so please keep that in mind. It's like
87:58 anything that you don't think is ideal here. Just like throw it to your coding
88:02 assistant, codeex, cursor, quad code, whatever, and just see what it can do.
88:06 Especially if you've gone through the Dynamus, aentic coding course and like
88:11 you have this whole architecture set up, like you have a system for coding,
88:14 you're going to be able to knock out things very easily. um adding on top of
88:23 Uh let's see how is Langfuse looking with this long running command. Uh so
88:27 yeah, you probably noticed that I have the Langfuse dashboard set up here, but
88:33 this is actually monitoring my agent like the application that we're working
88:37 on for the demo here. I don't have Langfuse incorporated into the remote
88:42 agent coding system. Um but I I think that would be cool to add. So that is um
88:46 I actually really appreciate you bringing this up because adding more
88:50 like observability remotely so we're not just like looking at artifacts and
88:55 looking at the logs and everything. Um that is something that I am considering
88:58 doing and I did actually do a proof of concept uh last month integrating the
89:05 codeex SDK with um Sentry. So Sentry is another good like agent observability
89:08 platform. They used to be like just generic application monitoring and
89:12 they've done some AI specific stuff as well. So, I mean, I probably would do
89:16 Langfuse. Um, but yeah, that is another thing that I'm considering. But yeah,
89:21 this one right here is um tracing and monitoring for the Pantic AI agent that
89:28 is operating under this application. But anyway, uh the priming is complete.
89:32 I'll get to more questions when we kick off the planning here. But yeah, it does
89:37 uh this prime command specifically, and again, these are all just markdown
89:40 documents that I have as commands that I've loaded into my system. And this one
89:44 I'm walking it through doing a pretty comprehensive overview of the project.
89:49 So it isn't just specific to this issue here, but like in doing this it's going
89:53 to understand um where it has to go like in the front end for example. So it
89:58 knows our text stack, it knows our core principles from our global rules.
90:03 And then also this is another really cool thing the remote agent coding
90:07 system whenever we start a conversation in a GitHub issue or pull request it
90:14 also automatically injects the context from the issue. So when we start off the
90:19 conversation here it knows like this is my issue number issue number nine. It
90:23 knows the description of the problem and the labels that we have on it for
90:27 example. And so we can see at the bottom of the prime here that it even knows our
90:33 current state and it knows, [clears throat] excuse me, it knows the different
90:37 commits that we've made recently. It knows that we're currently working on
90:40 this issue and that it is of type enhancement. So this is really cool.
90:44 Like the prime command, it's always good to review it to make sure that your
90:48 coding assistant really like understands your codebase and what you're currently
90:52 working on. And clearly it has a very good understanding here. And so now, I
90:57 mean, I could just like say remote agent um go on and build this, but I'm I'm
91:01 going to continue with my system here. Even though this is a simpler
91:05 implementation, I'm going to start by using our plan feature command. So, plan
91:10 feature GitHub. And then the one argument that we specify here is exactly
91:16 what we want to plan for. And so, since we already have the context of the
91:19 issue, there's not that much I have to say here. I'm just going to say let's go
91:25 ahead and plan out this issue and um we'll say with a simple plan. I just
91:29 want this to be quick for the demonstration. Uh let's just do um unit
91:36 testing for the validation. I don't want to spin up the site or anything. Keep
91:42 this brief. All right. Because my planning command, I have what's called
91:45 the validation pyramid where I go through linting and type checking and
91:50 unit testing and end toend testing. If I were to run this in a longer workflow,
91:53 I'd actually like spin up the website and have it visit it and make sure
91:57 things look good. But just for this demonstration, I do want it to be
92:01 quicker here. So, yep, we are invoking the command plan feature. And then I'm
92:04 using double quotes here because I want this to just be a single parameter
92:09 outlining the description of the feature that we are now planning for. And then
92:13 we'll take the plan and send it into our execute command. And so if you have been
92:19 in Dynamis or gone through the agent coding course, like what I'm doing here
92:24 is very standard to the piv loop that I always do within my systems even when
92:29 I'm developing things locally. And by the way, we can see here that it is
92:33 resuming the session. So we did the priming. Now we're staying in the same
92:37 session here and we are doing all our tool calls now to create that structure
92:42 plan in a new feature branch. So we have the the get native workflow built into
92:46 the command here where it knows that like okay once I do my planning here and
92:51 I create that plan markdown document I'm going to create a new branch and I'm
92:54 going to add the document there and then we can throw that into the execution
92:59 next. And so this is like very much the pivot loop that I was showing in our
93:05 architecture earlier. So um for the planning phase, the commands that we use
93:09 mainly are the prime and the plan feature and those are [snorts] the two
93:13 that we've already kicked off here. So we have gotten to the point now where we
93:17 are creating our structured plan. Now typically I would recommend spending a
93:22 lot more time in this planning. Like I'm I'm being very quick here for the sake
93:27 of the demonstration, but yeah, like [clears throat] sometimes
93:30 you'll spend a half hour, even an hour in the planning phase because there's a
93:34 lot like when you're doing a more complex implementation, you really have
93:39 to make sure that you and the AI coding assistant are on the same page with what
93:43 exactly to build and how to go about building it. And then once we have our
93:48 plan, you also generally want to iterate on the plan. like make sure that you
93:52 read through it and that it looks really good. I won't do that either here
93:56 because I I don't think that's like worth our time right now. So, we'll go
93:59 right into the implementation, but typically at each step of the way,
94:03 you're going to validate things. And the main main validation I want to focus on
94:07 for this demonstration is just like that artifact that we'll see at the end where
94:13 it publishes things to the staging URL automatically. So, that that's the main
94:20 thing that I want to focus on here. But yeah, while I uh wait for it here,
94:25 I'll just uh do some more questions. And so I'll I'll just kind of like show the
94:30 logs here. So we're still in process. So Um if I get the monthly subscription,
94:38 can I pause the membership for a couple months and come back the same $60?
94:42 Unfortunately, you cannot. So if you lock it in for life, you have to lock it
94:46 in now. and there is no ability to pause. Um, don't see the prime GitHub command
94:53 in the codebase. So, the prime github command is something that I have. So, that's that's
95:00 a good question, Sean, because I want to be very clear here. The commands that we
95:06 use in our remote coding system, they are in the repository we are working on,
95:12 not in the repository for the system. So when we load in our commands like if I
95:20 go into telegram here and I do um like here I'll just do a slashhelp again
95:23 to show you the commands like we have the um load commands when we give a path
95:29 here like commands this is a path in the repository that we're currently working in like I
95:38 am I have I'm currently working in the Dynamus AI agent repository so we're
95:42 using the command commands from there. So the reason you don't see it in the
95:46 remote agent coding system is because I'm not loading in commands from that
95:50 repo. I'm loading it in from the repo that I'm actually operating in with my
95:55 system. I hope that makes sense. U but don't worry all of the commands that I'm
95:59 covering here. I have that as a part of the agent coding course in the the
96:03 repository that we have in Dynamus there. So, um, yeah, we I got you
96:07 covered, Sean, because I know Sean that you are in the the Dynamist community.
96:13 So, have to run. Thanks, Cole. Good luck, everyone. Appreciate it. You're
96:16 probably gone now, but thanks for being here. Uh, you mentioned the use of GitLab as
96:22 well. How does that come into play? Yeah, so GitLab would have to be another
96:27 application that we integrate with. The reason I mention it is just because the
96:31 way that it works is going to be very similar to uh how we work with GitHub
96:36 because you still have like issues and pull requests there. Um but it's not
96:41 like this application right now out of the box would also work with GitLab.
96:45 We'd have to set up another integration, right? Like going back to the readme
96:50 here for the our remote coding system, it would be another platform adapter
96:56 that we'd have to build in. Um, but you could do that very easily. Like if you
96:59 are legitimately interested in using this system with GitLab, then you like I
97:04 said, I've set up this codebase in a way and I've architected it where it is very
97:11 easy to add more platforms into this thing. Like you can add a new platform
97:16 in without having to add the custom integration with each of the individual
97:21 coding assistants because we have this orchestrator as a middleman. So, I'll
97:25 get into the architecture more maybe towards the end of the stream. It's
97:28 really interesting the way that this is architected. But yeah, you could add in
97:32 GitLab um like along with GitHub and Telegram very easily, but it would have
97:37 to be something totally separate. Hope that makes sense. So, yeah, it
97:41 looks like going back to our um issue here, the plan is done. So, we'll get on
97:46 with this in a second. I just want to take a [clears throat] a little bit of
97:50 time to answer a couple more questions here. Is there a link to a GitHub repo I can
97:57 clone and try myself? Um, yeah, good question. So, well, honestly, I would
98:03 start with a private repository. Um, but if you do want to like try this
98:07 really quickly with a public repo, you could really just try any repository.
98:12 For example, you could use Archon. So, if well, if you just search if you just
98:17 Google Archon GitHub or if you go to archon.diy, DIY. You could pull in this
98:22 repository. So you could do like in Telegram. Let me copy this quick. You
98:27 would do slashclone and then paste in that link that I just copied from Archon
98:32 right here. So you copy this, you clone it. I'll actually do this right now. So
98:37 slashclone. Um, oh wait, maybe I shouldn't have done that because I need to be operating in
98:43 the in the the Dynamus agent for the the demo here. But anyway, once you clone
98:47 the repository and I do get current working directory, it is automatically going to change me
98:54 to this repository. And now since this is not your repo, it's not like you can
98:57 make commits and things, but you still can, for example, say um summarize the
99:02 read me. So you can do like read operations within the public repository
99:08 and you could even change things locally, but it's not like you could go
99:11 through the full workflow that I'm demonstrating here where you would like
99:14 make a pull request and stuff because obviously um you wouldn't have
99:18 permission to do that. Well, I guess you could because you can make pull requests
99:22 on public repos. So yeah, I mean honestly if you want to use a public
99:25 repo, I guess I have to correct myself a little bit. I mean you can pretty much
99:28 like do everything. you just won't be able to like merge a pull request or
99:32 make pushes directly. But you can see that like what I did right here is a
99:36 quick example of basically starting from scratch. I clone the repo and then I
99:40 just like asked it a simple question. I can also ask it to make an edit for
99:44 example. But yeah, I definitely want to like switch back to um the Dynamus
99:51 agent. So slashreos and then slash um so if I do slashhelp I'll show you that the
99:55 command that I need to run here to switch back is set current working
99:58 directory. So slash setcurren working directory then paste in the path to the
100:05 dynamis agent and so now now we're back. Okay, very cool. So now what we want to
100:12 do if we go back to let me clear the feature message. Okay, so if we go back
100:18 to our issue here, [snorts] we have our final summary. So here's the
100:22 implementation plan. It was created in this new branch that it made. And then
100:27 the plan is located here. So we can actually go to that now. If I go to if I
100:31 refresh, so I have access to the new branch. Uh and then I click into aent/plans.
100:38 Take a look at this markdown export. We have this document. So the structured
100:43 plan that I showed you earlier in our slash command where we have the user
100:47 story, the problem statement, the context references, it basically took
100:51 that template, did a bunch of analysis of what we want to build, and now it
100:55 created the structured plan that is all of the context I need to now send into
100:58 [snorts] my execute command. And so I'll just scroll through this like really quickly
101:03 here. we don't have to focus on this, but we have the implementation plan and
101:08 the step-by-step tasks that we have to knock out. So, there's a lot of
101:11 direction and guard rails that we're adding for our agent when we're working
101:15 with a structured plan this way. This is very much not vibe coding because you
101:19 can validate this as much as you want and be a part of the process. You
101:24 understand what it's going to build and how it is going to build it as well.
