Flow AI Bootcamp — April 19, 2026 · 1:06 PM–4:29 PM PDT

# Transcript: 2026-04-19

> 1 time blocks from 1:06 PM to 4:29 PM

---

### Technical setup discussion
1:06 PM - 4:29 PM PDT | *meeting*

Microphone:
I'll give it one or two more minutes. Okay, um, are you guys Gabby? I'm assuming you're on a Mac? Or no? And Jose, you're on a Windows PC? Yeah, mostly Windows 10. Do you guys use the Windows Subsystem for Linux? 100%. That's the only thing that I use. Okay, was that you? No. Okay, I don't. So that might present a little like conflict or dependency issue. It might be good because I have some ideas how we can make it easier to work as a team and pick up where other people left off by creating some tools that we can sort of install and keep up to date in our repos that we can use for things like deploys and doing queries against the database, mostly like them to work on. Holy shit, that was loud. At least on my part, I've always worked on the subsystem, so I'm not completely alien to all the issues that come. And I've already set up my environments to be pretty soft, so I rarely get those issues, but yeah, they do give issues; they're fucking annoying. I've learned to deal with that shit. But um, how about... I mean, I've worked on Linux; I pretty much exclusively work on Linux at work, so I'm familiar with Linux, I just don't use it at home.

Yeah, I was thinking maybe we could just get you to install WSL. Because that way, like, anything I build for us to use as tools will sort of be under the assumption that we're using a Linux system for our package installation. It's pretty sure it's just as simple as an install script and you can just run in and you've got a separate Linux interface on your Windows computer, but I haven't done it personally, so we can revisit that. It's not a huge deal; I'm just not sure what the compatibility is going to be with stuff like Wrangler and some of the Cloudflare tooling we've got. Things like that. Do you know how any of that stuff reacts with VMs? Um... It should be good, I do. So mostly, like in the setup, I can help with the setup; it's not that hard. It's just where you're going to harbor everything inside the subsystem for it to actually work better, and your computer needs to be compatible with at least WSL2. If not, then it's deprecated to WSL1. But once you enable the subsystem in Windows, you would need to restart your computer. You open your file explorer, click this PC, and you see at the end like Linux, and everything's gonna be harbored right there. And then for your projects folder, you are going to create a shortcut on your documents so that you can open any cursor or any IDE followed with the WSL path. And then you would need to route something so that the paths are for Linux instead of mounting on the Windows OS, and that's how you're going to be avoiding those issues. But it's not that hard.

And then if you're going to SSH into a VM... VM is going to have links so you can connect it by any IDE; you can host in Hostinger or Hetzner and you can have your VPS over there, and then you just SSH. That terminal is a WSL Ubuntu terminal, so you can just rock on it, you know? It wouldn't give you any issues in your Windows OS; it would just be on the VPS right, so it should be fine. Yeah, I normally work like on two VPS and my computer at the same time, so VPS being Virtual Private Server... yeah, Virtual Private Server where you have your VM inside, and then uh, you just SSH. So as long as you have the key and you can connect, you're good. The interface will just be your IDE and your computer, but all processes are over on the server, so you should be good. All right, we'll definitely have to come back to that. Yeah, yeah, yeah, don't worry. Or something? What's up guys? What's up Britter?

The first time is a headache, but then it's okay. Thank you. I wish I was chilling. You wish you were chilling? No, I wasn't. I've been working on some server-side issues with a project on the company's side trying to resolve it; they blocked the server so I'm doing some investigations, getting some evidence, and then submitting a request. They make you work on Sundays? No, I choose it because I like this. They're not paying me or anything. That sucks. No, yeah, but it does suck, but at the same time I like it, so it's bad even though. Okay, sorry about that, guys. Okay, Miguel's here, too. Great. Alrighty.

I don't mean to interrupt your guys' conversation. Were you talking about VMs or something? Yeah, I was just talking about how it would normally go. You need to activate the subsystem in the settings, make sure that it's enabled, and then download Ubuntu, and then you would get Linux on this PC, and that's where you need to path the projects or anything that's going to go there. So if you normally have the issue, it would come with your IDE; your IDE will be normally pathed to your MNTC, which is the Windows OS. You would need to rewrite that path so that it paths directly to the subsystem in Linux. So when you open your IDE, your terminal's actually there. But it's, to be honest, kind of aesthetic and a commodity because you don't need to do that; you can always open your IDE and then navigate through. You open PowerShell, right? You write WSL, and then you bash your way back to the root, and then path your way to the Linux system. You always do that, but if you don't want to do that repetitively, then yeah, you change the path for your IDE so your IDE can open like you directly into the subsystem. And then you're good. All dependencies are not everything; it's fine. You just download it and you won't have any problems because you're harboring it inside the subsystem which is Linux, so you should be good.

We'll sort of take that as it comes. Yeah, no way. I'm sure it's probably difficult to get all that over Zoom without written material and stuff. So, when required, we'll make sure that everyone's got the necessary instructions for setup.

If we go that route, even if it's easy to set up on Windows, we could probably just write Windows compatibility code to make sure it works for any system we're using. So, no need to worry about that. I appreciate your domain expertise there; no worries of course. If it comes down to it, I can help out—don't worry.

What about familiarity with Docker? Everyone has worked with that? Of course, yeah, I've worked with it. Okay, cool. Yeah, I haven't used it much in Flow, at least as far as compatibility issues go; like that's usually my go-to solution when developing something. So, if it comes to it—which I don't think it will—we can also explore that for our local development. But anyways, without further ado, I'll sort of dive into the weeds.

I guess I can start with a high-level overview of the three things I'm going to discuss today and hopefully answer some questions: what Flow Systems offers to anybody out there. We sort of spoke about this last week a little bit, but at a high level, we're designing and building websites; we're building AI-driven automations; and we're building chatbots—or chatbot relays, for lack of a better word—which are ways to interface with AI systems. That's sort of what we do at the core, and then we've got our bespoke integrations and builds, which are usually some sort of variance of those three things. As I mentioned last week, we're using Cloudflare primarily because of what they offer us as developers regarding their developer tools, what they can offer us as infrastructure providers through their Workers platform and persistent solutions like D1 and durable objects—a bunch of really great infrastructure tools for microservices—via the CLI.

With all that being said, I'll sort of jump in a little bit and start showing the code. This is sort of the repo that I've created; it's a little bit of a mess, but a handful of directories contain a bunch of different configuration dot files, a handful of contracts and service agreements (change the name), and then in here I do all this as like a monorepo style so everything is usually in one place. The benefits of a monorepo won't really go into; it's just for code organization stuff and the way that we do the microservices—it just lends itself better since you can just sort of keep things separate. And what it allows us to do is give us this opportunity to build templates for different packages that we would use in different customers. For example, like the events queue here—which might be hard to see; try and zoom in a little bit—the ingest queue or worker intake, nurture scoring worker: this is used in a couple of different places for a couple of different clients, just tweaked slightly. There's some temp and configuration stuff that I could do to make it even more easy to sort of drag and drop into different places, but it's something I'm going to work on; I'll probably do that. I'm open to it.

Then, I guess for starters, I'll show Wrangler CLI. Once you've got it installed, you can hit `wrangler login`. In this case, I've got some API keys, so I need to move to a different folder temporarily. There are bunch of different Cloudflare accounts on my computer which is a source of confusion for myself and my company agents, but you make do with this going forward, um, which is a little bit janky, but saves us money. We're going to share the info at email part of the or info at flowsystems.live email. Part of what we've done is all emails for the team and systems.live; everyone receives it, but it also goes to our distinct inbox. But for services similar with Slack too, which we haven't been doing a very good job at leveraging, I'm hoping that's something we're going to work on later though. Yeah, you know, it's something that with those teams so small doesn't necessarily make a whole lot of sense. No.

Here we go. So supposedly the name came first, but it's a searchable log of all conversation and knowledge which I didn't know that; I was taught that just a couple days ago. I love thinking of Slack as that idea, having like a queryable repository to operate as our team grows—Slack will make more and more sense. Nonetheless, I'm not trying to sell you guys on Slack; we can go on to the info discussion. So one minor thing that I do or started doing relatively recently is creating different environments. Maybe for the sake of this, I can also go into... And then `wrangler.jsonc`, thank you. Okay, so looking at a `wrangler.jsonc` file: this is how all of the workers and worker static assets that sites are configured and deployed. I think it's a very nice... Oh, is that a call? No, no, no, yeah, sorry, that was... All good, you're allowed to cough. Sorry, I already muted myself man come on.

So what we've got set up here is we're using configuration language; you can write this in TOML file format or JSONC. Here, it's fairly simple to define environments. You basically just add this `env` object to your `wrangler.jsonc`. You say, "This is the name of my environment: staging." This is the name of the bindings that we're going to be using. And bindings are the databases, the queues, the key-value stores—anything in the Cloudflare infrastructure that you attach to a worker. This simplifies things dramatically from a configuration standpoint; I wouldn't need to set a secret for my Postgres database and password because I set up D1; it's like already out of the box going to work directly with that. That's sort of at a high level how this works. I'll show you one more of these files so you can see how the different `wrangler.dgson` distinction in this versus the other is that there's an assets object defined here.

And this is just to say: if you have some sort of compiled code that you would like to serve on, like at your root path or at your Worker's URL, you can specify the directory where my compiled assets are located. When you deploy it as a Worker static asset deployment, it will be served at whatever the URL or custom domain is.

What the Cloudflare dashboard allows us to do, one thing is assign custom domains. In the case of FL Collection, we have these Workers dev URLs, which is how everything references one another. Like, if the site is trying to connect to the site, they want to name it to flbrokers.io again. We could change the custom domains for the Worker configuration, but the automatic deploys are only saving you like two seconds of time basically. And then beyond that, you've got some nice metrics building these microservices together, sort of as a combined infrastructure. It's really nice.

I agree. Miguel did a great job. Wow, I'm flattered. Thank you, Tim. Yeah, you've already heard my praises. So, yeah, I'll stop right there for a sec. I'm sure there's a little bit more I can go into, but questions, thoughts, concerns?

So currently, just to summarize and tidy everything up: We're using Wrangler CLI for deployment connected via Cloudflare. We have the different domains. We're doing process-of-the-art automation depending on what that process needs to be automated pertaining to the company or project that's been handed over. The UIs are being pushed out by Miguel; the backend code is being handled by us. Basically, the company would get a page, they would get a deployable frontend, and they would have their request in the frontend that we create. And it's mostly creating chatbots and processes or, depending on again what the project requires.

Yeah, I think obviously, like, it will get a little more complicated than that as we go. And I think it's also like... That's fine. Yeah, you can continue. And I try to keep things relatively facile here; it's not going to be helpful to you guys understanding how things work if I just get right into the nitty gritty. There are so many projects that we're working on and a handful of things that are just... But, yeah, the way you described it sounded good.

Yeah. To be honest, like, the remediation or organization is going to be subject to once the team grows, right? And then the processes that you do will be hardened when we try them ourselves. In terms of remediation, we could do a repo and then GitHub issues, where each project can get an issue for organization remediation in the terms, right? That really doesn't bother me; I think that will ultimately come. But yeah, the stack is very important—the stack that you use. If you guys are using chatbots on everything, have you used Flowise? Or have you heard of Flowise or anything?

Flowise is really... really nice. It allows you to create in-website chats with AI agents and it's very seamless, and it kind of works like an N8N in a way. You have like a canvas and you have the different nodes and the different requests, and then you just put the system instructions for the chat, and then you have your bot for your page. Pretty easy setup. Pretty cool, too. You can check it out.

Is it open source? Yeah. I do, yeah, I was using it; yeah, you could have it self-served, like N8N, and you can run it via Docker container and you can have it self-hosted, and then it's free but it should be open. So let me check. I'm looking at the repo. It appears open source. Yeah, it should be good. Seems like a cool tool.