101:28 Like it even has like pseudo code here that we can review to make sure that all
101:31 this looks good to us. And so if I wasn't doing a more brief demonstration
101:35 here, I would definitely dive into each one of these things and make sure it all
101:38 looks good. And that's how I can be in the loop. And so even with a remote
101:42 coding system, I could just like go to this file even on my phone and just
101:46 quickly review the markdown here for the structured plan before I send it off to
101:50 execution. So I'm going to say this looks good. And so then we can go on to our execute
101:58 here. And so it even gives me the next step. So I have this uh kind of like
102:03 built into the prompt for the plan feature GitHub command where it tells me
102:08 like after the structure plan is created you can use the command invoke we're
102:13 using the execute GitHub command now and then the two arguments that we need here
102:17 like I showed in the other uh really quick demonstration earlier is we need
102:21 the path to the structured plan and then obviously the feature branch that we're
102:25 operating in because we don't want to operate directly in our main or even our staging
102:31 at this point because what this is going to do is we make the execution in this
102:35 feature branch and then we're going to make a pull request to go into staging.
102:39 So I can ask it to like merge that pull request and then it'll automatically
102:44 deploy things to my website here so I can validate things and put myself in
102:49 the loop in that way. And so let me go down here and I will do uh command
102:56 invoke execute GitHub send this off. And boom, we are off to the races. And by
103:00 the way, the reason why we see two comments here is because I shouldn't
103:05 have switched to the Archon repo and switched back. It confused the the
103:10 coding agent. Um but that that's on me because I was just working right here.
103:16 So you can work in parallel with this system but just on a single repository
103:21 right now because if you switch the current working directory it's going to
103:26 confuse the coding agent. Um and so that is another thing that I'm working on for
103:30 evolving the system where we can have different applications or even different
103:34 threads in the same application like different Slack threads where we could
103:40 have like different working directories. And so that's a bit more of an advanced
103:43 thing that I'm looking to build out as well. But yeah, I'm gonna I'm gonna keep
103:47 plugging this because like this is important for me. Like I'm gonna keep
103:49 building this out for you guys. So if you want to be a part of the Dynamis
103:53 community where I'm going to take this system and continue to evolve it and you
103:57 want access to the complete Agentic coding course, like please take a look
104:01 at this. You're you don't want to miss out on this opportunity because it is
104:05 going away once our live stream is over here. And so I will uh put it in the
104:11 chat really quickly again. There we go. Send it off. So yeah, this
104:15 system like just like talking really quickly about how I want to continue to
104:20 evolve it. I want to integrate Slack and notion. I want to add in Open Code and
104:25 Klein as more options for coding assistance. I want to make it so that we
104:29 can work in different repositories in different applications at the exact same
104:33 time. Right now you can do parallel execution just on the same codebase. I
104:38 want to make it so that the commands are easier to work with. Even having like an
104:42 alias for the command invoke like we talked about earlier. There's and like
104:46 adding an observability like langu also like we talked about earlier. Like I I
104:51 really appreciate all of you guys like giving ideas here. U because like yeah
104:55 and these are awesome like I want to incorporate these. Um and then the other
105:00 big thing I Okay, so someone asked about Archon going out of beta. Let me
105:05 actually address this here. The reason that there are not that many updates on
105:11 Archon right now is because I am actually considering how we can take
105:17 this remote coding system and build it in as a part of Archon. Now, this is
105:21 something I wasn't planning on talking about too much during this live stream
105:25 because it's still like really like just we're in the initial stages of thinking
105:28 about how we would do this, but I think that you guys would find this exciting.
105:33 Um, Archon, if for those of you guys who don't know, that is my open source
105:39 project that's kind of like the the AI u, let me go to the repo here. It's sort
105:44 of like the AI agent command center for coding assistance. And so, this is fully open source and
105:50 free to use. I would highly recommend giving it a shot if you guys are curious
105:56 to like how can I add rag, how can I add task management, like actually good task
106:01 management into my AI coding assistance. That is what archon is for. And it was
106:06 even the number one repo on GitHub for three days in a row when I first
106:11 released the um the new beta version of Archon. And so, yeah, there's a lot of potential
106:16 with Archon, but one thing that we've realized, like me and and Raasmus and Sean and all the
106:25 guys on the Archon team, is that um as much as the the rag and task management
106:31 components of Archon are very powerful, we want to do something more where
106:35 Archon can be a place for us to actually orchestrate different coding agents as
106:41 well, doing a lot of the things that I'm talking about with the remote agentic
106:45 coding system and there the possibilities here are endless and I'm
106:49 actually like so excited about this that's why I have to talk about it right
106:52 now even though I wasn't really planning on it but like if we could build
106:57 something like this directly into Archon so that um man I could talk about this
107:01 for hours and hours but like having a place for you to kind of manage like
107:05 what we are calling our agent work orders and this is like already in alpha
107:09 like we're working on this where you can like kick off your different agents You
107:14 can assign the knowledge that it has access to for rag. You can assign the
107:18 projects that it's going to manage tasks in. And you can even like inject your
107:22 own commands and like create your own system directly in archon. I think that
107:27 that is just going to be so incredibly powerful. Um, but like because that's
107:33 such a big thing, that's why we've like kind of like I don't know, we've been
107:37 more like high level recently with Archon. Excuse me. And so that's why we don't
107:42 have like a ton of like new features that we built out recently because um
107:46 and this happens with open source projects really any software engineering
107:51 project in general is you go through these different waves of like sometimes
107:54 you know exactly what you want to build and it's just like go go go like let
107:58 let's uh build all these features that we know exactly what we want to create
108:03 and that's what Archon was back in like July August and part of September as
108:08 well but then like the other phase or like the other part of a wave with any
108:12 project is you go into kind of like the research phase where you're like okay we
108:16 have this strong base now I mean it's not perfect but we have a strong base
108:20 but then it's like oh actually the industry is going towards this direction
108:24 where it's like remote agent coding is huge like okay this is a huge
108:27 opportunity we need to build this out for archon and so then we go through
108:31 this phase of just like planning and thinking about how we build these kinds
108:37 of things and so the remote agent coding system I built this to be a core part of
108:41 the Dynamus Agentic Coding course. Like don't get me wrong, that's the primary
108:45 purpose, but then there was kind of this underlying secondary purpose of like
108:51 let's see what we can build here to integrate [snorts] a coding agent with
108:55 Slack and Telegram and GitHub and all these different applications like
108:59 actually using the agents where we work. That's the value proposition that we
109:05 have here. But then also what if Archon is one of those applications? What if
109:10 Archon is one of these platform adapters that we are covering during the setup?
109:13 So, it's like you can connect it to Telegram, you can connect it to GitHub,
109:17 or oh, you could also connect it to Archon. I think that would be freaking
109:21 awesome. Or even making this like a core part of Archon. Maybe I'm saying too
109:24 much like going down that route, but like that's the kind of thing that we're
109:28 thinking about right now. Uh, and so it's pretty exciting. I'd also like
109:31 really appreciate your guys' input on that as well. like how much you find
109:35 value in like just what we have with Archon right now and continuing to build
109:39 that out versus like incorporating this in along with the other things that we
109:44 already have for Archon. And so yeah, I'm definitely planning on doing some
109:49 workshops in the Dynamis community where I actually like integrate Archon as a
109:53 platform adapter and um yeah, just kind of like shape the vision with you guys,
109:57 which of course like I'll share stuff on my YouTube channel as well and like the
110:02 all of the free valuable stuff that I do on YouTube is never going to go away,
110:05 but [snorts] the Dynamus community is definitely the place to go deeper on these things. Like
110:12 if you're really interested in Archon or really interested in the remote agentic
110:16 coding system or really interested in building your own systems for AI coding
110:20 like that is what Dynamus is the place for because I'm I'm active on YouTube
110:26 and Dynamus every single day. Yes. But I'm not making a YouTube video every
110:29 day. I'm doing things in Dynamus literally every single day. Weekly
110:34 workshops, working on the courses. We actually have like a a live event every
110:39 single day. Like if I go to um the calendar here in Dynamis, I'll just show
110:43 you really quick. Now, there's like a couple of days off that I had this
110:47 month, but like yeah, there is something happening in the community every single
110:52 day and even some new stuff coming up that I can't talk to or talk about that
110:56 much yet. But like looking at October, you can see it's the exact same thing.
111:00 Besides the weekends, obviously last month and the month before, um there was
111:07 not a single day where there was not an event in the community because I'm just
111:11 pouring everything I have into this. Uh so I hope that sounds good to you. Like
111:15 I mean the amount of effort that I put in Dynamus is crazy. So if you want to
111:19 be a part of it, like I'll I'll put the link in the chat again.
111:23 Um because yeah, this special is going away after the stream. And what you get
111:28 in the community is um yeah like 1,200 people that we have in the community
111:31 right now that are it's like just the amount of engagement that we have on the
111:35 platform is insane. You get the AI agent mastery course, the agent coding course,
111:40 weekly workshops, daily support. Uh it's just it's a blast and it's actually just
111:45 like an incredible honor for me uh to be able to to lead this community every
111:50 single day. But yeah, let me uh go back here and um answer a couple other
111:54 questions. I saw a couple people saying that makes sense. So, um, yeah, Archon
111:59 is amazing. I appreciate it. Yeah. And more coming for you. Like I said, we
112:01 just are kind of in that phase right now 100% understood. We wait for greatness.
112:10 I appreciate it. Thank you for your patience and understanding. It's going
112:13 to be awesome. Yeah. Thank you, Sean. Hey, Sean's doing great work as he
112:23 Let's see. I vote for making this a core part especially building and sharing
112:26 your agentic workflows. Yeah, that's another thing is Archon can also be a
112:31 marketplace kind of like claude plugins if you guys have played around with
112:34 cloud plugins before but it's a way for us to share our systems like oh here are
112:39 my rules that work really well when I'm working on an Nex.js JS app or when I'm
112:43 working on creating an MCP server or here are the commands that I use for my
112:47 pivot loop like here's my prime command here's my plan command here's my
112:51 validate command and you can like give it to other people and use and it can be
112:55 like this ecosystem in archon where you can kick off an agent to start a new
113:00 project and then not only can it be like your own system but you can also like go
113:03 to this marketplace and pick out a system that someone already created and
113:07 like leverage their work or you [snorts] could like automatically incorporate
113:10 BMAD or automatically incorporate the getup spec kit and like that would be
113:14 injected as the process for your agent that you're kicking off from Archon. So
113:17 just yet another way that we could do some like really insane things with
113:22 Archon. I'm very very excited for um so let's see another good question just
113:29 focusing on AI coding in Dynamus. Are you planning to go back to agent
113:32 building at some point? Yes. So I still do a lot with agent building in Dynamus
113:36 even through like some of the weekly workshops. So like as I've been going
113:40 through the agent coding course, yes it has been more of a focus in the
113:44 community but definitely like still a lot that I'm doing for building AI
113:49 agents and um actually like now that the agentic coding course is complete the
113:53 next like set of work that I'm doing for the community. Obviously a third course
113:57 is coming at some point and like the weekly workshops I do want to go back to
114:03 focusing on building AI agents more as well. And also there are a lot of
114:06 opportunities to do both at the same time, right? Like we can use our systems
114:12 for AI coding to create AI agents. And I think that's like really really
114:15 important. Like you're never going to just like build an AI agent by um
114:19 looking at the pinantic AI documentation yourself now. I mean like you can do
114:23 that but like we can use archon or just have it look u use web search to find
114:27 the pinantic AI docs and then help you build the agent. And so there's a lot of
114:32 overlap there as well. like um one of the next courses that I'm thinking about
114:36 releasing for Dynamus. There's a few ideas that I have, but potentially one
114:41 of the next ones is a course where like each module is walking through creating
114:47 a specific AI agent end to end that actually has real world value. I'm very
114:52 excited for the possibility of doing that course. And so with that, it'd be
114:57 kind of like, so we have the AI agent mastery course where it's a ton of
115:00 modules going you through start to finish. Here is how you think about
115:04 building AI agents and how you practically do it. And then this new
115:07 course would be like taking the principles that we cover in that that
115:11 first course, the AI agent mastery course, and then applying it to specific
115:16 real world use cases one module at a time. But then the reason this goes back
115:22 to your question is um we'll be using AI coding assistance to build these agents.
115:26 Like we'll still cover the principles of like here's what we want to consider for
115:29 our system prompting and our tools and how we're maybe orchestrating multi-
115:33 aents. Like we're still getting into the technicalities of things for people that
115:36 like care about the deep stuff, but we are using AI coding assistance to speed
115:40 up the obviously the code generation part of it so we can you know have the
115:45 time dedicated to the really high leverage stuff when we build our agents.
115:49 And so I want to be clear when we're using coding assistants to build our
115:53 agents or really any application like we're doing here, we're not having the
115:57 coding assistant replace us. We're having it do the grunt work so that we
116:02 have even more time to dedicate to the higher level things like the
116:05 architecture and the validation of our system. So I hope that makes sense. But really
116:11 really good question. I appreciate you All right, cool. So, with that, I know there's a
116:22 lot. I was sorry. I was like looking at some of the other questions we have in
116:28 the chat here, but um yeah, I want to go back right now to the repo here because
116:35 we have our our implementation now. All right, hold on. We got an error.
116:41 Why' we get an error here? Oh yeah, because it the it did the Okay, so there
116:47 is unfortunately a bug with claw code where whenever it uses the kill shell
116:53 tool call, it crashes. It's really, really unfortunate. This is not an issue
116:56 with my system. I actually did some research because I had this happen last
117:01 week where when Claude Code tries to spin up a website and then kill it, it
117:06 actually ends itself. So Claude Code unfortunately killed itself here, uh,
117:10 which is very unfortunate. So, I'm going to see if I can kick it off again, but I
117:14 might have to restart the process. That is something that um yeah, so there
117:19 there's actually an issue out right now. Cloud code kills or no um kill shell crashes cloud agent
117:30 SDK. I think this Yeah. So, there's a a bug that's reported right now um that
117:35 unfortunately is not being addressed yet. And so like I've run into this a
117:39 couple of times now where cloud code will kill itself when it's uh trying to
117:43 terminate like a Node.js server for example. So what we had happened here is
117:49 as a part of the validation it tried to restart or like spin up the website and
117:54 and view the website and then when it killed the background task after that's
117:58 when it crashed and so yeah I might have to restart the implementation here. Um
118:05 well okay cuz it might it might try to do the same thing again. So there's
118:08 always these little kinks where it's like unfortunately we also have to work
118:11 with the limitations of the platform itself. Now this is a really specific
118:15 thing that claude code will probably address hopefully really soon. It's too
118:20 bad this is still an open issue. But also in our prompting, I should have
118:24 just been more clear that we don't want to do that deep of a validation because
118:29 unfortunately even though I said right here that um I just want to do simple
118:36 validation, it's not really enough because my my structured plan has the
118:41 full validation pyramid built in. So I just confuse the coding assistant like
118:45 that. That is unfortunately uh my problem here. So yeah, going back here.
118:50 It's restarting, so it'll probably be good. Otherwise, I'll have to just tell
118:55 it to um I'll just have to tell it to like explicitly like do not use the uh
119:02 structured plan u validation like you just want to do the unit testing. So
119:05 we'll let that go here. But anyway, that gives more time for questions. So it's
119:10 totally good. Uh Gemini does the same thing. Okay, that's really interesting.
119:13 Yeah. So it's it's not really Anthropic's fault. I mean to be fair
119:18 this these like little issues where coding assistants will like kill
119:21 themselves for example like it's it's really not surprising that it happens
119:27 because this whole idea of having agents work so autonomously within an
119:31 environment like these kinds of things are bound to come up until we evolve the
119:35 the platform and like have guard rails to avoid these kinds of things. Like the
119:40 core issue here is that it it should just be using a better command to end
119:43 the process so that it doesn't kill itself as well. And I would have to do
119:47 more research into like what exactly that is. But once I have that, I could
119:51 inject it directly as a part of the prompt in the system. So you wouldn't
119:56 even have to inject that into your command yourself. Like it' just
120:00 automatically be taken care of. And like those are the kinds of things that I can
120:04 customize with the system to make it even better for you. Or we could just
120:09 like build it into our prompts where it's like when you spin up the website,
120:13 make sure you kill it in this specific way so that you don't, you know, end
120:17 your own session as well. That's pretty unfortunate. But yeah, clog code,
120:22 they've got um 5,000 issues apparently in their repo, which is absolutely nuts.
120:25 I don't know when they would get to it exactly, but there is definitely a
120:28 workaround if I really wanted to dive into it. Um, so it's just like one of
120:32 those like little things where I mean I just haven't gotten to it yet, but we'll
120:36 get there at some point. So anyway, this implementation hopefully
120:41 I won't try to spin up the site again. Um, but I should actually look. See,
120:45 this is where like I didn't validate the structured plan because I wanted to be
120:49 brief for you, but I probably should have because there might be an issue in
120:53 the validation here where it is um telling it to like be really
120:57 comprehensive. Yes, it is. So, integration testing, it's starting, see,
121:01 it's starting the container here and then it is um it's going to try it's
121:05 going to try to kill the session again. So, we're going to see the same thing
121:08 happen, but I'll just prompt it again where I tell it not to do this or or I
121:15 could edit this file in place. So, I can do my own validation. I guess this is
121:18 like a really good opportunity to show you an example of like here's where the
121:22 coding assistant messed up. Like I told it not to be as um heavy in the
121:27 validation, but it more followed my structured plan instead of what I told
121:31 it to do. And by the way, the last time I tested this, it did listen, right?
121:34 Right. So, like it's just the unpredictability of LLMs. Like that's
121:37 why it's important for us to be in the loop even when we're using a remote
121:40 setup. And like I could go and do this from my phone as well. So like we this
121:45 still is very possible. [snorts] But yeah, it's even running the playright
121:47 tests here. Like I definitely don't want to do that right now. And so um for
121:54 validation commands, I'm going to say do very basic validation.
122:00 Don't spin up the site or anything. Right. And then like I'm just going to
122:05 delete uh levels three through five here because that's just too much for what we
122:08 want to do right now for our demonstration. And then I'd have to do
122:11 some more prompting here to avoid that issue that we were talking about
122:16 earlier. So yeah, this is maybe like a quick band-aid, but um I mean you get
122:21 the idea of what I'm doing here. So I'm going to remove this testing strategy.
122:25 [snorts] Remove edge cases. There we go. validation commands. Okay, this is
122:33 looking better. Completion checklist. I'm going to uh make sure that there's nothing here for
122:42 the end to end testing because yeah, playright test we don't
122:46 like it's too much right now obviously. Um so let's make sure I'm hitting on
122:52 everything here. Looks good. What is this? This might be too much as well. Oh, no. This is good.
123:01 Okay. Okay. Good. So, I'm going to commit this. And this unfortunately will
123:06 not apply to my current run. And so, I'll have it's going to do the same
123:10 thing again. I can almost guarantee. But that's okay because we'll just try it
123:13 again, right? Like we can we can retry just like we can do when we're working
123:17 with coding assistants locally. And so yeah, as much as it's like a little
123:20 stressful for me when this kind of thing comes up, it's a good opportunity like
123:26 it's a teaching moment in a live stream like how I adapt to this in real time.
123:30 And then just like talking about like what like or just like showing the fact
123:34 like this kind of thing happens when you're working with a new system, even
123:37 an existing one, you [snorts] have these little issues come up that you have to
123:40 address them. And it's never going to be perfect. That doesn't take away from the
123:44 power of this system. Like I already know a workaround. I already know two
123:47 ways that I could address this so it never happens again. And right here I I
123:54 know how to just quickly fix the problem and then restart the session which I
123:59 will do um just once it uh crashes here probably. So I'll let it I'll let it run
124:04 right now and then in the meantime I will go and answer some more questions.
124:10 Lovely thing is we can have AI fix that is that exactly right. I I used the
124:15 markdown editor and I changed it myself, but I literally could have like within
124:20 the same conversation just been like at remote agent um the plan we have here uh
124:28 has too much validation. Go and remove the last three uh layers, right? Like I
124:33 could just like have it do it itself and then boom, kick it off the execute again
124:37 and we'd be good to go. And so let's actually do that now. So it errored. It
124:41 did the exact same thing it did before, exactly like I predicted. But now if I
124:47 send this off here, now it won't because the remote agent is um
124:54 now going to read the updated plan where I took out that those validation steps.
124:59 And so yeah, let's go ahead and send this off. And it's starting a new
125:03 conversation every time, by the way. So you can see that it's starting a new
125:07 session. And that is good because when we start the execute command, I'm
125:12 actually doing that on purpose because the planning command creates the
125:15 structured plan and then the structured plan is the only context that we have to
125:19 send into execute. So we don't actually have to continue the conversation. We
125:25 can keep our context window concise by going into a new execute or a new
125:29 session with our execute and then having it read this document. And that's all it
125:33 needs, right? like reading this document is kind of the prime that it needs to do
125:39 this specific feature implementation. And going back to the logs here,
125:44 we are off to the races. And uh hopefully this time it will not try to
125:50 um spin up a site and have to kill it. Right? Like that's the core of our
125:58 Anyone using Droid heard it has excellent caching to reduce context
126:02 consumption. So, um, little bit of a spoiler. There is a YouTuber named Ray Fernando.
126:11 Some of you guys might have heard of him. He uses Droid as his primary coding
126:17 assistant. And, uh, next month we are going to be doing a live stream
126:20 together. And I I believe Droid is going to be what we will use. Um, just because
126:25 like I mean I love using any AI coding assistant and that's his thing. So,
126:28 we're going to be using it together for the live stream probably. So yeah, stay
126:34 tuned for that. It might be December 6th and it would be the same time as this
126:38 live stream. So it'd be another Saturday at 9:00 am central time because that's
126:42 the time that works for most time zones. Like I try to work with as many time
126:46 zones as I can. But yeah, that's going to be a very exciting live stream. We'll
126:51 be using Droid. So yes, they Droid is fantastic for the way that it compacts
126:56 memory. So you can have like a single conversation with your droid agent
127:00 that's like seven or eight million tokens long and it can still like
127:03 remember things all the way back to like the PRD that you have at the start of
127:08 the conversation. It is just bonkers. It is so so cool. So yeah, I'm excited for
127:14 that live stream. And by the way, the reason that I don't care what AI coding
127:19 assistant I use is all of the principles that I cover in the agentic coding
127:25 course and uh what I cover in uh on my YouTube channel like it's always
127:29 agnostic to the specific tool that I'm using like this remote coding agent.
127:35 Yes, right now it only works with claude code and codeex, but I've designed it in
127:40 a way where we can very easily add other coding assistants in and then the way
127:44 that we load our commands and work with our commands, it's going to be the exact
127:48 same no matter the tool that we use. And that's the beauty of this is
127:53 we never know like what coding assistant is going to be the best next week. Now,
127:57 Claude Code has been kind of the reigning champion ever since April, and
128:02 that is debatable, especially recently with Codeex and um Gemini 3, but like
128:08 generally Claude Code is considered to be the king. That was my really lame
128:13 attempt at drawing a crown there, but you get the idea. But um yeah, like
128:18 maybe next month um there's going to be like GBT 5.1 codeex and then it is just
128:24 like better than claw code. Well, the way that I've created my systems and
128:29 even what I teach in the agentic coding course, like you can apply all of this
128:34 no matter the AI coding assistant that you're using. So, if you don't like
128:39 cloud code because of the rate limits or you just like really love cursor because
128:43 of their 2.0 update, like no matter the tool, you can apply all of this. And so
128:49 yeah, also like please don't think that like just because cloud code is the tool
128:53 I prefer that like none of this would like all of that is still going to be
128:57 valuable no matter the coding assistant that you use. And there are slightly
129:01 different ways of doing things like slash commands for example. Um but like
129:07 all of that is going to work no matter the tool. And so, yeah, like if that
129:10 sounds good to you, like you want to know like here's how I build systems for
129:14 getting reliable and repeatable results with coding assistance, no matter the
129:18 tool that I use, then you should definitely check out Dynamus. We got the
129:22 special sale that is only available during this live stream. So, I'll paste
129:27 that in again right here. If you're interested in joining, I'd really
129:31 appreciate you being a part of Dynamis. It's where I go deeper. All the good
129:33 stuff that I cover in my YouTube channel, I will always keep posting it.