Yeah, I was looking into N8N... I do love like canvases and UIs for building these automations. Yeah. But honestly, I found myself like... Um, with Claude Code, I just feel like it's an extra step. I'm not going to say that there's anything wrong with that. And I think as we bring Luis and Miguel up to speed, a tool like this can be extremely powerful. Yeah. I mean, definitely, you know, obviously for you and Gabby and Paul... So, yeah, probably the hard-coded ways it's going to end up being, or not the hard-coded way, but, you know, you guys are probably going to be able to, like, actually make the best out of the play with code and whatnot. But obviously, Miguel and I are, like, newbies here, so any type of tool that makes things a little simpler is... you know, yeah?

If I can share my screen real quick if you don't mind—do you have uh... can you give up? Let me know if you can't and I'll see if I can edit it. I think I can actually. Right? Yep. So one of the things that I really like when I try to implement these things is Cursor; you can index, right? So I go to N8N dev documents directly from N8N and I go at a doc and let me try it and do it right now. So, because I don't want to read anything and I don't want to like learn that much... do it so I already know how to do it but let's say that you don't and you don't have the time and you just want to do something with N8N. You go to the docs and edit N8N, and you get the URL, and you just do this. This was ClickUp one, so let me just copy again. And then you... you put it in, and then you put the name that you want, you click confirm. I can't use this because I already have one named like that, so whatever, I already have it here.

And then I create my agent based on this documentation, so then I have an agent that is versed in N8N documentation directly. And then I go to my MCPs and then I connect N8N MCP over here, and then I take and I self-host it with Ngrok. I create a container and Docker so I don't have to pay the monthly due and I can have it self-hosted by a Docker, and then I can use all the capabilities for N8N and I don't need to pay for N8N so it's completely free, which is really, really nifty. So I set up with that.

And then basically, I say what I want to do: I create the PRD, and based on the PRD, I create the workflows. The way that it works is that the agent is already versed on how the nodes should be and all the updates for N8N. So I just need to copy the JSON and paste it in the canvas, and all the nodes will be there. The only thing I need to do is click the ones I need to configure with personal credentials, and that's it. I have my full automation workflow created in less than five minutes, which is really, really cool—really nifty. It can be done the same way with Flowwise for chats, bots, and anything that needs to be pushed like today.

If you're thinking about incorporating AI at an end, I would recommend this flow; it could help. You can test it out, see how it works for you and everything. But yeah, ngrok self-host makes it free. `mc index.create`, agent create workflow—that's it. Pretty simple, honest.

I think you'll find that our automations are actually simple to the point where these tools are probably like the canvas for most of our automations because of the way they're structured. It would just be like three nodes next to one another; all of them are just like three nodes. They work, man. They work. And that's all you need: stupid simple, make it fucking work. That's it. That's the dream, you know?

So I'd love to pick your brain about that a bit. I'll take a look at the code and see if there's anything we can do there. I'm sure we're gonna have some more advanced chatbot implementations. You'll see right now what we did for our client Eduardo; it's sort of like less of a chatbot, like a dispatch for Cloud Code. They came out with a bunch of features that basically allowed you to do the same thing. All that's to say that we basically built something up. There is now a native feature for Cloud Code. So it's not even a chatbot in the sense that you're probably imagining; it's more like an interface to relay messages to Cloud Code and have it send it back.

I think some of our... well, I'd like us not to be dependent on Claude Code and B2 to be a little more defensible. So it's all stuff that's at least top-of-mind for me. Yeah, and if that comes to the moment, we can always serve it locally. We could use Cloud Code or any coding CLI and have it serve locally so the data is not being pushed out. There's also a feature on Cloud Code: after 24 hours of account creation, you can put it in private mode, and then no data that is used in your Cloud Code will be sent to the model to be trained or anything. So it adds like a mini layer of safety.

The good thing about using the Cursor IDE is that you can symlink it with Cloud Code. So that MCP for N8N, those indexing docs for N8N, and everything that you do in that IDE can be accessed via Cloud Code as well. You don't need to set it natively inside Cloud Code; you can just set it in your IDE, symlink it, and then it's useful out of the box. So you don't need to use Cursor for it; you just leverage Cursor.

What folder are you symlinking to get your Cursor configuration in Cloud Code? Normally, you would have a `.cursor` file, right? I would link my `.cursor` file by symlinking it to my `.cloud-code` environment. Anything that I have in my Cloud Code, I can use in my Cursor chat, and anything that I have in my IDE, my Cloud Code can look into the software sources in my IDE and get them out as well.

I didn't realize they have the same configuration primitives as... Me neither. To be honest, I was just looking at a wall and started trying to reverse engineer it, then I found that. So I completely get it because if you haven't hit that wall, most likely you wouldn't know. I just mentioned it because I've been there, and it was horrible man. It's something that helped me, of course, which is why I'd like to share. That's just it.

Yeah, no, I mean honestly, I don't know. Are you there? Yeah, I'm eating lunch. Sorry, you're keeping up. Yeah, I'm following. Sorry, it's... I probably could have done a little bit better job at structuring this, but I was hoping that this could be more of an open discussion: sharing infrastructure, familiarization with the tools, and getting a better sense of how we're doing what we're doing currently.

I'm assuming now this is like out of left field for you currently? Yeah, I mean, I've barely ever used AI in the way you guys do, so I'm very new to this. I'm kept in a fucking locker for my workday. So all the stuff you guys are talking about is a little bit outlandish to me. Well, maybe that's... For the next bit of this, how about I spend a little bit of time talking about our AI development environment for Flow Systems and show what this looks like? This might be less exciting for Gabby, man, and more exciting for you.

Not at all. Maybe everything is exciting; it's good. That's what I like to hear. Anyways, this is what Cloud Code looks like when you open it up. This is basically the only way I've used our code at Flow Systems. I have basically not used an IDE once with our code currently, and that has all sorts of downsides. You could probably tell just by the code organization alone a little bit of this is to blame. But one thing worth noting is that you can sort of run bash commands in Cloud Code.

I just remembered you said Zbela, and you just reminded me that I need to do something real quick. I've got a couple things to do for Sabah too.

But anyways, a couple of things we were talking about probably went over your head, Jose. One is configuration primitives for the agents. What we mean by that, typically, is skills, sub-agents, MCPs, and hooks as they're called. In Cloud Code, there's also a concept of skills and slash commands, which recently were joined. As you see at the bottom of my screen, when I type in a slash, it shows a list of commands. We spoke about this in passing last week regarding the skills.

So if you type in a slash—for example, if I want my agent to open up Chrome—it will do that. I can go here; I'm not sure what a bridge looks like yet, but there we go. It's currently controlling my browser. I'll let it go do whatever AI-native practices are necessary. In here, if I go to skills and list them, you see where I'm heading? It just has all these instructions on how to use the Chrome DevTools node CLI package. When I call it via a slash command, it shows the agent this markdown file along with some other reference files. With that information, it can go ahead and do its thing. Oh, me being slow—we've created a couple of skills for different uses in this directory; let me see if there's anything here worth using. You also have controls for your settings and a bunch of configuration things you can look at using slash commands. It's fairly useful.

Then with sub-agents, this is where I think the most powerful and useful parts come into play, though these other primitives sort of help things along. For example, here I might say, "Can you please take or provision sub-agent marketing functions?" When I kick that off, it goes ahead and provisions sub-agents which share the context of my main thread. I have five sub-agents right now; it's crazy. This is my main thread where I'm directly interfacing with the model, and then it spins off the sub-agents. They are all sharing the same context but don't muddy the context of your main thread despite having shared access. The way to put it is they're sharing the context cache of the main thread. They also allow your main thread not to have to read through hundreds of thousands of tokens in order to get the same information. Equally capable models go and do this research and exploration without affecting the thread you're working on primarily.

Beyond that, you can customize them. There's a slash command here where you can create agents; it tells me which ones are running—ones I've created. You can also access a whole plugins marketplace where you can download new ones. There's a bunch of stuff that helps just configure it with what you'd like, and ways to create your own custom sub-agents or skills. I mean, there's a good chunk of this, but yeah. Beyond that, in a similar vein, you can do something like here: I've got a deploy command where when I call this and name a branch, it kicks off a deploy to our Heroku staging environment at my day job to send some new code up to that environment. That Cloud Code can leverage, but just about anything in a markdown file can be used in a way similar to skills and slash commands. It's just a matter of doing that; I recommend it, but it's not essential or the end of the world if you don't.

I mean, that's about it as far as like... I'm not going to bore you with this stuff. Don't worry, it's not boring, it's just a lot to dig into at once. I think what I'll do is send some resources. Also, just so you know, I kind of wanted Cam to meet with you and Gabi now so he can give you the rundown on how he's working on it. Ideally towards the end of this conversation with him, YouTube, you guys can start figuring out how you're going to share resources and how you're going to be able to use these resources that Canvas created. Likewise, if you have resources that you're going to provide, Gabby, and so on and so forth. Afterward, Cam is going to sit down with Miguel and me, and we're just going to go from like literally step one—setting up Cloud Code on the computer—and from there, Cam will help us get some resources and understand how to use it. You can stick around for that as well so you can get started; you'll probably pick it up a lot faster than us, but I usually learn better with hands-on approaches anyway. Learn as I go.

If you have the time, I'm thinking we're probably wrapping up here with the technical discussion. Maybe a quick water break or something? No, to be honest, I find it great; I can definitely share some resources as well. I created my own status line for Cloud Code that I think will be pretty good. I found one done in Python but reformatted it to Bash for faster execution. I made a dynamic so if you change models, it accurately tells which type of context you have left. I also made a dynamic for tokenization so it tells you how much money you're spending on that session and everything. I can add other stuffs as well, so I can share that; it will help a lot so we could monitor our sessions and see how we're doing with context—when we should compact, when we should delegate, or upload jobs as well. A few other things that I have in the locker room that I can take out if I don't get into trouble with my job, of course. Yeah, I could definitely help with that.

I don't think I will be needed for the setup. Yeah, I think you're probably good on that, Gabby. So I'll just try to resolve this server issue that I'm having with Hetzner right now. But yeah, it sounds great. I got an understanding of what you guys are doing and how you guys are doing it. Later on, we can talk about optimizations or environment setups and having a better re-envisioning. I would like to help if I can setting up GitHub and creating those CI/CD pipelines, and creating those GitHub issues and PR requests as well.

So what I'll do... What was that? No, no, no, go ahead. Go ahead. Yeah, what I was gonna say is I'm gonna create—and I believe I already have one—but I'll have sort of like our messy repo in its current state, make sure it's updated. I'll invite you as a collaborator; I'll invite everyone here with a GitHub account as a collaborator so we can sort of take a look. I'll flag the repo or anything. And then you can take a look. We can sort of discuss asynchronously what sort of changes we might want to make differently.

We could do a Discord server as well, honestly. I'm down for whatever we prefer. And Discord, the one advantage is that it's gonna be free, you know, for what we need it for. Yeah, which is really good too. There's also ClickUp. ClickUp is really good as well because ClickUp has all the formats for any type of document, and you can embed just HTML code, and it pops up as the page inside ClickUp. You have channels; you have Workspaces. This is free. You can connect it by API key. If you need to put docs, excels, PowerPoint slides, any type of formatted presentation of data—even everything, md files or anything—ClickUp is really good as well. And yeah, it's free obviously if we scale way bigger than there is a premium or paid version. But yeah, you guys can investigate ClickUp as well to see—you have Kanban boards and progress maps as well coverage. We had Slack going here fairly soon, hopefully. Assuming that's a good way for us to communicate if we have some sort of preference that we do it. Maps and a lot of things are very, very dense, very, very... But if you don't need it, then we can just keep with Discord and WhatsApp. But the good thing about Click is that it works like Slack; it works like Discord as well, and you would have a centralized source of all documents pertinent to each project that you can modularize the projects in different workspaces which promotes organization and keeps everybody on the same page. I don't know, I just mentioned it because I use it at work and it really does help.

I saw your MCP server for it, which caught my high. Yeah, it does have an MCP server. There are some ones that are outsourced but I normally use the API key because there's two MCP servers out there for ClickUp: there's one that's only with the premium account which I would say to stay away from that one. And the other one you can, well, you can use it super well with cloud code. But the reason why I would go with API is that API can technically do everything that the MCP can do in a way and it costs less tokens because that's the overall problem with the MCPs. MCPs drain usage and it's fucking insane. There is an open repo, open source repo that converts a lot of—100 and 50-something MCPs to CLI—which is really, really good. And I've been also working to see if I can myself convert some MCPs to the CLI, which would be better for economics because an MCP can roughly spend around 800 tokens per tool call, and the CLI is only like 60 or 80. So it's a huge difference. So yeah, but the MCPs, obviously, they're awesome but just keeping in mind that they can drain usage, you know? I've been converting all of my MCPs to CLIs recently and it's just more powerful to be able to fit so much more tools without flooding your context.

100%. 100%. And Jose, don't worry. We'll get into what MCPs are. We'll set one up today. It sounds like a scary acronym, but... It's really just a code name for tools in the context of AI. Yeah, software is all acronyms anyway, so it's okay. Yeah, it's easy to use to that, 100%. If you guys ever end up taking Cybersecurity Plus from CompTIA, it's literally just an acronym test. That's an acronym test. I believe you. And if you do, I have the Quizlet that will have all of the acronyms, so let me know. Hell yeah! Bro, this is a plug. This is a plug. We're about to be cybersecurity experts. I'm telling you.

Also, quick little side thing. Having taken Cybersecurity Plus, I know NPM is a very common target for cyber hackers. They can basically poison the package registry and just, you know, put some malware where you were expecting—I don't know, Yum or some shit. Yeah, and there's some CI stuff that you can do to try to avoid that stuff. Depending... I think the way that our Cloudflare, uh, for a lot of our tools is great because I don't even think as an admin I have the ability to control the server in the same way that like if I was able to gain remote control execution of a true VPS. Like I think it's fundamentally different in their edge network using these runtimes that they have. But that's not to say that we're not at risk of supply chain attack at all. For sure. But I mean, I think Cloudflare would probably handle that before we would have to, right? Yeah. I believe Cloudflare is, to sort of prevent you from making the mistakes that would lead you down those sorts of paths. But it'll be great to sort of have you because... Like my AIs have been making some poor cybersecurity decisions. And I haven't been in for time; I've been trading off on spending a lot of time doing security reviews and stuff.

So I'm sure there will come a time where it'll just make sense to go over that stuff, figure out where the vectors of attack exist and how we can address that. But first, I need to know how the fuck this shit works. So let's start there. 100%. Sounds like a good place to start for sure. All right, well in that case Gabby, I'll make sure to share the GitHub repo with you. Please send over.

I've saved Flow for a while ago, like a bookmark collector called Raindrop. I like it as a way to share bookmarks effectively. It's not that there's much of a great need for that or anything, but I've really enjoyed it for just amassing different links online, especially for tools and stuff. Yeah, this looks awesome man; I just looked it up and it has an API and everything. Hell yeah, okay. So I'm excited. I'm glad you're sharing all these tools with us. I think it'll be great to sort of beef up our stack a bit. Hopefully get us... I'd love to have input from everyone on our tooling as I'm not very opinionated myself. We'll figure that out as we go, and yeah, we can move on to our non-technical training here shortly.

All right, our baby steps training. That sounds great man. All right, so guys, I'm going to head out. Have fun. It's fucking fun; it is fun. So just try to have fun. I'll see you in the next one, right? Take care, take it easy. Have a good one Gabby. "Try" is the operative word in that sentence. All right, do we want to take like a quick three to five here? Get some water. Yeah, I have to poop. Okay, boy. Yeah, yeah, yeah.

Okay, so let's do... add seven minutes from now, sound good? Uh yeah, all right. Do we have time constraints by the way? No, I'm good, I'm good. Uh yeah, I assume it's going to be like another 45 to 60-ish. Yeah, more or less. Okay, we'll get at it like at your pace. But if we're going to download the MCPs just as a tutorial, I would like to download the Figma one for me. Oh yeah, totally. Because I have the connector, but I'm pretty sure I need to install the MCP; it's a separate thing. Not sure exactly how it works, I'm pretty sure it should be almost identical, but I can show you either way. Like we'll step through it. I'll like all switch off.

Oh, there's one thing that I think everyone should download. It's... and I believe uh Louis, just in case I need to connect to anyone's computer to help, that'll be good. Yeah, I'm gonna refill my water real quick but I'll... I think Habla. Caron, me estoy dando cuenta que lo que me tienen haciendo en Lockheed, while it is technically relevant, is something you've never touched before. I'm paying attention and I don't understand, and I just like holy shit, there's so much shit here that I've never seen before in my life. What the fuck is a kilometer? Be right back. Thank you.

I mean they're speaking Cantonese to me, you know? Como que yo todavía no sé qué carajo un MCP en verdad yo no entiendo muy bien qué... I don't know anything, but I understand an MCP is like a plugin, basically for AI for accesses. Chrome MCP, do you know what MCP stands for? Because if I knew what it stood for maybe I would. There you go: Model Context Protocol. MCP is an open source. Okay, okay, now I understand. Exactly. All right, that makes a lot more sense now and the other shit starts clicking. I understand what they say when they say, "Oh yeah, spin up this instance and run the VM," all that shit. Coding language? Yeah, I mean it's not even coding; that's just like tools that we use. Right? Cause like, a VM is a virtual machine. That's basically a computer that you spin up inside of your computer. It's like a program. Imagínate como cuando tú abres Google Chrome. Uh-huh. Instead, you just open up Windows. Yes, that's a VM. That's a VM right? But then they say shit like, "Yeah, just get all these plugins and MCPs and shit," and I'm like, what the fuck is going on?

Yeah, I mean I think we'll see that a lot of... I think you muted Cam. Oh my bad. Yeah sorry, I was laughing at what you were saying. I think a lot of this AI configuration language is just rebranded concepts from like what you're already familiar with. I mean hooks, for example which we didn't even really go into are just like hooks. Yeah like webhooks right? Yeah, I mean basically like except instead of a network request response cycle on a web server it's the network prompt response cycle in an AI coding environment. So it's like okay kind of.

So I'm not going to make a set up hooks but just so everyone understands how this works: you open up Clip Cloud Code, the session starts, you send a prompt, and then inside these boxes you sort of have what will occur in like... it's called turns when you have your prompt AI response. This is the agenda loop. Pre-tool use is like before it calls an MCP or before it reads a file or before it does whatever. Permission request: like does it have permission to call this tool? Tool execution, sub-agents; we've looked at that a little bit. So what is it? A sub-agent? That's basically just another cloud instance that it spins up to go do a specific task. Yeah, to my understanding and they don't really elaborate too exactly what I'm looking for. I'll send this link though. But basically if I understand it correctly, sub-agents are like an additional model instance that's spawned using a shared context cache. Which is like if I go here into this conversation like all of this text up here both my messages and the model's responses are cached in context. So when I say "can you spawn a sub-agent to do this?" it'll keep all of that and start all that stuff in mind when it does. Yes right exactly, and so the subagent can go do that; it can go read a bunch of files and while it does that... Muddy? Oh you're back.

It reads those files, muddies up the sub-agent's context window, and reports back to the primary or main thread with a summary of everything the sub-agent did instead of flooding its own context window with that information. So it's mostly a way to divide and conquer and save context. This is mentioned in the documentation. All this stuff is more than what you need; all I'm going to show you today is how to use Cloud Code and how to use skills because those are like the two most important things in my mind. Everything else is fun or useful, oh, and MCPs—which I'll explain why they're a problem, but it's not going to be a problem for us today.

Without further ado, is everyone ready and looking at their computer? Yeah? Okay, great. Let's get started. I'm going to send over this right here; we're just going to work through this document. For starters, we're going to want to install Claude Code and Cloud Desktop on our computers. Mine hasn't been working, so I think I'm going to uninstall it and reinstall it. Does this need to be with Linux? For now, just do it on Windows and we'll revisit it. I'll send you the Windows PowerShell commands. Essentially, you'll just use a free account for now. Oh wait, I'll give you a guest pass; you could use the InfoRad account. Well, I'm worried right now because there are some reports that people are getting banned for sharing accounts across devices. So I'm going to say not right now, but Miguel is going to be using it during this session. We'll revisit that as soon as possible.

I'm also going to get my guest pass here; I'm going to give him a guest pass for Cloud Code. How do I install this on Windows? Use one of those two commands, yeah, and then you can also use this link to set up a guest account for Cloud Code. Should I set up the... And then do I use a team or enterprise plan? No, no, no. Just sign up through wherever that referral link takes you; it should just be like a normal onboarding. Gotcha. What if I already have a Cloud account? Oh, if you already have a Cloud account, great—just use that. Are you paying for it? No? Then use that referral link. Oh wait, I don't know if that'll work. Did you guys have to create your accounts in order to make use of those referral links? Miguel, Luis, did you have to create your accounts? I literally accessed the link and it made me create an account so I did it right then and there. Okay, I just clicked the link, said "Use for Free," and it redirected me to my account. Okay, I guess it should work. Yeah, I mean, it might not give you the free pro plan, but we'll make do. And worst-case scenario, we can give you hack access to the InfoRad account. It might actually be good to see if it'll work, but I do worry about us getting popped for it.

Wait, let me send you... give me one second; yeah, I'm gonna be right back. I already have all of this installed, so I'll be right back, okay? Sounds good. Yeah, can you give me a five? I just really need to do this thing for Sema like really fast. Yeah, go ahead. Thanks. All right, Jose, let's see if we can get you logged in here using this other... What am I? Shit out of luck. I don't have my computer; I don't have my laptop, which is where I have all the... Shit. It's okay. Just give me a second. I'll go more; I'll probably figure this out. Okay. Do you use Chrome, Jose? Firefox. Can you do like profiles on there? Yeah. Okay, so then I think I'll ask you to create a new profile. Neither. I'm gonna mute myself. So if you can create a new profile for this email, we'll get you logged in here. I know what OAuth is; I know you do. I'm just saying, I hope there's not some change in Claude Code that's going to... I gotta take my hair meds. Okay. So you want me to open up a new profile on Firefox with that? Yes, with that email. Yeah, one sec, my bad. I just needed Seva's ManyChat sign-in. You got it now? I think so. To approve that. And now I need to log in there. That's a... Where do I log in to? Oh, here we go. That's giving me a call. We're good. Sweet. Okay. I'm assuming you're not going to have to log in again to Google, but if you do, just let me know and I'll confirm again. Okay. Well, I mean, I just made a Mozilla Firefox profile. Okay. So then, yeah, that's... Why don't you go to Google or accounts.google.com or something? And we're gonna see if this totally boots everyone from it or not, which will be a good test. All right, I'm now getting a prompt for a Google update; find a cater 70-70 should be approved. Yep. Okay, great.

Okay, so then I think you're good there. What you're gonna need to do is... We got Cloud Code installed, I'm assuming, right? I need to install Cloud Code. Is Claw to Code apart from the desktop app? Yes. We're going to be using both. So Jose, you'll want to install one of these in this desktop app section. And then Luis, although I'm wondering if we might be able to just do it entirely from the desktop application too. Are you on Mac, Luis? No. Windows. Yeah. So Hosea, I'm going to suggest that you install the CLI version. I'm a little undecided here because I think Luis and Miguel might be able to benefit from the new user interface they've created for Cloud Code where you can use it as if you're running it from the terminal, and configuring the terminal as well if needed. Sure. Sorry. Okay, I'm installing the desktop app right now. All right, great. And Miguel, I know you've got that installed already, so that should be good.

And once you're installed, hopefully... again, I'm almost slightly worried that you guys both prompting at the same time it's gonna be a problem. But should I be logging in with the Info Systems account or should I be using mine?