129:37 always giving a ton of free value like I'm doing in this live stream right now,
129:41 but this is just an opportunity to go even deeper and there is a lot more
129:50 And yes, uh Droid does have their own agent infrastructure. It is pretty
129:55 freaking fantastic. Um does anyone have a link to the Droid coding assistant? Uh well, I think you
130:01 can just go to it's called factory.ai. If you just go to factory.ai, AI. Um,
130:06 that's the name of the company where like Droid is their coding agent. So,
130:11 yeah. Didn't mean to droid Jack your live stream. No, you're good. It's a
130:16 fantastic tool. I love talking about other tools and coding assistants.
130:22 So, okay, going back here. Uh, still working on the implementation.
130:26 Hopefully, it will not spin up another website. So, we'll kind of be on the
130:30 lookout here for a kill shell command. But if it happens yet again, then like
130:34 literally all I'm going to do to like absolutely make sure that it doesn't do
130:37 it is just like as a part of the prompt, I will tell it like do not spin up the
130:42 website or I could go and and change the execute GitHub command in real time and
130:46 then I just have to reload the commands and run it again. So again, there's
130:49 always solutions to the little problems here. Like it doesn't faze me that we
130:54 might have to um adapt or like improvise a little bit, right? And like that's
130:57 that's what live streams are all about. You just have to improvise a little bit.
131:03 So, yeah, in the meantime here, while we're waiting for this, I'll go ahead
131:09 and uh pull up some more questions here. [snorts] let's see.
131:18 It's uh a live call that Oh, sorry. I was answering some other question. I'm
131:22 trying to find the the next comment to showcase here. Scott said, "You and
131:26 Ray." Cool. Yeah, it is going to be a good time. I have never actually done a
131:31 live stream with a another YouTuber before, a content creator. Like it's
131:35 always just been me like me and you like it is right now. U that's going to be
131:38 really fun. I've done a couple of YouTube videos with other people um like
131:43 with Raasmus and Sean for some Archon and context engineering stuff, but yeah,
131:48 that's going to be really exciting. Um let's see. And then someone asked, u are you going
131:58 to Hawaii for the Ray Fernando stream? That would be really cool. Uh but no,
132:03 I'm not going to Hawaii. Um Oh, shoot. It failed again. Okay, it
132:07 keeps Okay, I'm [clears throat] g I'm just going to tell it like
132:13 cuz um I don't Well, okay, this time I don't actually know why it did this, but
132:17 here I'm going to do this. And do not maybe I for Oh, I might have
132:21 committed to the wrong branch. Maybe do not start the front end.
132:29 Just do very basic validation with unit tests. Okay, I'm going to kick this off
132:35 again. I don't I don't know. I must have edited the wrong place or there's something
132:40 else in the plan that I had to do. Yeah, unfortunately. Uh but I'll just add this
132:45 in and then we'll we'll be good. I'll say a kill shell crashes the whole
132:52 system. [cough and clears throat] It's kind of funny. I guess I can laugh at it. Well,
132:58 maybe more once the stream is done, but it's all good. So, we'll kick it off
133:02 again. I've never had to do this before, by the way. It's literally Murphy's law.
133:05 Like, when something can go wrong during a live stream, of course it does. But
133:08 yeah, like I said, it's a good lesson just to show how we can improvise here
133:12 because you can work with coding assistants however you want. Like I'm
133:15 just like giving the path of the plan and then I'm just saying like, "Oh,
133:18 here's something else to take note of." Or if I wanted to, I could have
133:22 restarted the process um after I corrected the my prompt for the plan.
133:27 Like I should have just like taken out the playright stuff because I'm not
133:30 using the playright stuff for validation right now. Like that's not the point of
133:34 this system. I don't want it to go that indepth when I'm working on this kind of
133:38 feature, right? And I definitely could and I could just like adjust my prompt, but that's just
133:48 So, yeah. What else do we got here? PID files were created decades ago to
134:01 avoid issues like this. Yes. Yeah. Right. I could build something like that
134:05 as well, like a system where it like is able to track the exact process and run
134:08 the command to like only end that because I think what it's doing is it's
134:14 running something like um let me go back to this issue. It's doing
134:17 something like pkill-f node like it's some more like generic command where
134:22 it's killing itself as well as the website. So, if it had like the actual P
134:27 ID, which it could just search for the P ID, and I wish it would. Um, but like
134:31 yeah, it's doing something more generic like this, and that's what's causing the
134:34 problem here. So, as much as it sucks that it's happening, it actually is like
134:37 really interesting just to like think about when coding assistants run these
134:41 commands, we need to like instruct them on how to be specific and like that
134:47 should be a part of our system as well. But yeah, now it now it will probably
134:56 Oh, this is a good question. [clears throat] What are you looking for
135:00 in coding assistants? What makes them good or bad for you? Okay, that that's
135:05 an awesome question. I appreciate that. So, the main thing that I'm looking in
135:10 for coding assistance is I'm looking for autonomy. And what I mean by autonomy is
135:16 I want them to take in my request for what I want them to build and I want
135:21 them to actually do it end to end. A lot of times what we h what happens with
135:26 coding assistants is they get very lazy where you'll give them like a structured
135:32 plan of I want you to um do some research and then edit this part of the
135:35 codebase and this part and then I want you to do these four things for
135:39 validation, right? like you have a a very structured process and they'll get
135:43 lazy and they'll only implement one thing or they'll only do half of the
135:47 validation and then they'll literally tell you that they're done or they'll
135:51 just say like oh this last part I'll leave for you or something like that and
135:55 that's really annoying and we're seeing that less and less as really a lot of
135:58 different coding assistants are getting more powerful um but you still do see
136:03 that and so the best coding assistants like CL the reason why Claude code
136:08 impressed me so much and I started using it like pretty much exclusively all the
136:13 way back in April is because it was and still to my in my mind is the best
136:18 coding assistant for actually taking a request and handling it end to end. Now
136:23 of course a lot of of how well it does that depends on your system as well like
136:26 your instructions and your rules and in your commands but as just like the base
136:32 harness around the claw model cloud code is very impress impressive to me for how
136:37 it can handle that kind of thing. And then obviously um Oh, okay. It actually worked here.
136:47 [laughter] Finally. Finally. Okay, good. So, I didn't do the the stupid extra
136:51 validation now. But anyway, I'll keep answering your question really quickly.
136:54 Then then we got like the final big payoff here. Let's go. But yeah, the
136:57 last thing I'll say here is the other thing I appreciate in coding assistance
137:02 um besides the obvious of like it just works the best for the code is just the
137:07 features that are built in. Cloud code uh was and still is very ahead of its
137:10 time. Like it's the first to implement slash commands and sub aents. Now we've
137:15 got the whole idea of skills that work directly with cloud code and like that's
137:19 really important for saving context because of all the issues with MCP
137:23 servers taking thousands and thousands of tokens of context just being loaded
137:27 up front. Um and so like yeah, just having access to the latest cutting edge
137:30 features is pretty important to me as well. Um so yeah, good question. I hope
137:35 that those are helpful for you. But uh yeah, anyway, we finally have our
137:41 our big payoff here. So, we're operating in the same feature branch where we
137:45 created our plan and now we have our pull request. Let's go. So, it is from
137:51 this branch into staging because we are going to create basically an artifact
137:56 for us to review by uh once we merge this pull request, it's going to update
138:01 this staging URL for our website. Now before we do that obviously we can go
138:04 and look here and we can take a look at what was changed. So I can review its
138:08 summary of the poll request that it also created autonomously. And so remember
138:12 when we set up the remote agentic coding system we used the GitHub CLI like we're
138:17 integrating the GitHub CLI directly in to our agent. So it has the ability to
138:21 do things like create a pull request. The cloud code actually generated this
138:25 pull request here as a part of a work that it did. like one of these bash
138:29 calls towards the bottom of our logs. Is it making that pull request with the
138:34 GitHub CLI? I can look at the files that were changed. Um like it created a
138:39 utility here for exporting the conversation as markdown. We have the
138:45 chat layout um so that we can add in the um conversation title. Looks like it
138:49 needs to include that. And then we have the function to do the exporting um that
138:55 I already showed that. And then we have the chat input. So, we're adding
138:57 something to the bottom here where it's going to be that button we can click to
139:01 export the conversation. And so, now to really like validate this
139:05 as a user, that is where I can merge this pull request because things are
139:10 looking generally good to me. And so now what I can do, I'll actually show you
139:13 this in Telegram. I think it'll be cool just to like show you that I can be
139:17 using an entirely different application to continue this process here. So I can
139:24 say I just created a PR going into the staging branch, find that recent PR and then merge it,
139:34 which obviously I can merge it myself, but I'm just showing you how we can have
139:38 the AI coding assistant like handle this completely end to end for us. And I
139:42 could have like sent off this request from my phone as well. And so yeah, so
139:46 it's using the GitHub CLI just like it did in GitHub. It found PR number 10 and
139:51 now it's running the merge like boom boom boom we are good to go. Cool. So it
139:55 has successfully merged the pull request in. So if we go back to our codebase
140:01 here it has been flagged as merged now. Very cool. And then if we go into render
140:05 we can see that it is automatically started it has automatically started the
140:10 new deployment. So as of 11:20 a.m. which central time that is right now.
140:14 Very cool. And so within render, if I just show you the settings really quick,
140:18 I just have it set up for that anytime there is a commit that is made to the
140:23 staging branch, that is when it's going to automatically build and deploy this
140:28 application for me. And so I don't have to like build or like copy over files to
140:33 a server or anything. Like if you're a more old school engineer, you probably
140:37 remember doing that with like FTP or something like this is all like I make
140:40 the commit or rather the coding assistant makes the commit by making the
140:44 merge for me and then it automatically deploys things. Very very cool. So I'm going to refresh
140:52 here and um yep. So it's in the middle of deploying. So I just have to wait a
140:55 second for the static application and render. It should only take like a
140:58 minute or two. Like it's going to be All right. Hi, your content is very
141:06 valuable. Can you please leave this up for an hour or two? I missed the start.
141:11 Yeah. So, okay. For those of you who are joining the live stream like now or or a
141:16 bit later, I will leave the the link up for the Black Friday sale for like a
141:22 little bit after the live stream just so it's not like immediately once it's
141:25 done. If you join at the very end, you just like totally lose the opportunity.
141:28 But, it's only going to be for like an hour or something and then it's coming.
141:31 It's going down. So, if you are interested in joining Dynamus and taking
141:35 advantage of this sale, I'm not going to have something like this again. 40% off
141:40 the annual, 25% off the monthly with the celebration of completing the entire
141:44 Agentic Coding course, then yeah, definitely get in on this now. So, I
141:47 appreciate you asking because yeah, I do definitely want to make sure that um you
141:51 don't join at like the very end and then just like feel like you lost the
141:55 opportunity. So, yeah. Also, someone said it's hard to hear. I'll move the microphone a bit
142:00 closer to my face. Sorry about that. It's also just hard for me to talk as
142:04 loud right now. Um, but yeah, I'll try to to speak up a bit. So, all right.
142:10 There. I hope that is better. All right. Any video any chance you're doing a
142:17 video on a DSPY? So, I am actually planning on doing one. Yeah. Um, so stay tuned for that. That
142:25 might be next month. We'll see. I'm still working on planning out my content
142:29 calendar right now, but I I do have that on my radar 100%. So, okay, let's go back here. All right,
142:36 so we are now deployed. So, this is the moment of truth. No promises that this is perfect like this
142:43 is a true live demo where I have never built the markdown export feature into
142:48 this application ever before. Um, so we'll see if it works. So, there we go.
142:53 I refresh the page and now we have the button to export a conversation. Very
142:58 cool. So, I'm going to select one here and then I'll click on export as
143:03 markdown. And there we go. Okay. Did it work? Oh, yes. There we go. Nice. We
143:10 just one-shotted it. I was actually like a little bit nervous. Not because I
143:13 don't trust my system, but because of Murphy's law with live streams, right?
143:17 But all right. Our system just oneshotted this new feature for
143:21 application. And the important part is, well, first of all, it's just freaking
143:23 awesome that this worked. We got the markdown export of our conversation now.