I think for now, let's try yours for now. And then at the end, let's try signing into the Info Ad account because I just don't want to have us all messed up right in the beginning if we all get logged out simultaneously. Yeah. So we'll start there. And hopefully you're able to use Claude Code. The limits are a little close; I finished with set-off, so we're good to go. All right, great. Okay.

So, um, I think we'll first and foremost... are we all logged in? Uh, yeah, I just logged into the desktop app. I'm on the desktop app, yes. Okay, great. So I think the first thing that we want to do is create a folder for us to work from. We can create a folder anywhere and we'll sort of work from there. I'm going to work in my Flow folder, and if anyone has any questions on how to create one... I'll walk you guys through it. Setting a folder should not be very difficult, but um, it might be difficult for Louis; that's what I'm here for. Okay, so once everyone's got a folder, let me know. I'm ready. All right, ready? Mm-hmm.

Okay, so yeah, first and foremost, I'm going to go... we're navigating to there are three sections: Chat, Co-work, and Cloud Code, and we're navigating to the Code section. Does everyone have access to that, and can they see that? Yeah. Hold on, where is that at? So do you see my screen? Yes. I want the Cloud Code section. Should I share? Yeah, why don't you do that?

I'm wondering if it's because you're on the free plan and it's not allowing you to? Maybe. It's possible. No. Sure. All right, so this is what I get on my context menu: New Chat, Projects, Artifacts, and Customize... that hamburger menu at the top. What happens when you click that? It's blanked out. Okay, shit. Yeah, I'm thinking that we need to sign you in here. Yeah, also you get like a two-week free trial, but you can maybe sign up for the... Well, that... The guest pass didn't work; it seems. Ah, okay.

Yeah, so I mean, the options would be we create a new account in the Cloud Desktop using the same account on two devices? It pulls that pretty badly; it's been dropping so much lately. Yeah, it's been crazy. Okay, so some people are saying you can't... All right, there is an email now. Okay. Let's see. Google Alert. That works, and once you're signed in, let me know. I am in.

Okay, Miguel, can you send a prompt? What? Can you send a prompt to Claude? Anything—chat, message, whatever. You are through the... So that's essentially the magic link there. Alright. Let me know if it's okay with the Info account? Yes. Okay, let me sign in. Cause—Oh wait, are you using your own? Yeah. Okay, that's fine. That's actually better; your own is great. Yeah.

Okay, so this is what it looks like now. I have my Chat, and you've got your Code. Okay. Great, great. That's perfect. Okay. So... This is good. This is good. Okay. So now let's get everyone set up. We're going to want to be in Code, Jose. So move over there. Perfect. And so yeah, if we go to the bottom of the screen, we're going to want everything on local, which it appears you're already on, Jose. And then we're going to select that folder we created. So you can click that there. Yep, you're seeing Jose do it now.

And yeah, once we've got that folder set, we should be good. I got a little pop-up. It says, "Git for Windows is required to run local sessions. If it's already installed, set the Cloud Code Git Bash Pack environment variable." I have to download that? Yeah, I think... Although I don't know why I wouldn't say that in the overview stuff; it seems like a strange thing to overlook. And then... Oh, okay. Native window setups requires Git for Windows. So here's the link here. I'm pretty sure this is a... Oh, where do I put this though? Do I put that on the code file? It doesn't show up. I don't know if you're... Can you see my screen? I can just see the Chrome browser.

Okay, did you guys see the new Cloud Design tool? I'm sorry, I'm super distracted right now. I did see that. They also dropped; I'm like salivating a little bit. I haven't done anything with it, but I'm accessing it right now. I'm just going to get a little taste of it. Yeah, that's good, Luis. Yep. Oh, wait, actually, can you go back one? Yeah, it's fine.

Thanks, that was insightful. I mean, it's not going to make much of a difference on any front; I mean, all those other options are like if you're a fucking nerd, I guess. But you don't have any preferences on your Git, right? I don't even know what I get. It is... That's okay, you don't have to get it to get it. Um, okay, so you can finish. What? What happens when you ask, uh... Um... Can you fix my Git? Okay, cool. All right, sick, sick, okay, cool. Do I change the model? No, just keep it on 4.7. Okay.

Okay, so now we've got our Cloud Code set up. Um, why don't I start sharing my screen now? Um, I'll present here. Okay, so just so everyone understands at a high level how this works: Essentially, we've selected the folder that we're working in, and from there it's going to create a new folder. A Claude agent will be spawned in that directory. I always run it in bypass permissions; this is probably making Jose shutter a little bit, and if he's not shuddering, it's because he doesn't understand what this means. It's essentially turning off everything that would prevent Claude from performing an action without your permission. Um, so normally there's like a permission workflow, uh, or I guess sequence where the agent decides it needs to call a tool. And I'll turn it on for example here... for file collection. Let's see. Hopefully this will give me an example here shortly. Oh, shit.

I think you told us already, but running it through the app now, Sai, we have the same power as running it through the terminal or something else, right? Yes. Okay, that's amazing. Which is awesome and great for non-technical users because most people are just not familiar with this in the same way.

So here's what I was looking for: I wanted to get a... but this is what happens when you have permissions on. If I turn it on "bypass permissions," it won't ask me. Luis and Miguel, you're not familiar with some of the commands that are dangerous; it might not be as helpful. So I think it'll be in particular. It's really important to do a little bit of configuration here, which I'll sort of walk you through. One of the first and most important things you can do is initialize a `claude.md` file. There's a `/init` file or a slash init skill or command that you can use to do this. If your folder is empty, it's of no use to you currently because it's going to try and read all the files in your folder memory.

So I'm going to send this to you guys here just so I can understand why: Why would we need to initiate like a new `claude.md`? What is the function of it? That's a great question. Your `claude.md` file is essentially instructions about what you're doing in this folder.

Hey guys, I'm sorry; I had to walk away for a little minute. Can I get caught back up? Yeah. So real quick, what we were talking about is configuring context or memory files, as they're called, in any project that you're working in with Claude Code or this folder that we've created for us. You'll create a `claude.md` file. If we take a look at my... and I ask Claude to make that? Yes. One thing you'll want to note is what I was showing Miguel: there's a command called `/init` that you can use to create a new `claude.md` file, but if you don't have any files in your folder, it's not of any use to you yet. So it's something good to know. When I give you access to the repo, it'll be... but for now, just to keep things simple, we're just gonna focus on configuration and doing some fun little setup stuff here today. We won't go too deep into that, but just so you guys can see: in my `claude.md`, I've got some context about what we're doing and what the agent needs to do when working in this repository.

The start is like, "Make sure you're documenting heavily, write for my future self." And things like that. So it plays something there that said "fluence." What? I thought I read it too. Yeah, it says the fluence namespace is deprecated; do not use it for new packages. But as you see, it's instructions for the agent. Like, if you're in the `fluence` folder, you're doing something wrong. Or um, here this is dead ass wrong too: it should not be logged in on my work email doing this. What I've got a weird setup where I've got my work Cloudflare account and I've got my personal Cloudflare account, and I need to instruct the agent to say, "Okay, here's like the issue you're gonna run into and this is how you get around it." And that's what I'm telling it here. Like this is the API token; use that. Your `.env` file will pick it up. So it's just instructions like that. Just like help, especially if your agent makes a mistake once; it can be very helpful to write something in your `claude.md` so that it doesn't make that same mistake again. That's primarily what I'm using it for.

So for the time being, should I just ask it to make a blank project? Yeah. You can say, "Hey, can you create a `claude.md` file for my first Cloud Code project? I want this environment to be used for building out my Cloud Code configuration. I want to create some skills, sub-agents, and configure some MCPs." Write a `claude.md` to instruct it as effectively as possible to complete that. Something along the lines of that. I'm not going to tell you how to prompt it, but like at a high level, that's how I would approach it. And I would assume that whatever your first intuition is, is correct. That's sort of like the first core configuration thing that you can do. There's also sort of what they call memory files and rule files, which are also a way to instruct on like what it needs to do, what it's not allowed to do, things of that nature.

And you said MCPs, skills, and what else? NCP skills and sub-agents. And that should be just about all that you would need. Could you send me that link, Cam? Or did you send it already? I believe I did, but I'll send it. Okay. See, see... so I can read up a little bit more on that later. Yeah, I mean, right now this docs tree is the root of it, but all of these docs are very useful. I've read every single page in here at least once. Here, these are like all the configuration primitives you're calling for Cloud Code: your `claude.md`, your skills, MCPs, sub-agents. Do we have... I'm not exactly sure what the plugins marketplace looks like or if it's just in the... oh, it might be in a separate one.

I feel like generally we should maybe try to teach us how to build something simple, like maybe something that can be used efficiently in terms of staying on top of to-dos or something along those lines. So that way we can just kind of get our hands dirty and start understanding what the workflow feels like.

And then I feel like as we decide to start building stuff on our own, we can always cross-reference with you. It's like, "Hey, I'm having a problem here; how should I fix it?" etc. Yep, okay. So in that case, what I'm going to suggest we do is select the folder, and you can add other folders if you'd like, but we're going to start here. For now, I think what email will give us the ability to do a daily digest of the information that exists in our Google Workspace—in our calendar and in our to-dos or whatever—and create a daily digest of what is going on this week, what needs to get done, etc. We'll just use the info ad email for now; it's not relevant to everybody, but I'm hoping that this example will certainly... So first, let's see here.

I think the first thing we're gonna wanna do is go to manage your own connectors. If you go down here, does everyone see this at the bottom? If you go to the plus sign and then to Connectors, you can manage or add connectors. Sorry, Cam, is that connector the same thing as an MCP? Yeah, I mean, the distinction isn't super important, but basically a connector is like a remote MCP that's handled through Claude's user interface. It is shared across your desktop app, your web app, and your cloud code in your terminal. So all those connectors are accessible from those three environments. Oh, great. Whereas an MCP would only be configured for one environment. So for the sake of simplicity, we'll set up three connectors: Google Drive, which I'm adding as an example. Oh, shit, same problem since I'm on my personal account. Yeah, okay. I'm trying to think through a good workaround, but our options might be limited here. I just did it with my personal email; if it's for the sake of figuring out how to build it, it's not a huge job. Okay, well, let's do that then for the sake of simplicity. We'll set up the Google connectors for our personal emails and we can go from there.

So I should add a Gmail? Uh yes, I will. We'll add Google Drive, Google Calendar, and Gmail for our intent and purposes today. Let's get back. Should I be signing in as Flow Systems? Yeah, for now, just to keep things simple; it'll be good because you're connecting it for all of us. But yeah, here we'll continue working through. I'll add some more. Do you have the verification email or should I just sign into the email? I'll get it right here. Let's see. Thank you guys for working with me here. No, thank you; you're trying to teach a class to three people. Come on, right there, and we'll sort of do the same thing. I'm assuming it's not gonna have you magic link for all three of those, but if it does, we'll just work through each. And I think Google Calendar, Google Drive, and Gmail will be a good start.

And then when it says "Choose an account for Gmail," do I use my personal? Okay, sweet. Yeah, I'll save you here; I got my Gmail connected. Okay, great. Like I already had. I know I'm going on my own tangents, but I already have the Figma one downloaded, so I just haven't played around with it. So I guess we don't need to go into that well... We'll uh, I'll spend a little bit of time with you at the end to just like show you how that works. Um, so I'm assuming even if not everyone has all three set up, it's fine. Let's just go ahead and let's start a routine. What's the third one? It's Gmail, Google Drive, and what else? Calendar. Calendar, yeah. And the only reason why I would say get all three set up is because what we're going to set up next is going to have it instructed to go through all three. But we can modify that as needed. Okay, I got all three. Okay, cool.

So now if we go back to our quad desktops and we go to Routines, we can set up... I don't see that. Oh, this is not available on fucking Windows. Wait, wait, wait, wait. Yeah, I have Dispatch and Customize. Okay. If you go down here, it doesn't say like, if you click Customize. Yeah, I did that; I only have Dispatch and Customize. Okay. Yeah, I also don't have the Routines. It must be a Windows compatibility thing. Okay, in that case, we'll create a skill, and there are ways to sort of also schedule tasks through other mechanisms. But I'm basically going to show you how to fumble your way through a problem like this with Cloud Code. Because whenever I hit an issue like this—compatibility or configuration or what have you—I build around it. So with all that being said, I'll sort of kick off... Like what I would do when creating a skill. Um, although, let's actually add one plugin before we begin. If we go to the bottom here where we added connectors, let's say I'm gonna be pissed. No, I got it. Yeah, I got it. Okay, cool.

Yeah, so why not everyone download that or add that? It should become available to you in Cloud Code. And let me know if anyone hits a hiccup there. Skills skill to write a proper skill folder and... What the fuck? Okay. Bro, Cloud is so good. Like I got lost for a second; I didn't download the writing skill skill, and I'm just asking it to download it for me now. So I didn't even... They make your life so easy, Ro. Go on. It's truly crazy. And then, yeah, it's the skill that you're doing. The Rails ecosystem is also really cool because there are so many people participating in it and there are so many powerful and interesting ways that... just for everyone's information. Right now it's creating the skills folder, and in there it's going to create the skill.md file, and create the skills folder. This is sort of like what the skill.md file looks like; it'll have sort of just simple natural language to describe how it should be, how a certain workflow should be completed. You'll see here that you might add additional files with examples or scripts, references, or... that you could think of or anticipate. Okay, so here, this is the skill that it created for us; I think all of our skills will be different.

And then once this is set up, let me know if everyone has created the skill and if anyone had any issues there. Yes, I created it. It said to try it: start a new conversation and say "give me my daily digest." The description should trigger it. Okay, that's strange. All right, so this is cool. But for example, how would we go about having something like this be autonomous or recurring? Well, normally I would suggest using the native Teams routines feature, but in this case, it will be different for everyone depending on what device they're on. For example, in the same channel chat, you might say, "Can you please create a cron job to schedule this?" We'll see how it goes from there. What is this you sent us? Sorry, another prompt. Yep. So what I'm doing here is saying, "Can you please create a job or a scheduled job that your computer is going to schedule?" It might have to give... There was something I did; I got to remember how I did it. Yes, I believe there's some sort of Windows native app you can use. It's called Task Scheduler. Yeah, there we go. I believe that's what Luis mentioned. That's actually what it's telling me. So I gave it pretty much that prompt and said there are three options: You can do an in-session cron with the `--durable` flag to persist the schedule across CC restarts, but CC still has to be running at 8 AM for it to fire; or auto-expire after seven days where recurring jobs self-delete and need recreation (good only if you keep Claude open on this machine); or a OS-level Windows Task Scheduler running `claude-p run` the daily digest skill at 08:00. Truly headless, laptop awake, plus network, plus off tokens, fresh digested.

So that brings us to an option for us: we could do the `/schedule skill`, which is like, "Can you schedule this?" And it's loading that skill. So there are a couple of ways to skin a cat or bake a cake. Depending on... I'm a bit behind, but I'm keeping up; I'm putting in the prompt now for the cron job. What is cron? Cron is just like, I don't recall what it stands for, but it's basically a Linux utility that runs tasks on clock instead of you telling it to do something. It checks the clocks and sees, "Hey, it's 8 PM; it's exactly when the user told me to do something." So I'm gonna do that thing that the user told me to do. For me, since I'm on a Mac, it's still cron, right? I should be able to set a routine which I can go into. Yes, I am able to. Okay, okay. So I would just ask Claude, "Can you help me create a little prompt to set a routine to run this daily?" And that'll help for you, which is much nicer. What the frick is it talking about? Cam, I have a question; I'm gonna try that. What the hell is up with this though? I'm so confused. Are you using the Flow Systems account? Who? Me? Yeah, I know. Is it possible that... well, if you're not using it, then it doesn't matter. Yes, because I am currently running something on mine. I think this is... Yeah, but routines, hopefully they'll add this to Windows soon; it's basically just a way to set a cron job via their UI, which is really cool. Doing on the Windows machines here. But Miguel, if you have any questions, just let me know. It should be fairly straightforward. Yes, it is. I'm setting up a routine right now. Great. As you see here, I've got a little calendar; it's running things every day. It's got these instructions here. I think I got it. Can you ask it to test? Yeah, it's testing it right now. Okay, cool. My brother just got here. Cam, I'm a little curious: what's your dream? That was a test that I put together, although I'm not sure where it is. I would have expected it to be here somewhere; I thought I saw it in your hand.

So the Cloud Code has a feature called Dreaming, which is like a strange hidden feature where if you go into Cloud Code... Sometimes no, but if it works, it'll pop a little blue message that says "dreaming" or "do you dream?" This has worked sometimes and other times not. It does this review of the project that you're working in, and it basically fixes your memory files on a nightly basis and stuff. So I tried to see if I could do something similar and have it do like this meta-analysis regularly to see if it can improve things over time. I haven't been super impressed with the results, but it does do a lot of work; it is reading a lot of stuff and writing a lot of stuff. It might just be wasting my usage though; I'm not sure. And so anyways, everyone's routines or scheduled tasks are running all right? Did everyone see what the outputs of their skills looked like? Miguel, you're catching up or I'm setting up the routine but you can keep going; I'm still listening and keeping up. Okay, yeah, cool. So that's sort of at a high level how you would set up a routine. It's a relatively new feature and not as robust as schedule tasks in Cloud Code. So beyond that, maybe what we can do is... Not sure if that's the best way to go here, but I was kind of interested because you did say that you had like a bunch of stuff that you wanted to also give us so we could start maybe setting up so you can have some sort of knowledge base to go off for certain functions and whatnot. Yes, those would be the skills. Although... Do Wrangler setup or something here? Cam, I have a question before you keep going into that; it's already got everything ready for the routine but it's not letting me create it. I imagine it's because I haven't selected an environment, but it's not showing any options when I click on it. No, not even a local environment? No. Or repository?

Yeah, so when you click to... let me see your... Are you seeing that right here? I chose remote. Maybe that was the problem. Oh, that is the issue. Do not do it remote. All right. Sorry, I should have specified. No worries. Make sure you save your prompts because it might lose. I just copied it just in case. Sweet.

Okay, and then so I'm going to send you guys some docs here. I'm assuming Jose, you probably have Node on your computer, maybe, right? Probably not. I don't use this computer. If I have a Git Bash, is that good enough? Yes. Or well, I'm not sure if you'll be able to install it without Node unless you use like MPX to call Wrangler, which is like using it without installing it. When you say no, do you mean the package manager? Yes, and I guess the JS runtime. It's sort of like all in one.

It'll come. It's called... what was that? Oh yeah, Claude can do all this for you by the way—you know, you can just say, "Hey, can you..." That's a great opportunity here by the way: make it the start and shut up. What is it, Paul? What is it? Where do you want to go? And I fucking love that Claude can do everything for me. We're not... you know... typing like pilgrims. Yeah, hey, sorry, sorry—I'm kidding, I'm kidding. You're still typing in a prompt. Shut the fuck up. You're right, you're right. I mean, I assume the bots will be prompting for me and I'll be working at McDonald's. So here we go. That's if AI doesn't take that job too. Yeah, it's all just gonna be fucking computer screens and we'll be the underclass begging for scraps.

Right. So quick question: It told me to run the Daily Digest tool once. It says it's running it now. I'm wondering where it's going to end up. Is it going to come through the info email or... is it gonna come through to mine? So I think it's going to create a .md file in the folder that you have it working in. What I think we can do once we set this up would get you guys set up with Wrangler is we'll have it deploy a static asset deployment. I've seen this before and I'll share it with you too, Jose. But currently, this entire conversation has been recorded and I have... I'll have Claude Code after I get a transcription of a meeting go through and do—I'll show you right now, actually. It'll go through. That sounds like a baller ass fucking tool that I can definitely use.

Yeah, I mean, so my hope is to get you all set up with that. There's just... I don't know, I don't have a Windows machine to test on, and I could obviously get, like, VMware to test that stuff out. But, you know, once you guys are set up, my hope would be that you guys can just... like what was discussed—and it also does like a comparison against multiple models. I have local LLMs running on my computer that it can also use for the inference locally. And it's nice to, I love having these meeting transcriptions because I'm just a forgetful guy. But I mean, I'm hoping that you'll, once these pieces are in place, you can sort of see how this might work for you and what sort of... It is currently running. Okay, great. Anyone having trouble or not able to install Wrangler currently?

I'm good with it. Okay. So, yeah, well, once installed and/or logged in? I haven't installed already, but when you say login, should I just make an account for Cloudflare and connect it? It's either that or use the info at FlowSysSystems.live email and I can't connect that even though I'm on my personal account. And yes, yes—that shouldn't be an issue; that's just an issue with the Cloudflare connectors to my understanding. Okay, but could you send me the password? Yep, I'll send it in the chat. I promise I have better cybersecurity sense than this, Jose. It's all good, bro. When it comes to threat actors, we're not really the target anyway. Yeah. Uh, yeah, until they find out about how... how we're running our infrastructure or something. Um, yeah. More like until they find out how much money we make. Yeah, because that's all that really matters for them. And thankfully, not... I need to make an NPM account.

No, no, no, no. Do you want to share your screen? You told Claude to install Wrangler in Node? Or no? No, no, my bad—that was fixing something on the service automation. Yeah, so here, I'll send it all in one message. Just send this here, like word for word, bar for bar. In a new session? No, in a new session or existing session, it should be fine. Cam, it says email and password do not match. Are you signing into the correct email? Info add FlowSystems dot live? Info at flowsystems.live? Uh-huh. And you typed in... And the password I just copy-paste it. I mean that's exactly what I sent. You're sure it's that one, right? So maybe I'll... Yeah, can you share your screen? Oh, wait, no, no, no. Okay, sorry. What I'm saying is you need to use the Google OAuth. So try clicking that Google sign-in. Say, "And if you've got the info ad, that's the one you want to use." And don't save that. Yeah. And that's the password there. Sorry. No worries.

Hmm. Concept, yeah, do you understand? This... yeah, hopefully it's not too much of a certain degree. Yeah, yeah—it'll start clicking as we get to do a little bit more of this? Yeah. Like I said, the more I actually do, the more I'll learn. It's just for the time being, like hitting me with the fire hoses information; there's only so much I'm gonna retain. Yeah, so I'm hoping that the transcript will help with that. And then I think, I believe very strongly that this Cloudflare development flow is a great way to sort of like help the Claude Code magic click of it. So I'm on the right. I'm excited to sort of get your feel on this.

Okay, and question, Cam: This prompt that you've sent—that instead of writing an MD file locally, whatever... Could I paste that into my same routine? Would it work the same for... Yeah, you would just need to replace wherever it says the DonMD stuff. Stuff. Um, okay. And then, yeah, but you don't have to jump all the way there.

Let's make it able to log in with Wrangler first. I'll walk through the login process one more time from here. We'll do... Let me know if anyone has any trouble there. This guy's going on 10 minutes for the install. Holy shit, that's crazy. Okay, boys. Bye.

But I have a question. Yeah. Since I have a routine and the N-Style chat separately, they're not like... Well, yeah, so you might not be able to prompt it. What you might say? That's my question. Yeah, I think what I would suggest you do, or what I would do in that situation, is I would take this here and I would take it to a chat that I have. If you asked a chat to help you write that routine, um, you know, go there and say instead of writing a .md file, can you help me do this? Rewrite the routine. Okay, yeah, I'll just reference the routine and ask it to do that. We'll see it.