143:29 But the most important thing is is that this is not in our main branch yet. And
143:34 so if there was anything that went wrong with this implementation, it's not
143:38 deployed to production. We have the opportunity to come back here and within
143:42 the same conversation, we can say, you know, like at remote- agent, this failed
143:49 miserably. You need to fix xyz. like we can continue to iterate with our remote
143:53 agentic coding system just like we could do with our coding assistant that we're
143:56 working with in our local development environment. And so in this case, it's
144:01 like actually perfect. Like you can see when I do a new chat, it's grayed out.
144:05 Like I can't actually like try to click it when there's no conversation. And
144:08 then I immediately can I can go back to old conversations before I even
144:11 implemented this feature. I have the little um toast in the bottom right. I
144:15 know my face is mostly covering it, but the conversation exported like this is
144:20 freaking awesome. And so now what I can do is I can merge this into the main
144:25 branch. So I can update the primary version of the website as well. Like you
144:29 can see that if I go to chat.dynamus.ai, we don't have that button yet because we
144:33 have only updated the staging environment. This is this is beautiful.
144:36 like the human in the loop here is just so powerful because even when we trust
144:40 our system, we still want to be able to validate any kind of artifact and maybe
144:44 the code as well before we go into production. And we can do that. And so
144:48 let's let's go ahead and do that. So I'm going to go to the Telegram here again
144:54 because of course I want the coding assistant to take care of this for me.
144:58 So I'm going to say okay, I merged or okay, things are looking good. Now
145:05 create a PR from staging into main. I want to move this into production, make
145:11 a simple PR description, and then actually merge it as well. Now, I could
145:15 like review the poll request before I merge or whatever, but like I know
145:19 exactly what I'm bringing from staging into main. So, I'm just going to have it
145:23 do this entire thing for me here. So, it's going to fetch the origin. It's
145:28 going to pull the diff here because it's going to use this as context to create a
145:32 good pull request description for me. Like, it it's handling everything. There
145:35 we go. So creating the pull request, we can go back to GitHub and we can see
145:40 this in real time. So let me scroll all the way back up to the top and go to
145:45 pull requests. And there we go. We have staging going into main. And we can see
145:49 the notification from Telegram. All this is happening in real time. It created
145:53 the pull request and then immediately merge it into main just like I asked it
145:58 to do. Perfect. And so now going back to render immediately we can see that all
146:03 of our services in our production environment are now redeploying. And so
146:07 just like with staging we don't have to do any work now. We validated things but
146:11 now we leave it up to the coding assistant and the automation and render
146:15 to do our complete deployment. And then we can just go to chat.dynamus.ai after
146:20 and validate everything once we are deployed. So, I just have to wait a little bit
146:27 here because the rag pipeline and agent API, uh, they take a few minutes to
146:31 deploy because they're obviously bigger services than a static front end. Um,
146:35 that's part of the reason why I didn't include them in the staging environment
146:39 here. Uh, but yeah, the agent API will will be the one that takes a couple of
146:43 minutes here. Cool. Raasm is out for today. Great stream. Yep. See you, Dynamus. Raasmus,
146:48 I appreciate it. Cool. Audio is crystal clear. Uh, turn way down. Maybe turn up the volume on
146:54 your system. Oh, okay. Interesting. Yeah, that's always a tough thing with
146:57 streams and YouTube videos is different people have their audio at different
147:00 levels. So, it kind of has to be like you just have to, you know, adjust it
147:08 When there's a problem, fix the system, not just the prompt. Yes, that is very
147:13 true. Actually, yet another reason it's good that I kind of had this um teaching
147:16 moment here. So, let me go back to the issue. So, one of the core things that I cover
147:22 in the agent coding course and this this is like actually crucial like please
147:26 lean into this like this is important when we encounter an issue in our
147:32 process like we did here a very real problem unfortunately that we ran into.
147:38 Sure, I can go and just put a band-aid on it like I did here. And by the way, I
147:41 did this just because I want to be brief for our demonstration here. But when you
147:45 encounter a problem like this, instead of doing a one-off fix, take this as an
147:52 opportunity to evolve your system. And so let me show the diagram for this very
147:57 quickly. So we go through a pivot loop here. This is the process that we did in
148:01 GitHub. We did our vibe planning. We created our structure plan and then we
148:05 passed the structure plan to the implementation which also did a simple
148:10 version of the validation for us. So this is the loop that we go through
148:14 whenever we want to implement a new feature. But after our piv loop is done, we have
148:21 a decision point here. So let me zoom in on this. If we encounter no major issues
148:26 through the piv loop, right? Like the implementation went pretty well. Maybe
148:29 there was like a couple of issues that we had to address, but overall like
148:33 things went really well and there's no issues that we saw come up as a pattern.
148:38 If that's the case, then we can just go directly onto the next pivot loop to
148:42 implement our next feature. Like I could go and make another change to our front
148:47 end because things worked pretty well. But then we have this other side of the
148:51 coin. If we did encounter a major problem with our pivot loop, and I guess
148:55 you could technically say that we encountered a a repeated problem, we had
148:59 a pattern that developed where it's like, okay, something should clearly be
149:04 changed in our system. So now what we do here is instead of just going back to
149:09 implementing the next feature after we apply our band-aid, instead we're going
149:13 to do some retrospection, we're going to go back to our system and think about
149:18 what could we have updated so that this problem would not happen again. So I'm
149:22 actually like really glad that this issue did happen because it's a good
149:25 opportunity to talk about this with a real example. And so we're kind of doing
149:30 a metal loop here, right? Like when we first set up our project, this is where
149:35 we create our PRD. And I go over all of this in the Agenta coding course. By the
149:39 way, I even like call out the specific modules here where I cover these
149:43 different things. So we create our PRD. We define our global rules and our on
149:46 demand context. We create all of our commands like the ones that we've seen
149:50 today for priming, planning, executing. Similar thing for brownfield development
149:54 like if we're applying AI to an existing codebase. So we set up our project. This
150:00 is kind of like our AI layer I like to call it. And then we go into our pivot
150:04 loops. And then we kind of have this like smaller loop here. Excuse me. We
150:09 have this smaller loop here where after we do one feature implementation, we can
150:13 just like go on to the next. But if we find an opportunity to evolve our
150:18 system, then basically what we do is we look back at what we created for our
150:22 project initially. like okay maybe there's something I need to change in my
150:26 commands or maybe I need to be very clear about something in my global rules
150:30 or like add something to my global rules and so you use this as an opportunity to
150:35 address the problem in a way where it is not going to happen again that is the
150:40 the high lever skill here of not just going back and doing another loop
150:42 because you might encounter that problem again like I'd probably encounter the
150:46 same problem where it like spins up the website and then it kills the shell and
150:50 then that ends cloud code as well. So instead, I'm going to go through this
150:53 like kind of metal loop here where I zoom out, I improve something like my
150:58 command, then I go back into my next pivot loop. That is so important because
151:03 like what I would do for the issue that we ran into specifically today is I
151:10 would probably update my execute command and I would say like when you are
151:14 spinning up a website and you have to like kill the docker container or kill
151:19 the node process at the end do it in this way so that you don't end the cloud
151:25 code um agent SDK session as well. Now, exactly what this way is, I don't know.
151:30 I'd have to do some research, but that's where you can use your AI coding
151:33 assistant to also help you do some research. Like, I can say, "Hey, Claude,
151:37 code, you killed the shell in a way where you also killed yourself." Like,
151:43 please help me do some research here and find like what is the right kill command
151:48 that I can use so that I can target a specific PD, for example, um, like we
151:52 talked about earlier. And so that way you make sure that you're not killing
151:55 yourself when you're ending the process that you spun up for validation. So we
151:59 can include that as just like a little addition to our execute command or maybe
152:02 we include it as a part of our structured plan. So we are evolving our
152:07 system so we don't encounter that issue again. And that is so so powerful
152:12 because you start with this base system where you have these problems come up
152:16 like I did but then you address them one at a time. You do a couple of piv loops.
152:20 you figure out this thing keeps happening so you address it and then
152:24 maybe something else comes up when you do the next set of piv loops but that's
152:28 okay because it's an opportunity you treat every problem as an opportunity so
152:31 eventually you get to the point where like you've evolved your system to
152:35 address all these things so you just go to that next level of how much you can
152:40 trust what you've created and that's when you really get to the point with
152:44 coding assistance where you have this insane level of power at your disposal
152:48 and that's what I feel like I have Like let me just tell you this entire process
152:53 that I used in the agentic coding course that I cover for you in the course. I
153:00 use this exact process to a to build out the remote agentic coding system. And so
153:07 like literally like if you join Dynamus and you do and you go to the GitHub
153:10 repository like let me go to this right now. If I go to the Dynamis community
153:14 you can see like all the repos that we have as resources for the community. If
153:18 I go to the remote coding agent, you can literally go through the 41 commits that
153:23 I've done for this project and you can see like all the structure plans. You
153:27 can see all of the commands that I built in. You can see each of the individual
153:31 piv loops that I've gone through and all the validation that I've done. Like all
153:35 of this is like using the course exactly. So I'm I'm drinking from my own
153:40 or what what's the phrase? I'm I'm dog feeding. I'm forgetting I'm I'm brain
153:44 farting on the phrase, but like exactly what I teach is exactly what I use to
153:48 build this and I was able to build this entire thing in like a very short amount
153:55 of time because the process was just so reliable and any problems that I ran
153:59 into in the first few piv loops that I did, I just evolved my system so that
154:03 the validation was better or the structured plan was better, like
154:06 whatever it took to make it so that that thing didn't happen again. And that's
154:10 why I'm so confident that when I go back here and I add in Slack or I add in Open
154:16 Code like any more platforms or coding assistants, I know that it's going to
154:21 work so well because any issues that I encountered when I built in the Telegram
154:26 integration, I evolved my system so the GitHub integration went like two times
154:30 better. And so then now I know that any next integration is also going to go
154:34 that much better. And it's the same thing for building and coding assistance
154:39 as well. And so that's like kind of like one of the main things that I cover in
154:44 the agentic coding course is that every time you encounter an issue in your
154:48 code. Don't just fix the bug. I mean obviously you have to fix the bug but
154:53 also go back and address the problem in your system if it's something that's
154:57 come up as a pattern like especially if you encounter this issue you know three
155:01 times or more. It's like, okay, yeah, please take a step back, think about how
155:06 you can address the system so that thing doesn't happen again. So, if that sounds
155:09 good to you, if you want to have a system that you get to evolve over time
155:14 and just make bulletproof, like please check out this page here. I'll link it
155:17 in the chat again. An opportunity that you are not going to have again for the
155:23 discount joining Dynamis celebrating the completion. This agentic coding course
155:28 is fully complete. Of course, aside from more improvements that I want to make to
155:32 the course, Raasmus and I, Raasmus, by the way, uh, helped me make this course.
155:35 Um, huge shout out to him. I know that he left the stream earlier, but yeah, we
155:39 put so much effort into building this course. And even though like it's
155:43 released and it's done in the sense of like the full core modules are released,
155:48 we're going to keep improving things. We're going to keep building out the
155:51 remote agent coding system. Like I've been saying, we're going to keep
155:54 building out the Obsidian AI agent, which is like the primary use case for
155:59 this course. And just whenever there's new best practices for AI coding, we're
156:04 going to make sure that we incorporate them in this course. Like let me tell
156:08 you, just going to the MCP module right here in module 10, um, I actually
156:13 released this module slightly late because I knew that like, okay, I have
156:18 to talk about skills. Like when I first planned this module like way back in
156:23 August, skills weren't even a thing. So this module is actually only going to be
156:27 about MCP servers, but then I adjusted it like basically in real time to
156:32 incorporate skills as well because I always want to make sure that like
156:35 you're on the cutting edge with the content that I have for you. I don't
156:38 want you to be like slightly behind or anything. And so that's a commitment
156:41 that I have to you is like I'm going to work really hard to keep this stuff up
156:44 to date. I mean, like, obviously if skills came out tomorrow, I wouldn't
156:48 like have skills incorporated the next day, but I mean, there's going to be
156:52 like a little bit of a delay, but it's going to be like pretty crazy like how
156:56 quick I get those things out there for you. So, yeah, hope that sounds good to you. So,
157:03 all right. Well, [clears throat] anyway, going back to render here, everything is
157:08 deployed to our main branch. And so now if I refresh, this is the the last
157:12 moment of truth here because we have things deployed in production. I refresh
157:17 and boom, there we go. We have our export button. So if I go to a
157:20 conversation, I mean, none of this should like totally shock you now
157:23 because it's exactly what we did in our staging branch, but now we have it
157:27 deployed to production after we did our validation. So very, very cool.