Luis, has your cloud code been able to install Wrangler? It's saying... okay, let me actually. If you want to share your screen. Yeah. Okay. Okay, yeah. So copy and paste that into a PowerShell and I'll walk you through it. This? Yeah, that. What's a PowerShell? You should type in search PowerShell. All right. In your... and this might also be your default browser, which is Edge, I'm assuming. So yeah, but I have it on my... Like I already have, like I'm logged in on, I'm on my other screen and it's logged in. Well, yeah, just copy and paste that URL and paste it. Yeah, so that link right there, can you try and like right click on it to copy it? The link. Are you able to copy it? The link from that page. You see it says... but yeah, can you copy it to your clipboard? Or like? Yeah, I have it. You got it? Now, but it's all good. Clicking always allow on everything is giving me PTSD. If you want to avoid that, you just turn on bypass permissions and uh deal with it in your nightmares. I mean, it's already giving me nightmares.

Um, okay, Luis, it doesn't seem to be fucking working. Um, let's, uh, let's stop it really quick before it goes and does something reckless. Um... open. Okay, I think it's done. Okay, sweet. And then do you want me to walk through the steps for the Wrangler or is Luis about to walk through it? So um all right, uh type in `wrangler login` in Cloud. Or yeah so you can do it in cloud with just going to... if it doesn't pop the link open it in a PowerShell and uh you'll type in `wrangler login` there. Um I'm hoping um I'm logging into Cloudflare now. Do I just use the Infosystems Live? Yes, yes. In just a second, I saw that. Same password? Yes, same password. Except make sure you're using Google OAuth. To clarify I'll be right back. Sorry guys.

How do I know if I'm using Google Earth? What oh it because you to sign into Cloudflare and you can sign into Cloudflare first. You see, yeah, this is what Luis is doing is exactly what you're trying to do which is you get logged into Cloudflare, you use Google to sign in, and once you're logged in, opening that Wrangler login URL is going to take you to this permission page. Could I, like in theory, now that we're tapped in here... um VERIFY NOW? Okay yeah it's just saying password required and then when I try to use a password it says it doesn't match okay let's go to power. Do you want me to stop streaming? Oh yeah, yeah. Jose, can you share your screen? Sure. Cool. So type in `Wrangler login`. Space login? Yes. Can you type in... I'm wondering what environment... I'll try to stop this guy. Hit escape on your computer. Or there's that square too on the right in the prompt bar. Yeah, just hit that. I mean, it's strange that Wrangler wouldn't be installed. Did it say that? That it installed it properly in PowerShell. It says here all three are installed and working: Node.js, NPM, and Wrangler. Okay. Oh wait. So I see what happened. Can you type in the prompt bar? Can you do exclamation point `Wrangler login` URL? See, I don't understand why I wouldn't error there. Do you know how to like reset or like reload the script? Like something that you're not familiar with and frankly well typically that would mean just running source um oh yeah so that's exactly what I'm thinking in my head I'm just thinking of Linux. Can you try running `source` and trying `Wrangler login` again? Sure. How do I stop this here? It would have pop the URL if it works. So don't even worry. It did take me to a site the first couple times. Oh, it did? Okay, we'll see. Yes, and then when I tried to log in there... When I tried to do a single sign-on, it said I needed a password, and then when I put in the password, it said it didn't match. Is there any way you could show me the window where it was doing that prompt again? Make sure you do the exclamation point. If I just run this, then it opens up the win. Okay yeah so it maybe it has your like source loaded in this bash environment. Let me know if it pops it. Oh and then can you ask it to run it again? Yep and so hit Google at the top and uh can you I mean I know this is not um the way The way to fix this typically in my experience is you need to make sure that your, the profile that you're using is the one you've had selected last so that like when... Okay, and then hopefully a prompt here. Yep okay just hit Review permissions authorized. Okay. Okay. Great. We are all set up now. Miguel, you're good too right? Did you sort that out? Is he good? Yeah. All right sweet.

Okay so now without further ado, for Luis and Jose this is going to be editing the skill that we've set. And for Miguel and you know myself it'll be modifying our routine. I got ahead of that too. Yeah so everything's working for me perfect. Great. Okay so we'll quickly knock out not the... I'm sharing still. OK. Cam, I have a question. Yeah. Run the last one. Sorry what was that?

The question is: how is this going to be making a deployed website each time? Isn't that gonna make a mess in the cloud? Fair, yeah, but that's why everything's gonna be named and we'll do some organizational stuff. But don't even worry about that; that's not your problem at the moment. Okay, if for any reason none of us use this stuff again, we'll just delete it like it's of no cost to us. Yeah, you can make a mess in Cloudflare as much as you like. And then... I guess this is just for the sake of practice. Really, I just wanted to... hooray for sycophantic AI! The sycophancy gets on my nerves real bad, but then suddenly they released this feature not too long ago called "Buddy," where they have these little Unicode art snarkiness that they coded it with. Okay, it shipped? It shipped? Okay. And then, does it have a Daily Digest? Luis Daily Digest, Miguel Daily Digest. I'm the only one with that one. Let's see, so everyone sees, "Oh wow, Miguel, why is he so sexy?" What the fuck? I actually did no input on the sign things; I imagine it's just from free. And also, it's not going to make a mess because every week it's just going to rewrite on the same URL, so we're not gonna have like a thousand Miguel Brief URLs. You know, "Yo, give me... I'll be back in 10 bro." I might think my room is fucking blackout drunk or something. Okay, that's the luck with that. Nice. Fucking retard. Be all right, bye.

Okay, so where do I see that? Yeah, so first, why don't you ask your code: "Where did you deploy?" I mean, I see that it's deployed and I've got your link right here, but I'm also trying to teach you how to work with your cloud to figure these things out too. It looks like it's still running something. Oh yeah, so maybe it just deployed first and now it's doing the actual daily digest piece, which would make sense given what I'm seeing here. Okay, now it's telling me where it is. Okay, great. Um, well... can you ask: "Can you run a real daily digest?" And you know, I'm doing daily digest. Great. Miguel, so just so you understand why this is important to you: this is exactly the same flow that we would use when deploying like Francisco's website for example. You know, like we create a worker for static asset deployments, which is exactly what it did here. And what it did was push this up to Cloudflare. So long as that's treated as the static asset—which in this case is like an .html file—I have a skill for you guys that I can send along.

A skill for what? It's a skill for deploying Cloudflare worker static asset deployments, probably. Like I said, a genuine question is why would you need a skill if it's like this; it's already deploying shit for me by just asking for it directly. Oh yeah, you know the skill... just so you understand like what skills are used for most commonly: it's for repeated workflows. So if I look at my skill here in my skill.md file, it's just specifying a couple of rules as far as like what would be expected when it's built again. Okay, okay, perfect. So it's basically like you already have that skill—a pre-written prompt with a bunch of instructions—so you don't have to type it super specifically each time. Mm-hmm. Yeah, you described it perfectly; that's exactly what it is. Okay, perfect. Yeah, so I'll send this along and then, whenever you have a website, you just like open up a new Cloud Code session and select the folder that the website's in with Cloud Code. Nice, okay. And that's what I would hope would be sort of like the general workflow when you need to deploy a Cloudflare.

I asked it to run: "Let me run digest again." Yeah, I'm wondering if there's something weird with like, are you running it via the task schedule? Or no, I'm running it through the routine. Okay, yeah, I'd be curious why it wouldn't be working, but yeah, I know we're transitioning so I work on the websites alone. But with this new Cloudflare ship, bro, revisions are going to be way easier. I've been seeing some features that they don't need to be like text-based; you always write something in, but I can literally circle "you're going to work on this only" so I don't have to prompt every time: "Don't touch anything else." Oh dude, you need to be so hyper specific. It uh... it made drafts in my email. Oh shit, what? Well, that must be an instruction issue. I'll be right back. Why don't you... I mean, so I believe the solution is it has something in the skill routine that you created that's confusing it about like what the deployment is; it created a draft in my email for no apparent reason. I'd say that and hopefully it fucking fixes it for you. That's what it is. It's interesting how it will interpret your instructions sometimes. Yeah, I mean it probably did it when it told me like "hey dude, I have three options to do all this shit" (this was before I installed the Node.js and stuff), so it probably did it then and it's like "okay, I'll just use your email instead." And type shit.

But I mean, honestly, like, it feels like this is mostly just prompt engineering, which is great because that means it's not going to be hard for me. Yeah, I know. I think you as an engineer are best equipped to use a tool like this because the language required to prompt an agent effectively—you are one of the most qualified individuals to be prompting in the first place. And yeah, it's a shame in my eyes, but I think it also sort of allows you to explore some exciting stuff outside the context of work. It's totally changed how I work; where I was like, my entire life revolved around my day job for so long.

And like this allowed me to be able to do so much more outside of my day job without having, um, without like needing an extra eight hours in every single day because I recently clawed it was my favorite model on this planet and some of my favorite tooling for AI development. And like very recently they started... Claude is still probably the best one. Yeah, that's why I didn't pivot before this meeting suddenly. As far as I understand it, the hope is that they'll get that shit under control here shortly.

I was reading too that that has been happening a lot recently at um okay Claude come on guys saw this video that Claude sometimes doesn't even read your prompt entirely. I think it has to do with this effort. You can say like its intelligence level and its effort level. With some new updates they lowered that at um I don't know I guess for less usage and shit at them. They recently axed the usage limits for all subscription plans recently to be able to handle the compute requirements for their most recent releases and supposedly this Mythos model release that they've got is like a big focus requiring a lot of resources, and that's like to blame partially.

Let's see. Okay, there we go. Got a little daily digest here. But do you think it's something that's gonna... Like, is this the new norm or is it gonna go back when Mythos is finally released? I think it's going to get back to normal but I just don't know how long it's gonna be because just by the nature of how compute constraints work for these AI labs, they basically have to decide a year in advance how much compute they have. And I mean if you've seen the Claude revenue growth over the last couple of months... I mean nobody, nobody in a million years could have predicted this sort of growth rate. I mean, they went from $1 billion to $30 billion in just over a year. Um, absolute insanity.

And you have to think in a year they have to predict exactly how many data centers they have to be able to service all of these inference needs across their API users, their enterprise teams, um, their uh, their subscription users, and then they need the compute for training new models as well. So it's this whole fucking... like circus that they're trying to figure this out. You can get it horribly wrong. They can overshoot, undershoot. I would not want to do their job. It sounds awesome, um, but yeah, so I, I, I have hope for them and I think they've done really, really amazing things. So I wouldn't expect them to keep fucking up like this. Um, but I've also felt a little bit of my hope dwindle as a reason. Thank you.

So, yeah. Real quick here, I mean it's looking like we've got... No, okay, it did it now. Okay. What did I miss? Not much, we were just looking. So if this is going to be the typical workday, I just prompt it until I get what I want is kind of what I'm doing. Yeah. Basically. All right. That's much easier. Now I'm not as worried. There we go. That was the moment I was hoping for. It probably feels a little foreign until you start getting in the weeds. Well, yeah. Like I said, the more I do, the more I'll learn. I learn better when I'm hands-on.

Yeah, I think, yeah, these upcoming projects that we have... and I figure there's a couple other like little fun things that we can do sort of as a what-have-you to enable all of us effectively. I think that we could set something up pretty easily, actually. If you say let's say that you as a user would send the info at systems email account an email every morning and say hey this is what we talked about last session... You can keep track of these action items using the AI. Oh basically send the information info at systems on email with action items for everybody, and then have Claude go and grab that action list and then divvy out the tasks for each person on a daily basis.

So basically we have a Trello board or a Kanban board of some type, and then you just populate it using Claude, run through and transcribe everything and create our like the transcription board, and then create the divvied up emails. I'll sort of set up everything, everyone's email aliases so they can see those action items go to each of the inboxes and stuff. Yep. And I also think that what we could do is kind of like reverse our daily digest: instead of polling our email accounts, we poll Info's, and then it sends us an email or some sort of like meeting calendar thing. Wow. It's amazing.

Anticipate that being difficult for any one of us here to do that... Um yeah for sure. I mean if you want uh when are we going to do this next? Like is there at any point going to be like a like I know this was just getting everybody up to speed these last couple meetings but uh do we have any plans like tasking? Are we currently working something? So we're definitely hoping to do this on like a try to strive towards like a weekly session like this, A, to sync up as a team, and B, to sort of... Like a small workshop. So basically just a weekly scrum. Yeah. And yeah, it would function both as a sync for the team and for us to sort of enable everybody and make sure that we're all working in the same direction and getting everyone to work up as time goes on and working on some of the stuff that we've discussed over the last couple of hours as far as like what we need.

I feel like now I'm set up to control. Yeah, I think so too. And now the balls are in my court to make sure that you have everything that you need to be able to... Like, I need to set up GitHub basically for our team. Like that's the biggest blocker that's holding you up and Gabby up. I'll need to make sure to get that done tonight. In that case, the email I sent you, the jmarques3000, that's the email I use for my GitHub as well. Well um I find it funny that Claude made a fucking action item for me to go buy a car. Um do you need a car? No, I don't. Oh OK. Claude, I don't know what the fuck you're talking about then.

It's just some guy sent me an email about a car I was looking at online, and I put in a little too much information, so they said, "Hey, buy it." And now they won't stop messaging me. I mean, hey, maybe it's time to support it? I don't know. I wish... It's okay. We'll get this AI money, and then we'll all get our matching sports cars. That'd be fantastic. Your color code is green.

Okay, so with all that being said, I know this is probably a little less productive than Luis was hoping. Want to go over Miguel? You specifically had a couple things. One quick question: Have I been the only one using the Infosystems account, or are we using this over multiple people? Currently, it's being used by you, one of our clients, and... Miguel has given access to it. Yeah, but I haven't... No, but I mean, like, right now in this session. Like, am I the only one prompting? Oh, yeah. There should be. Okay.

I would like to see if we can have multiple sessions across different devices, because now that I've set all of this up and I'm using that account, I'm scared that if I swap to my account, I won't be able to access any of this? Yeah, so the one thing about how things are set up is that... well, actually, there are a couple things. First and foremost, all of this should, I believe, be available through... the Claude Code. And maybe it's not okay, so what you can do—and I've had to do this before for different reasons—is ask any of your open sessions to write a `.md` file summarizing everything that happened in that session. Sort of just make sure I can transfer it to a different model and have that file available so that if you sign in on your account, I could transfer... Yeah, and it'll just be able to read all that, whatever.

I need to figure out exactly how to do this multilaterally tendency thing and if it's going to work. Um, so that's obviously on me too to figure out like what the best way for our team to go moving forward is. It might be that we need to get a couple of like team subscriptions for us as opposed to trying to do this max plan sharing thing. Why don't we ask Claude to write a script to create an empty file that will transfer it all for us anyway? There we go. See, you're already thinking like a prompt engineer now. Now we're thinking with magnets.

Yeah, so I mean, yeah, the context transfer is great and would be awesome to do across teams. The one thing is that we have no control over our client using this account, so like... I assume if it's going to be an issue, it's gonna be like the client using the same account as one of us. So I'll need to sort through that research; there's a lot of it. Yeah, I think if that's the case—if there is an account that is used by our clients—that should be solely a client account. We should not use it for development. Yeah, I know, I agree wholeheartedly. This is just sort of like... I was using my personal account on our clients in the interim and then I switched over to this one, and then obviously this came up and we wanted to train everybody so I wanted to make sure everyone could access Claude Code. But yeah, we'll get that. Yeah, for sure. We'll talk about it more at our next meeting, I guess.

And then I do have something on Sunday. I'm going to be on the play. And yeah, then we'll sort out these account issues and figure out the best way to go forward. The account also... To save you some time, come over if I... You know, I really appreciate that you're doing this. Come on, that he knows maybe coming in sitting down every Sunday might be too much since you're giving us basically a fucking class. So I think some other options would be: you could give us homework. Because I feel like a lot of the learning process, at least for me, has been trying to... you know, I need to fuck up some things. So if I just follow your things word by word, I'm obviously going to get it done perfectly. You know?

I also think that it should be more, like not so much "homework," it should be more like us leaning into ideas we might have and trying to build it. That's the homework, okay? So not gonna be like, "Oh, create this box, like this." It's fake-ass task; everything can be useful. You know, just... "Ah, guys, try to build a bot today that does this." Or you can say, "Everything should be personal." We're not down to this institutional structure that we need to follow things by the book. But he can give us some direction on things that might be helpful or might be good learning experiences, and we just go off of it and explore our own options.

I also think that it's important that we draw up a diagram of like how we should do things because, like what you said earlier: if the codebase is a mess and everything is spread around and everybody's using different accounts, then there's no real structure and that can lead to problems. So from the developer side, I would say the first thing we should probably get to is setting up a workflow—basically having everybody where they have their own tools and stuff like that, but maybe some shared resources are in place. And then any customer accounts remain customer accounts. And that way we have single dependencies. Yeah, I'm totally on the same page there. And we're... as much as I've said to fans, say the contrary—we're not far off from that. Yeah, I agree.

So what I'll do is: step one is going to be setting up the repo. I'll sort of create a new one—there's already an existing one—but I'll create a new one and that can sort of be like our shared canvas to create the new flow systems. And we'll sort of implement some structure. We can do the sort of development software development lifecycle thing in GitHub.

Obviously, Jose, Gabby, and myself were more familiar with that, so we could sort of begin putting the structure in place. Miguel and Luis will think of some ways that you can begin building out skills and workflows generally for us to use across the team. I'll send you guys some resources because what I want your next project to be is something like a long-running bot—similar to, although I don't know yet—I want to see if we can create personal bots that we can control from our phones.

I would love that; definitely what I want to get to. I do have a couple of things I already know I want to try working on. A lot of the endpoints for that type of stuff are going to be remote-controlled, like setting it up on my phone or someone else's phone so they can use it. Hopefully, we can cross that threshold soon.

I need to do some research into Windows compatibility stuff because that was honestly a footgun I wasn't expecting to hit here, though I should have seen it coming: I had no idea what the difference is in cloud code. To be fair, I do have a Mac. If I could actually have my computer on at all times working as a bot—leaving it on a set setting where it's not consuming a lot of energy but keeping it on—it would just be way more useful once I actually set it up. You know what I'm saying?

I also think Claude is such a powerful tool that you can kind of circumnavigate all that stuff with prompts. Totally. It was more like diminishing my class structure here, which honestly, um, I had a plan that sort of just completely veered off course as the first thing. So, I'm hoping we might be able to add a little bit more structure. Whenever another working session comes up, some of the ideas we're having and want to work through will require the terminal interface, which I think is probably going to be the last thing we need, like an "as long as this one."

But I do think that if we want our team communications to occur on Discord, then we have a Discord bot that we can message to kick off remote sessions using Cloud Code. Anyways, there's a lot of ideas and interesting things we can do. Obviously, we wanna make sure everyone is able to do as much as they can. Okay, I'll stop there. Questions?

Miguel, I know you've got some stuff that you want to go into. Honestly, man, I had a couple of questions, but seeing this whole cloud design now, Ted, it might bring a whole new approach. I had some other questions about workflows and how you went about the website for La Mosa, but I think I should explore cloud design first because I think that's going to be a workflow where you're gonna get the most bang for your buck for what you're doing.

Yes, exactly. And I already have the Figma MCP downloaded, so that was the other thing. Good with that? At them? So I'm good over here; I just need to explore this new tool. And I don't want you to get too caught up on like what we did for La Mosa. I think we already went through everything; it just might not be clear initially what we did and how it, you know... I know I brushed over it a little bit, but the daily digest is what we do for La Mosa. It's just the same process with different content. We're deploying.

Yeah, um, but what I'm gonna do, as promised—obviously delayed—I'm gonna put together some ideas for projects that we can work through this week to collaborate. Also, if you want us to read up on anything so you don't have to explain everything yourself: send it over. You know, anything like the cloud forums you sent earlier? Just send it our way so we can speed up time and the learning process with that, and also save some of your attention for them.

Yeah, I appreciate that. I did send a bunch of links over; I'll send them in WhatsApp again. Code docs are essential. Like I said, I've read them a bunch and still go to them constantly; it's a great resource. All these pages, I'll put you on game here really quick: you can copy the page and just put it directly into Claude. You don't even need to understand; you could just have it go or open Cloud Code directly. Of course, that's just broken though... um, I guess it worked? But yeah, I mean that's also a good way to go through these docs. Most docs for AI-forward companies are going to have those copy buttons that you can use. You know, stuff like that.

Yeah, hoping everyone feels a little bit better. Hopefully it feels better; we've got a lot of potential here. I think so too. Glad you agree. Anything else that we want to discuss or work through while we're all on this call? I think we're pretty good for today; we got a lot of stuff in. I'm going to take tomorrow morning and try to just start building something, then hit you up with my inevitable bottlenecks.

Please do succeed.

You know, and like that's how I learned. That's how you will learn. I promise you it'll go quicker, and you might end up getting some support sort of like... I don't know, maybe like Stockholm Syndrome with Claude, and begin to appreciate its company because, or whatever. But yeah, I'm excited for everyone here. I think this is going to be great for us. I think it will make us as a team better and work in the same way that a team of 50 people might work; we might be able to work with just five. So I'm very excited for what the future holds. Thank you guys for the crash course in AI. We'll get it popping, plus I put it to good use. Sir? I'll try. All right, boys. Thanks a lot, Cam. I really appreciate it. Yeah, I appreciate you coming. That's my pleasure, guys. And I promise I'll get all those resources sent over. I'll send over the transcripts so that everyone can review what we spoke about. Alright, sounds good. Okay, I'll talk to you all async.

System Audio:
Thank you. Bye.

Yeah, mostly Windows 10. That's the only thing I use. No conflicts, yep. At least on my part, I've always worked on subsystems and systems, so I'm not completely alien to all the issues that come up. I've already set up my environments to be pretty solid so I rarely get those issues, but yeah, they do give issues and they're fucking annoying. I've learned to deal with that shit, but yeah.

I mean, I've worked on Linux; I pretty much exclusively work on Linux at work, so I'm familiar with it. Do you know how any of that stuff reacts with VMs? It should be good. So mostly, like in the setup, I can help with the setup. It's not that hard. It's just where you're going to harbor; you need to harbor everything inside the subsystem for it to actually work better and your computer needs to be compatible with at least WSL2. If not, then it's deprecated to WSL 1. But once you enable the subsystem in Windows, you would need to restart your computer. You open your file explorer, click "This PC," and you see Linux at the end; everything is going to be harbored right there.

Then for your projects folder, you are going to create a shortcut on your Documents so that you can open any cursor or any IDE followed by the WSL path. Then you would need to route something so that the paths are for Linux instead of like mounting on the Windows OS. And that's how you're going to be avoiding those issues. But it's not that hard.

And then if you're going to SSH into a VM, it's going to have Linux, so you can connect it by any IDE. So you can host in Hostinger or Hetzner and you can have your VPS over there, and then you just SSH. That terminal is the Ubuntu terminal, so you can just rock on it; it wouldn't give you any issues in your Windows OS. It would just be on the VPS, right? So it should be fine. Yeah, I normally work like on two VPSs and my computer at the same time.

VPS being Virtual Private Server? Yeah, virtual private server where you have your VM inside and then you just SSH. So as long as you have the key and you can connect, you're good. The interface will just be your IDE and your computer, but all processes are over on the server, so you should be good. Alright, we'll definitely have to come back to that. Yeah, yeah, yeah, don't worry. The first time is a headache, but then it's okay.

I'm going to go ahead and grab the door. Something? Oh, what's up guys? So brother, I wish I was chilling. You wish you were chilling? No, I wasn't. I've been working on some server-side issues with a project on the company side, trying to resolve it. They blocked the server, so I'm doing some investigations, getting some evidence, and then submitting a request.