157:35 All right. So, that's pretty much all I wanted to do for the live demo here.
157:40 Now, the the other thing is like there's a lot more that we can do as well. Like
157:44 for example, I could run a code review on the poll request. Like there's a lot
157:48 more commands that I have in my system. Um things that I cover in the agentic
157:52 coding course like a code review or a system review that actually helps you
157:56 with the system evolution that I was talking about earlier. So, there's a lot
158:01 more things that I could get into, but I think what I want to do to wrap up the
158:05 live stream here is just address any questions that you guys have and then
158:09 maybe talk a little bit more about the architecture that I have for this
158:13 system. Um, because yeah, at this point, like we covered how to set things up,
158:17 how it works, why it's important over other systems that you already have like
158:21 factory or cloud code for the web. like the power that you have in a custom
158:25 system where you really have control over the commands that you inject and
158:29 things like MCP like that's a big deal and we went through the full demo. Um,
158:33 so yeah, like I I think like that's how I want to wrap things up here is answer
158:36 you guys' questions and then cover some of the architecture as well. So what I'm
158:41 going to do is I'm going to take a quick bathroom break here uh because I yeah I
158:47 need to like refill on water as well. So, I'm going to mute myself and I'm
158:52 going to turn off my camera just for a couple of minutes here and then I will
158:56 be back to talk to you guys and go through some questions here. So, I'm
159:00 going to say be right back. I'm just going to like tack this in really
159:05 cheekily right here. Uh, there we go. Be right back. So, I will mute and I will
161:18 All right, we are back. Actually, I forgot to drink water. That's all good.
161:24 There we go. Got to make sure I keep [clears throat] doing that. But yeah,
161:27 let me delete this. Cool. All right, so yeah, I'll go over the architecture in a
161:31 little bit. I' I've talked about it a good amount already. But uh yeah, I want to uh just like go
161:37 through some of the questions that we got here as well. Got to go. Thanks,
161:41 Cole, for all your time and effort. You're welcome. See you all in Dynamus.
161:46 That is right. Cool. Uh isn't there a way to blacklist the kill shell tool in code/config if it's
161:53 known to break things? Yes, Tony. So, you can set up granular permissions if
161:58 you want for cloud code, even the agents SDK. The problem though is I can't just
162:05 blacklist it and have that as a solution because it's it might get confused and
162:10 then uh just never tear down the website that it spins up and so then I kind of
162:13 have a memory leak because it's going to turn it's going to spin up a process
162:17 every time it validates something and then never tear it down within my
162:21 container environment. So that's why it's better to like address the system
162:25 directly and either tell it to like not do that validation just for my remote
162:29 system and like do other kinds of ways of checking or just tell it to like use
162:33 a different command. Um but you could do both. Like you could like blacklist the
162:37 kill shell and then have it just like use bash with a different um kill where
162:43 it like uses a specific PD for example. So it's like a two-step process of list
162:47 out the running applications and then grab the P and kill that. So yeah,
162:51 there's definitely a few different ways to tackle the problem. That's why I'm
162:53 not worried about it. It's like, oh, I I could easily evolve my system. I can
162:57 even like make a correction in the remote agent coding system itself, which
163:01 definitely we'll be looking into because like I said, I want to keep evolving
163:05 these things for you. So yeah, Sean just joined with the yearly sub.
163:10 Amazing. Sean, welcome to Dynamis, my friend. Appreciate it. This was a great
163:14 show. See you next time. Yeah, I appreciate it very much. You guys are
163:18 all very encouraging. I appreciate it a lot. Uh, this will have a discount too.
163:23 I don't know if they meant to like send something else there, but um, yes, the
163:30 we will have a discount for right now in the live stream is a discount that you
163:33 will not get again. And then we have the the normal Black Friday one that goes
163:37 through Cyber Monday. That's almost as good. So, if you watch the recording and
163:40 you miss the discount like in the live stream, there's still an insane deal
163:49 By the way, you're amazing for talking three hours straight. Yeah, I appreciate
163:52 even with a horse voice, I managed. It has not gotten worse really through the
163:56 uh live stream. In fact, it's only gotten a little bit better. So, that's
164:00 pretty cool. I'm glad to hear that. I'm glad that it's uh yeah, I mean, it is a
164:04 lot of talking. Like, I probably will lose my voice tomorrow, but I'm not
164:08 doing any talking tomorrow. I I think I will be taking tomorrow off.
164:13 Um yeah, just to like take some time to rest. Uh, I took most of yesterday off
164:21 and then Thanksgiving as well, which is like the first time I've taken like
164:27 about 72 hours off in a week um since like 3 years ago cuz even with like
164:32 vacations [snorts] that I've done this year, I'm just like and there's like so
164:35 much to do in Dynamus that like I'm still in the community every day because
164:39 that's my commitment to you guys, right? And like that's part of the the value
164:42 that I provide is just like constantly being there to make sure that like we got exciting
164:48 things going on and just to make sure that I'm facilitating like awesome
164:52 discussions. Um but yeah, also there are a lot of people in Dynamis and if if you
164:58 if you know you know like if you're one of those people like thank you for being
165:02 in Dynamis like every single day and creating really engaging discussions as
165:06 well. All right, you're the man Cole. Thank you so much. See you in Dynamus. I
165:11 appreciate it, Sean. See you in the community, man. Uh referring to a cup bearer who drinks
165:21 from the cup of the king to prove it's not poison. Right. Yeah. And Sean said
165:25 eating your own dog food. I mean that basic is what I'm doing as well. But
165:29 yeah, going back to that this analogy that's kind of true, right? Like well I
165:33 mean first of all what I teach in the agentic coding course obviously goes
165:37 like way far back as far as like the things that I've incorporated for myself
165:41 and research myself. But like yes, doing the remote agent coding system and
165:46 building it following the exact commands that I give as a resource in the
165:50 community and the course and like the exact same rules like that definitely is
165:55 like I wanted to do that to prove to myself and to be able to tell you guys
166:00 that like there aren't like little additions that I did to simplify things
166:03 for the course or anything that like affect the process. like I actually used
166:09 the exact same everything to build out this application that like yeah it's not
166:14 going to be like some Fortune 500 super complicated codebase but it is still
166:18 pretty detailed and like there's a lot of thought I put into the architecture
166:22 when I created the initial PRD. Maybe this is like a good time to dive into
166:26 this really quickly. Um, but yeah, like I built all of this to like prove that
166:30 like like [snorts] I I can say with full confidence if you go through the exact
166:36 flow that I lay out in the Aenta coding system and you build out your coding
166:42 system or sorry, let me let me back up a sec. If you follow the exact flow that I
166:47 go through in the Aenta coding course and you build your own system using
166:51 these commands and rules and just like the principles behind it all, like I can
166:54 just absolutely guarantee that you'll be blown away by the results that you get
166:58 using a coding assistant. Like for all the people that say that uh like AI
167:04 coding is not reliable, it's like yes, that's true, but like try having a
167:09 system like this. And when [snorts] you put yourself in the loop in the right
167:12 ways, it's okay when the coding assistant messes up because like I said,
167:16 we have our guardrails. We have our validation. We do our system evolution.
167:21 So, we address these things. We don't ignore it. We don't act like coding
167:25 assistants are always perfect, but we take their imperfections as an
167:30 opportunity to avoid them going forward. And yeah, it's just so cool the kinds of
167:33 things that that we are able to build that I'm able to build like this is just
167:38 proving it. Cool. Locked in for a year. Let's build. Awesome. Welcome to Dynamus. I
167:45 appreciate it a lot. Cool. Cool. I was building a website for
167:49 someone and that cup bearer in the book of uh dial was right in front of me.
167:52 Nice. [gasps] Nice. That's awesome. Not able to purchase a course. No
168:00 payment option for PayPal. Yeah. Well, I thought about integrating PayPal, but um
168:05 yeah, I mean there you can do link as well is something that's integrated into
168:08 Stripe, so you don't have to like directly give your credit card info. I
168:12 hope that helps. Um but yeah, I don't have a PayPal integration at this time.
168:16 [cough] Excuse me. Um can you please post the diagrams that you showcased?
168:24 Yeah, so I don't really know exactly where I would post them. Maybe I well
168:29 yeah because even the agentic the remote coding system repo will be torn down at
168:32 the end of the stream. But I will say that these are all in Dynamus. So this
168:37 one right here with the architecture as well as my full course guide. This is a
168:42 part of the Dynamus agent coding course. So if you join the community and you go
168:45 through the course like these are resources and I actually built this one
168:49 specifically as a way for you to have something to look back on constantly. So as you are
168:55 going through the course and building out your own system, um you can refer
169:00 back to this, [snorts] excuse me, so that you can think about
169:04 for yourself like, okay, I'm currently in this part of my process or I'm trying
169:09 to work on an existing messy codebase. I'm at this part of my process. And so
169:12 you can know like where you're currently at, what's coming next, what module in
169:17 the course I talk about that thing. And so it's kind of your um guide or your
169:21 map to go through the course and then just kind of have like all of the core
169:25 takeaways available to you as well. Obviously within the course itself
169:28 though diving a lot deeper into [snorts] how we do all these things like how do
169:32 we create our product requirement document that guides our MVP for example
169:37 and then having the course progression laid out with all the different modules
169:41 that we have as well. uh which maybe I can talk about this really quickly like
169:45 module one is kind of like a baseline exercise just introducing you to why we
169:50 need a system for AI coding and then module two is when I introduce the idea
169:54 of the piv loop which is the kind of like core process that I've talked about
169:58 a lot during the stream at this point it's just mental models right and then
170:02 we get into building our system components like our global rules and our
170:06 commands these are really the things in our on demand context as well and I
170:09 explain what all that is in the course but like these are the things that are
170:12 really the foundation of any system that we have, any process that we guide our
170:17 coding assistant through. And I talk about how it applies both to starting
170:20 new projects and working on existing ones. And then modules 5 through 7 are
170:27 like the big core modules. And this is kind of where like everything starts to
170:30 come together with the piv loop and our global rules and creating our PRD and
170:34 getting into our commands and how we tie that together to create our full system.
170:39 going through the piv loops and even getting into the ideas of um system
170:44 review and evolving our system in module 7. And then modules 8 through 12 are
170:48 kind of more of like the advanced modules where we get into the remote
170:52 agentic coding. So like module 9 is where I get really deep into the system
170:55 that I'm giving away to you guys in the live stream today. And then also
170:59 thinking about how we can use GitHub actions as another way to automate
171:02 things in GitHub repos and like building coding assistance into our CI/CD for
171:08 example. And then modules 10 through 11 is just thinking about like more ways to
171:12 give more power and tooling to our coding assistants like MCP servers and
171:17 skills. Um sub agents, how we can use sub agents and different coding
171:20 assistants and then also how we can um kick off different coding agents in
171:25 parallel working on the same codebase or different code bases at the same time
171:29 and strategies for that like uh git work trees for example. So yeah, and then I
171:33 got the advanced um kind of like bonus resources as well going into the
171:37 different strategies for context engineering uh like the BMAD method, PRP
171:42 and GitHub spec kit talking about how we can like try out these strategies and
171:46 not just like use it off the shelf because we can always outperform those
171:52 strategies when we um use them or like when we build our own system. But it's
171:56 still it's still great to like have it as an opportunity to learn how do these
172:01 systems work. Let me go back to the course here. It's like when I cover
172:04 these ones, it's all about like how can we learn the from these strategies to
172:09 evolve our own systems like finding the parts from these that we appreciate and
172:12 then building it into our own process like updating our plan command or maybe
172:16 adding a new command that we have as a part of our piv loop. Whatever that
172:19 might be for you, whatever you want to optimize for your system and the
172:23 projects that you are working on. That's when the true power is unlocked. So
172:28 yeah, another one that another time I just want to like plug really quick like
172:31 if if this sounds good to you and you want to go through and learn what it
172:34 takes to build reliable and repeatable systems for AI coding, come take
172:39 advantage of this this opportunity like this live stream is going to be done
172:42 pretty soon here. I'm going to wrap it up and then very soon after this this is
172:47 going away and we'll still have the regular Black Friday Cyber Monday sale,
172:51 but like this is a special discount that will only apply to the live stream right
172:54 now. So, I'd love to have you in the community. [snorts] All right,
173:00 cool. Cool. Now, I can really start to feel my voice going a little bit, but
173:09 All right. Hey, Chris. Uh, do we have access to this agent in Dynamus or should I clone
173:15 it now? If so, then share the link one more time, please. [clears throat] Yeah,
173:19 so I will share the link right now. Um, but Chris, this is available in
173:24 Dynamus. So, I cover this all in module nine of the agent coding course and I
173:29 also have a link to the repo for the version of this that I'm going to keep
173:32 evolving in the community. So, for those of you who are curious and just um
173:35 because you asked Chris, I'll put the link to the public one that this is the
173:39 one that's going to be torn down um within like an hour of the live stream
173:43 ending. But um yeah, you can clone this one if you want. That right now what I'm
173:49 showing right here is the exact same as the private one for the Agentic coding
173:52 course because I'm true to my word. Like I am giving away the full thing for free
173:58 for this live stream, but just over time the one that we have in Dynamis will
174:03 become more evolved. Hope that makes sense. But yeah, thanks Sean for helping out
174:13 All right. Oops, didn't mean to show that. Just finished implementing my Jira
174:18 adapter. Got the first comment in Jira via my local Docker. Wow, that is so
174:24 cool, Peter. Because uh Jira is one of the ones that I was thinking about
174:28 integrating. Did you um so did you like you're you're saying
174:33 like you actually forked the this system or you're saying it's something else
174:35 that you built because if you built this directly into the remote agent coding
174:39 system, that is so cool and that just like proves how simple it is. I mean,
174:43 I'm sure you're a really smart guy, Peter, but still being able to like do
174:46 that during the live stream is amazing. Um so that's very cool. Uh, and then you
174:51 also asked if the if it's torn down, will the forks remain? Yep. So, your
174:54 your fork will still be there. You won't lose the fork. At least I'm pretty sure.
174:59 Um, that's actually Well, maybe I should be very sure on that. Let me actually just
175:08 ask claw desktop quick. If I make a repo private that was public, do people still
175:15 keep their forks? I cuz I think like once you fork it, it is like your own
175:19 private reposi or well I guess it has to be public but it's your own repository.
175:22 [snorts] U yeah okay this is the caveat. So you still keep the fork but it becomes
175:28 disconnected. So it's not going to tell you that like your fork is now five
175:34 commits behind or 10 commits ahead of the main one. It becomes your own
175:38 standalone copy. So yeah, there's no issue for you once I u make it private.
175:43 Like you don't lose what you have. Um which yeah, you building a Jira adapter
175:47 is so freaking cool. Like I really hope you won't lose what you have. So yeah,
175:50 you'll be good. And then for any anyone even if you're not making changes, but
175:53 you just like want to have the repo still, you will because that is um the
175:58 luxury that you have being a part of the stream here is uh you get this like I'm
176:01 giving it to you. It's not something that I'll keep evolving like the one in
176:05 Dynamis, but you have this to now to evolve however you want and use right
176:09 now. Uh Tony said, "When you delete a public repo, the oldest active public fork is
176:15 chosen to be the new upstream repo." Interesting. That's not what Claw
176:19 Desktop told me, but maybe I don't know. Maybe it maybe it wasn't quite right.
176:22 So, I don't think I'm going to delete the repo, but I'll probably just make it
176:26 private or stash it or archive it and make it private. I don't know. I'll
176:34 During this live stream, I finished my terminalbased AI agent leveraging Olama
176:39 and add an additional feature all using my phone. Let's go, Sean. That's so
176:49 Aqua voice doesn't transcribe horse. I if you mean like a horse voice.
176:52 Actually, Aqua Voice, I've been using it the last couple of days, just as I've
176:56 been prepping for the live stream here, and it it actually understands me almost
177:03 as well. Um, the funny thing with Aqua Voice is, um, let me actually show you
177:07 really quick. So, for those of you guys who don't know, I talk about this in the
177:11 Agenta coding course as well, but I would highly highly recommend using some
177:16 kind of speechto text tool to talk to your coding assistant because you can talk like two to three
177:25 times faster than you can type. And a lot of times when we're trying to
177:30 correct our coding assistant or we um just want to like describe in a lot of
177:35 detail a feature we wanted to build, we are much more inclined to be very
177:39 thorough in our explanations if [snorts] we don't have that barrier of having to
177:44 type out every single word. And so I I've found my results with coding
177:49 assistants actually get better when I use a speechtoext tool because I'm more
177:53 encouraged to just be thorough. Like literally that's all it is. And so I use
177:57 these all of the time. You can see here that with my Aqua Voice, I started my
178:02 subscription at the end of April. I have spoken almost 400,000 words into Aqua
178:10 Voice. And my words per minute is 230. You cannot tell me that you can type
178:16 that fast. There is no way that you can. And and you also have to consider that
178:22 the 230 words per minute is also incorporating the fact that like
178:26 sometimes I have pauses in my talking just as I'm thinking. Like there might
178:30 be 30 seconds when I have Aqua Voice like live transcribing but I'm not
178:33 actually saying something. And even then it's 230 words per minute. It's crazy.
178:39 So yeah, Aqua is fantastic. It's only $10 a month. I'm not affiliated with
178:44 them or anything. If you want a free and kind of like open source version, you
178:49 could also look at um Epic Center Whispering. Um so yeah, that's the wrong
178:54 one. Sorry, I meant to go to the GitHub repo here. So Epicenter whispering is um
178:58 something kind of like Aqua Voice, but it's free and open source. So yeah, I'm
179:03 not saying you have to pay for Aqua or you have to use a um commercial tool,
179:09 but like just get something, please. It is so amazing having a speechto text
179:15 tool for coding. So yeah, [clears throat] uh Whisper Flow. Yeah, Whisper Flow is
179:21 also fantastic. Yep, there's a lot of good ones out there. Um and then Mac has
179:26 its own builtin if you if you have a Mac. I think that might actually be
179:30 Whisper Flow, but I don't use a Mac very frequently at all. So, um, yeah, excuse
179:35 my ignorance there, but yeah, there's a Uh, okay, this is a good question. I
179:43 don't have a formal CS or AI background, but I want to build my own AI agents and
179:46 systems like this. If you were starting today with no degree, what would you
179:51 learn first? Yeah. Okay, very, very good question. So, uh, here's what I would
179:55 say. Even with AI coding assistants, you still want to understand the
180:03 fundamentals of programming if you want to do anything technical like if you
180:06 want to like build any kind of software. So I'm not going to be that person that
180:10 tells you to just learn how to use a coding assistant and then don't learn
180:14 anything with like Python or TypeScript. I would recommend starting with the
180:18 basics of Python. making sure that you understand just generally goes that what
180:22 goes into building a program, understanding the logic of you know like
180:26 loops and classes and functions and arguments and things like that. But you
180:30 don't you definitely don't have to go as deep into like understanding syntax and
180:36 memorizing syntax like you did before because that's the you know grunt work
180:40 that you want to pass on to coding assistants. And so I try to have AI
180:45 coding assistants do all of the code for me now. But I still very much value my
180:50 understanding of architecture and just like how things are coded in general
180:55 different design patterns that you you know you learn with a degree or you can
180:59 teach yourself because that allows me to actually properly validate the AI coding
181:03 assistant. So learn the fundamentals of Python then get into learning how to
181:08 leverage AI coding assistance. I mean the exact same things that I'm covering
181:12 in this live stream. The things that I'm I'm you know constantly saying that I
181:15 cover in the agentic coding course when you learn how to like apply a pivotal
181:18 loop for planning implementing validating you learn about system
181:22 evolution and how to take advantage of tools like MCP servers and sub agents
181:27 like that is where you should go next. And now as far as building AI agents,
181:31 then once you have the foundation of like here's how I code, here's how I use
181:35 coding assistance, now you learn the specifics of how AI agents work like
181:39 specifically in the code and the higher level principles of like the React
181:44 pattern and how you incorporate tools in your agent and how you design good
181:48 system prompts. Um, so then you get into that next like that's like the main
181:53 three steps to really get in. And then if you want to go even further to
181:55 thinking about how you like actually monetize your agents, that's when you
181:59 start to like dive into research for like, okay, what's like a niche and like
182:03 an area that I can build my agents that like people want and will pay me for.
182:07 You can do freelancing or create a SAS or um you could just like do like cold
182:11 outreach or something like Upwork. I don't know if you're interested in like
182:14 getting an FR to like monetizing or if you want to like actually work for a
182:17 company and be like a an AI engineer or just like a software engineer on like a
182:24 forward deployed engineer team like I don't know exactly what your goals are
182:27 but yeah that definitely all comes like after just like getting the fundamentals
182:30 of coding coding assistance and building agents and like [snorts] definitely like
182:36 I mean I got to say because you like gave me some uh like a free ball here
182:41 like I cover all of this in the community because if I go to courses
182:44 here like I got the AI agent mastery course to show you how to build agents
182:47 from start to finish and I have the agent coding course for how to um build
182:53 with coding assistants and then if you want to learn like the fundamentals of
182:56 Python I mean I don't cover that in the course because or in the community
182:59 because that's like all over the internet already and I just think that
183:02 like there are great resources even free ones on YouTube for how to learn Python
183:08 like Socratica for example has a great um YouTube series for learning the
183:12 fundamentals of Python So yeah, appreciate it, Ty. Thank you,
183:17 Sean. Sounds good. All right, Peter. Peter is going crazy here. I love it. I now I mean going
183:25 crazy in a good way. I have Bitbucket and Jira. I put the Jira comment in. It
183:29 clone my Bitbucket and then manages the comment. Also, it will do an automated
183:33 PR review and a PR is raised. That is so cool. I love that, Peter. Really nice
183:40 Uh, you will learn a lot of coding skills through osmosis and dynamis.
183:43 Yeah, I appreciate you saying that, Scott. Yeah, just ask your favorite
183:47 LLM's questions constantly. That's really good advice. I think
183:49 that's another thing that I want to call out here. And I talk about this in the
183:54 Aentic coding course as well. Like trust me, you do not need to be an expert
183:58 software engineer to get a lot out of the Aenta coding course because I talk
184:03 about constantly like here's the part of the process where if you don't
184:06 understand the code like just ask the coding assistant like ask it away
184:10 question after question and so like a lot of like uh going back to the diagram
184:15 here a lot of the the vibe of planning for example like I said earlier this is
184:19 your time to get on the same page with coding assistant exactly what you're
184:24 going to build. Now, getting on the same page doesn't just mean the coding
184:28 assistant understands. It also means you understand. So, if it's presenting this
184:33 architecture to you or it's like calling out this syntax in Python or TypeScript
184:36 or whatever and you just don't understand it, just ask it. When you're
184:40 newer to coding or using coding assistance, it's really important to use
184:45 the coding assistant [snorts] as a teacher as much as it is the tool that
184:51 you're using to delegate the code. And so you build up your own understanding,
184:54 you know, like through osmosis, other people like Dynamus, like Scott said,
184:57 and then also definitely through just asking a ton of questions. Same kind of
185:02 thing with validation. I never recommend vibe coding. I always recommend you do a
185:06 code review. Even with the remote agent coding, like I showed earlier, we could
185:09 literally go to our phone and we could like look at the pull request and look
185:13 at the diff. Um, you can look at the code to review things no matter what.
185:17 [snorts] And I do recommend doing that. But even if you don't understand the
185:21 code very deeply, well, you still can go in and ask questions. Like I can go into
185:28 the poll request for example. Um, this one right here, it was closed now, but I can go in and I
185:35 can just be like, you know, at remote agent explain to me how the uh markdown or how
185:45 the messages are um aggregated for the markdown download. right?
185:51 Like it can be this simple. Um where you just ask a question where
185:54 it's like explain the code to me. Like if I'm a more experienced engineer, then
185:58 I can like click into the files change and I can read that myself. But um now
186:02 this is going to like actually kick off a task where it becomes the teacher,
186:06 right? And so maybe I'll go on to another question, but we can see its uh
186:12 answer in a second here. Uh where's the link to buy? Yeah, sorry.
186:15 Let me let me pull it up again. So yeah, if you want to join Dynamus and all of
186:19 this sounds great to you, I will put the link in the chat right now for you to
186:24 join Dynamus, it is the opportunity of a lifetime. I make sure every single day
186:27 that is the case, especially with the discount that we have right now that is
186:30 going away once the stream is over and the stream is going to end pretty soon
186:34 here. [snorts] Okay, so let me go back. All right, so here we go. Just to continue the uh your
186:42 the first question other question here. So So we have our utils. This is where
186:47 we [snorts] excuse me this is where we have the markdown export feature. So it explains
186:53 the input the header generation. It talks about how we iterate through our
186:56 messages that we have in the current conversation. I could ask it follow-up
187:00 questions like explain the code in more detail. But yeah, I mean this is like
187:04 fantastic like you can ask these kind of questions constantly so that you build
187:08 your own understanding of how your codebase works and how coding works in
187:13 general as you're working on the different features in your piv loop that
187:17 also helps you evolve your system over time as well. Real world development in there. That is
187:25 right. That is right. This is one of the best AI communities.
187:27 Keep up the good work, Cole. I'll be here for at least a year. Awesome. I
187:31 appreciate it very much. I mean, I I respect a year commitment. Like, that is
187:35 a long time. So, thank you very much. I 100 bucks. He says, "TypeScript next."
187:45 Sorry, I don't I I don't know like when you said this, so what it's in reference
187:50 to, but um or if we meant like what languages to learn. So, yeah. U by the
187:54 way, kind of going back to the original or the the question a few questions ago
187:59 about what to learn. Um, it's funny you said TypeScript next because that is my
188:02 second favorite language. So, I use Python the most. Um, Typescript is my
188:06 second favorite language. And I mean the whole remote agentic coding system, I
188:10 actually did build this entirely in Typescript. And the reason for that is
188:14 the cloud agent SDK supports TypeScript and Python, but Codeex and Open Code, a
188:19 lot of the other SDKs I want to integrate are only for TypeScript.
188:24 So, yeah, the the language the right language depends on the code base. And
188:28 so it's good to learn general programming principles that apply no
188:33 matter the coding assistant that you use and um or like the codebase that you're
188:37 working on. And then also the other thing is in the agent coding course I am
188:43 always working on systems that are programming language agnostic. So
188:47 everything I cover in the course and really like a lot of what I do on
188:50 YouTube as well like the systems and coding strategies that I share for free
188:54 on YouTube. I mean, all of it is like this is going to work no matter the
188:57 specific programming language that you use. So, in the end, you can just pick
189:00 the one that's best for what you're working on and then you know that that
189:04 your system that you apply to the codebase and maybe customize over time.
189:08 It it's going to work for any language. And so, like I picked TypeScript for a
189:12 reason for this language or for this project. A lot of times I'm using Python
189:16 because I think it's the simplest and also Python has the large largest
189:21 ecosystem of AI libraries like um graffiti for knowledge graphs is
189:27 only Python. Dockling for uh file extraction and rag is only for Python.
189:30 Like there's just a lot of those libraries that I really like using that
189:33 are only Python. Now there are some like like langraph for example that is Python
189:38 and Typescript but um not many are like that. And then once you go beyond Python
189:43 and TypeScript or JavaScript, that's when you other languages are very
189:47 limited in the AI libraries that that you have. Um like if I were to say like
189:51 my third, fourth, and fifth favorite languages, it would probably be um C,
189:57 Rust, and Go. But those three languages have a lot less as far as AI agent
190:01 frameworks and other open source libraries for rag and AI coding like
190:05 whatever. um right like Python and TypeScript are the king as far as um
190:11 just the open-source ecosystem of tools that you can leverage um like the agents
190:16 SDK and codeex SDK that I'm using here Oh, cool. Founding member as well. Cool.
190:25 Cool. That's how I got Typescript right. Osmosis. That's awesome. Yep. And then
190:30 Sean said, um been a part of Dinos from the beginning. learned a lot from other
190:33 people in the community just as much as the courses. I've actually met quite a
190:37 few people I continue to work with. That is very cool. I appreciate you sharing
190:40 Sean because yeah, that's another part of the community that I haven't even
190:42 like talked about that much during the live stream is like yeah of course I am
190:47 working my absolute heart out every single day to provide an insane amount
190:50 of value to you. But also like what I give isn't even the only thing. It's
190:54 also just everyone in there that's um also very active and people are very
190:59 collaborative in the community and yeah, it's just awesome. Like even I'm
191:02 learning things every single day because like let me tell you I reply to every
191:08 DM, every post, every reply, every comment on the courses and I'm I'm doing
191:11 it like every single day except for Sundays. I take Sundays off. Um but I'll
191:16 still catch up on everything. So like I'm constantly a part of it and that
191:19 also means that I get to learn a lot from you guys as well. So it's just an
191:23 absolute pleasure to be in the Dynamis community and driving it forward for us
191:27 all every single day. So yeah. All right. Um, with that, I'm I
191:34 think I'm gonna go ahead and and wrap it up here because we are already over
191:37 three hours, but yeah, this was absolutely fantastic. Um, so yeah, I think like the last thing
191:44 is I didn't really cover the architecture fully for this, but I know
191:47 I talked about this a lot already. Like the main thing here for this
191:51 architecture that I didn't quite hit on is I have this orchestrator that defines
191:56 like these common interfaces for our different integrations for the platforms
192:00 and the coding assistants. And so as long as a platform or coding assistant
192:05 like implements these functions like this is how I receive a message, this is
192:09 how I send a message then the orchestrator can immediately attach to
192:14 them and it acts as a middleman. So that way when we create another platform,
192:18 it's not like that platform has to have a custom integration with Claude and
192:22 Codeex and then whatever coding assistant I integrate. So I can add in
192:27 10 platforms here and then 10 platforms here and then it still only has to be
192:32 like one integration at a time because we just have to work with the
192:35 orchestrator and then the orchestrator is the glue that connects each platform
192:41 to each coding assistant. So I can use GitHub with codeex, I can use telegram
192:45 with cloud code, but the application doesn't care the coding assistant I'm
192:49 using and vice versa. That's the beauty of the architecture here. And I planned
192:53 this out in the PRD before I even implemented anything in the code. So I
192:57 did a ton of research and just like thinking with the coding assistant. I
193:01 created that PRD and that's one of the things I cover in the agentic coding
193:04 course like how we go through that process. And I thought about like how I
193:07 want to load commands and manage that in the orchestrator as well and persist
193:11 everything like our conversations and code bases and the commands that we've
193:14 loaded in the database. Like I thought through all that up front and then I
193:18 just did piv loops like multiple pivot loops to split the PRD into granular
193:23 sets of work so that I wasn't trying to do too much at once and I knew that I
193:27 had a reliable system that can knock out all those things one at a time. And of
193:31 course it wasn't perfect. I encountered issues, but I only had to iterate a
193:35 couple of times to build out each of the things here like the orchestrator and
193:39 the different platforms and coding assistants. And so I really have
193:43 designed this optimally where like I'm very confident that like when I add
193:46 Slack and by the way I'm going to be adding Slack to this system within one
193:53 of the next workshops for the course or for Dynamist. So if I go to events here,
193:58 I actually have this planned for Friday, next Friday. So, Friday, December 5th is
194:03 when I'm going to be live with you adding Slack as a platform to our remote
194:08 AI coding system, which that's going to be a blast. And I know that I'm going to
194:11 be able to do this very effectively because when I add Slack, it's not like
194:14 I have to integrate Slack with cloud code and with codecs. I just have to
194:20 build on top of this um interface, this I platform adapter. I plug it into the
194:23 orchestrator and then the orchestrator handles sending everything to each of
194:28 the coding assistants. It is a beautiful thing. So yeah, I wanted to end the
194:32 stream by just talking about the the architecture really quick. So I I know I
194:35 said I'd get into it more, but I hope that makes sense. It's very cool. So
194:40 yeah, with that, the last thing I want to say before I close off our stream for
194:46 today is just thank you everyone for being here. All you who joined Dynamus
194:51 as well, taking advantage of the special exclusive discount that we have for this
194:55 stream. [snorts] If you like just joined in the last like 10 minutes or whatever and you haven't
195:00 had a chance to take advantage of this, I will leave this up for yeah like
195:04 probably about an hour after and then that's when I'll tear down the discount
195:09 as well as the repository for the remote agent coding system. And so yeah,
195:13 definitely like download the remote agent coding system as well because I'm
195:17 giving that away to you for free for this live stream. And um yeah, last
195:21 thing that I'll say going back to the course here is that um I don't even know
195:28 I can't even begin to estimate how much time I've put into creating this course.
195:35 Like 18 hours of content is hundreds and hundreds of hours of prepping and
195:39 recording and editing and getting descriptions up and thumbnails and and
195:43 redoing things a million times because you know that that's how it goes. I'm
195:47 very much a perfectionist. And if you've ever made a YouTube video yourself, you
195:51 know that when you have a 10-minute video, that does not mean it is a
195:56 10minute recording session or a 20-minute video is not a 20-minute
196:01 recording session. It is multiple hours. So, yeah, it is it's been insane. So, I
196:06 hope that you enjoy the course. If you are not in the community yet, like this
196:11 is the best chance you'll have as far as the discount goes. the Black Friday
196:15 discount will go through Cyber Monday if you are watching the recording and you
196:18 missed it. The last thing I will say is I mentioned this at the very start of
196:21 the stream, but four o'clock central time on Monday. I will open it up again
196:25 for just like another hour u if you want to like grab this link from the stream
196:28 right now and go to it then just for those like in Australia where the time
196:32 zone is just really bad. So I wanted to um give that as an opportunity as well.
196:35 But otherwise like this is going down after the live stream. So take advantage
196:41 of it now. And yeah, we even got like the next evolution of the remote coding
196:44 system like literally next week. Like I'm I'm making do on the promises
196:49 already uh because we got this coming in um for the next workshop. And I do these
196:53 Friday workshops every single week. They're always like super engaging,
196:58 super educational. We usually got like somewhere between like 70 and 120 people
197:03 in those live every single week. So yeah, like a lot of activity um in these
197:08 workshops. I mean, like that's like a considerable number considering just
197:11 like we have them every single week and the attendance is still like really
197:15 high. So, it's always an honor. Um, so yeah. All right. I'm going to go ahead
197:20 and go to the full frame here and uh send us off right now. So, let me uh
197:26 turn down Well, I got to turn off the comment here, too. All right. You're an amazing guide for
197:32 the tsunami of AI. I appreciate it, Scott. That's what I try to be for you
197:37 guys. And um yeah, both for all the free stuff I do on YouTube and in the Dynamis
197:41 community. Both of those things are always going to be my core focus. So I
197:45 appreciate it. And yes, you're very, very welcome. Cool. All right. Yes. So hopefully my
197:52 voice will be better um after tomorrow because I'm getting right into recording
197:56 the next stuff for YouTube before next week. So yeah, no video for tomorrow
197:59 because I'm kind of doing the live stream instead of a weekend video. I've
198:03 been doing that a lot more recently. But yeah, got another video coming out on
198:08 Wednesday like usual. And then yeah, the events every single day in Dynamis. So
198:12 all right, with that, thank you everyone for being here. This was a fantastic
198:16 live stream. All you guys' engagement super appreciated all of you joining
198:19 Dynamus, thank you very much. And all of your great questions as well. Hope that
198:24 you guys have a fantastic rest of your weekend. And I hope to see you all in
198:28 the Dynamist community as well. Take