They make you work on Sundays? No, I choose it because I like this. They're not paying me or anything. That sucks. No, yeah, but it does suck, but at the same time, I like it, so fair. Yeah, I was just talking about how it would normally go: you need to activate the subsystem in the settings, make sure that it's enabled, and then download Ubuntu, and then you would get Linux in "This PC." That's where you need to path the projects or anything that's going to go there.

So if you normally have the issue come with your IDE, your IDE will be normally pathed to your MNTC, which is the Windows OS. You would need to rewrite that path so that it paths directly to the subsystem in Linux so when you open your IDE, your terminal is actually there. But to be honest, it's kind of aesthetic and a commodity because you don't need to do that; you can always open your IDE and then like navigate through.

You open PowerShell, right? You write WSL, and then you bash your way back to the root, and then you path your way to the Linux system. You can always write it. But if you don't want to do that repetitively, then yeah, you change the path for your IDE so your IDE can open directly into the subsystem, and then you're good. And then all dependencies—everything is fine. You just download it and you won't have any problems because you're harboring it inside the subsystem that is Linux. So you should be good.

Okay, yeah, no worries of course if it comes to it, then I can help out. Don't worry. Of course. Yeah. Containers, images. Yeah. Thank you. We're going to work on that later though. Actually, no. No. Hmm. Thank you. No, no, no, yeah, I'm sorry, I was...

No, you're not. No, you're not. Don't do that again. I'm sorry. I already muted myself, man. Come on. That's really nice. Nice. Thank you. Wow, I'm flattered. Thank you, Cam.

So currently, just to summarize and tidy everything up: We're using Wrangler CLI for deployment connected via Cloudflare. We have the different domains. We're doing processes for automation depending on what that process needs to be automated pertaining to the company or project that's been handed. The UI is being pushed out by Miguel. The backend code is being handled by us.

So basically, the company would get a page; they would get a deployable frontend and they would have their request in that frontend and their automation in that frontend that we create. It's mostly creating chatbots and processes or depending on again what the project requires. Yeah, I think obviously like it will get a little more complicated than that as we go, and I think that's fine. That's fine, yeah. You can continue. Mm-hmm.

Yeah, to be honest, like the remediation of organization is going to be subject to once the team grows, right? And then the processes that you do will ultimately be hardened when we try them ourselves.

And in terms of remediation, we could do a repo and then GitHub issues, where each project can get an issue for organization remediation. In the terms, right? That really doesn't bother me; I think that will ultimately come. But yeah, the stack is very important—the stack that you use. If you guys are using chatbots and everything, have you used Flowise or have you heard of Flowise or anything? Flowise is really nice. It allows you to create chats with AI agents on websites, and it's very seamless. It kind of works like n8n in a way; you have a canvas with different nodes and requests, and then you just put the system instructions for the chat and your bot for your page. Pretty easy setup, pretty cool too. You can check it out. I was using it... Yeah, you could have it self-served, like n8n, and run it via Docker container. You can have it self-hosted, and then it's free, but it should be open source. So let me check Flowise. Yeah, and it should be good. We were... Yeah.

I mean, definitely you know, obviously if for you, Gabby, and Falsa the hard way is going to end up being or not the quick way, but you guys are probably going to be able to actually make the best out of like Quad code and whatnot. But obviously Miguel and I are newbies here. So any type of tool that makes things a little simpler... Yeah, if I can share my screen real quick, if you don't mind? Do you have uh, can you give up my host permission? Uh, I think I can actually. Right.

So one of the things that I really like when I try to implement these things is Cursor; you can index right so I go to n8n dev documents directly from n8n and I go at a doc. Let me try it and do it right now because I don't want to read anything and I don't want to like learn that much, as I've had to do. It so I already know how to do it, but let's say that you don't and you don't have the time and you just want to do something with n8n. You go to the docs, edit n8n, and then you get the URL and you just do this. This was ClickUp one, so let me just copy again. And then you put it in, and then you put the name that you want, you click confirm. I can't use this because I already have a name like that, so whatever, I already have it here.

And then I create my agent based on this documentation. So then I have an agent that is versed in n8n documentation directly. And then I go to my MCPs and connect N8N MCP over here. Then I self-host it with ngrok; I create a container in Docker. I don't have to pay the monthly due, and I can have it self-hosted by Docker. And then I can use all the capabilities for n8n, and I don't need to pay for n8n. So it's completely free, which is really, really nifty.

So I set up with that, and then basically I say what I want to do; I create the PRD, and based on the PRD, then I would create the workflows. And the way that it works is that the agent is already versed on how the nodes should be and all the updates for n8n. So I just need to copy the JSON and paste it in the canvas, and all the nodes will be there. The only thing that I need to do is configure the nodes with personal credentials, and that's it. And I have my full automation workflow created in less than five minutes, which is really, really, really cool, really nifty. And it can be done the same with Flowise for chats and bots and anything that needs to be pushed like today.

If you're thinking about incorporating at an end, I would recommend this flow; it could help you. You can test it out and see how it works for you and everything. But yeah, ngrok, self-host, make it free, MCP, index doc, create agent, create workflow. That's it. Pretty simple, to be honest. Yeah. Probably, yeah. They work, man. They work. And that's all you need. Like stupidly simple, make it fucking work. That's it. That's the dream, you know?

Yeah, and if that comes to the moment, we can always serve it locally. So we could use Cloud Code or any coding CLI and have it serve locally so the data is not being pushed out. There's also a feature on Cloud Code where after 24 hours of account creation, you can put the private mode, and then no data that is used in your Cloud Code will be sent to the model so they can't be trained or anything. So it adds like a mini layer of safety. And the good thing about using the Cursor IDE is that you can link it with Cloud Code—that MCP for N8N, those indexing docs for n8n—and everything that you do in that IDE can be accessed via Cloud Code as well.

So you don't need to set it native inside Cloud Code; you can just set it in your IDE, link it, and then it's useful out of the box so you don't need to use Cursor for it; you just leverage Cursor, you get me? Which is pretty cool. So normally you would have a .cursor file, right? The same, and then I would link my dot cursor; I would symlink it to my dot Cloud Code environment, and then anything that I have in my Cloud Code, I can use in my Cursor chat. And anything that I have in my IDE, my Cloud Code can look into the source in my IDE and get them out as well.

Me neither, to be honest. I was just myself; I was looking at a wall and I just started like trying to reverse engineer it. And then I found that. So I completely get it because if you haven't hit that wall, very likely you wouldn't know. And I just mentioned it because I've been there, and it was fucking horrible, man. So yeah, something that helped me; of course, I would like to share that's just it.

Also, are you there? Yeah, I'm eating lunch. Sorry. You're keeping up, yeah? Yeah, I'm following. For sure.

Um, yeah, I mean I've barely ever used AI in the way you guys do, so I'm very new to this. I'm kept in a fucking locker for my workday. So, like, all the stuff you guys are talking about is a little bit outlandish to me to be fair. You're strange, but it's good, right? Not at all. For me, everything's exciting. It's good. I just remembered you said that one; you just reminded me that I need to do something right. Okay, the skills this Mama separate. Don't worry, it's not boring. It's just a lot to take in at once.

Yeah, so also, Jose, you know uh, I kind of wanted Cam and Gabby now so he can give you like the rundown on how he's kind of working on it. And, you know, I think ideally towards the end of this conversation with you too, you guys can start figuring out how you're going to share resources and how you guys are going to be able to kind of use these resources that Canvas created. Likewise, Gabby, if you have resources that you're gonna provide and so on and so forth. But afterwards, Cam is gonna sit down with Miguel and I, and we're just gonna go from like literally step one, like fucking setting up Cloud Code on the computer, and from there Cam just helping us get some resources and understanding how to use it so you can just stick around for that as well. So that way you can just like get started. You'll probably pick it up a lot faster than us, but you know, just for sure. And I usually learn better with like a hands-on approach; I'm more of a "learn as I go" kind of guy to be honest.

I find it great. I can definitely share some resources as well. I created my own status line for Cloud Code. I think it will be pretty good. I found one that was done in Python, but I reformatted it to Bash for faster execution. And I made a dynamic so if you change models, it tells which type of context you have left, and I made a dynamic per tokenization so it tells you also how much money you're paying or spending on that session and everything. And I can add other stuff as well. So I can share that; that will help a lot so we can monitor our sessions and see how we're doing in terms of context when we should compact, when we should delegate, or upload jobs as well. A few other things that I have in the locker room that I can take out if I don't get into trouble with my job, of course. But yeah, I can definitely help with that.

I don't think I will be needed for the setup. So I'll just... Yeah, so I'll just try to resolve this server issue that I'm having with Hetzner right now. But yeah, no, it sounds great. I got the understanding of what you guys are doing and how you guys are doing it. Later on we can talk about optimizations or environment setups and having a better repo. I would like to help if I can setting up GitHub and creating those CI/CD pipelines and creating those GitHub issues and PR requests as well. Yeah, just, we'll sit down. No, no, no, go ahead, go ahead, go ahead.

All right, we could do a Discord server as well, honestly. Yeah, yeah, which is really good too. There's also ClickUp; ClickUp is really good as well because ClickUp has all the formats for any type of document, and you can embed just HTML code, and it pops up as the page inside ClickUp. You have channels, you have workspaces. It's free. You can connect it by API key. If you need to put docs, excels, PowerPoint slides, any type of formatted presentation of data, even MP3, MD files or anything, ClickUp is really good as well. And yeah, it's free. Obviously, if we scale way bigger than there is a premium or paid version. But yeah, you guys can investigate ClickUp as well to see. You have Kanban boards and progress maps as well, coverage maps, a lot of things. It's very, very dense, very, very... but if you don't need it then we can just keep with Discord and WhatsApp.

But the good thing about ClickUp is that it works like Slack; it works like Discord as well. And you would have a centralized source of all documents pertinent to each project that you can modularize the projects in different workspaces, which promotes organization and everybody on the same page. I don't know, I just mentioned it because I use it at work and it really does help. Yeah, it does have an MCP server. There are some ones that are out but I normally use the API key because there are two MCP servers out there for ClickUp. There's one that's only with the premium account, which is I would say to stay away from that one. And the other one, you can use it super well with Cloud Code. But the reason why I would go through with API is that API can technically do everything that the MCP can do in a way. And it costs less tokens because that's the overall problem with the MCPs; MCPs drain usage and it's fucking insane.

There is an open repo, open source repo that converts a lot of—it converts 150 something MCPs to CLI—which is really, really good. And I've been also working to see if I can myself convert some MCPs to CLI. Which would be better for economics because an MCP can roughly spend around 800 tokens per tool call; it's only like 60 or 80. So it's a huge difference. So yeah, but the MCP, obviously, it's fucking awesome, but just keeping in mind that they can drain if... you know? Yeah, you know? Yeah, yeah, yeah, yeah.

100%. 100%. Oh, dude. Yeah. Yeah. Software is all acronyms anyway, so it's okay. Yeah, yeah. Easy use to that 100% if you ever—if you guys ever end up taking Cybersecurity Plus from CompTIA—it's literally just an acronyms test I believe. And if you do, I have the Quizlet that will have all of the acronyms. So let me know. Hell yeah, bro. Bro, there's a plug, there's a plug. Yeah, also quick little side thing: taken Cybersecurity Plus, I know NPM is a very common target for cyber hackers. They can basically poison the package registry and just put some malware where you were expecting, I don't know, Yum or some shit. Yeah, yeah.

I don't... For sure. But I mean, I think Cloudflare would probably handle that before we would have to, right? Yeah, I appreciate it. I need to know how the fuck this shit works so let's start that 100%. Yeah, this looks awesome, man. I just looked it up and it has API and everything. Hell yeah. Okay. All right, our baby steps training—that sounds great, man.

All right, so guys, I'm going to head out. Have fun. It's fucking fun. It is fun. So just try to have fun. I'll see you in the next one. All right. Take care. Take it easy. "Try" is the operative word in that sentence. Yeah, I have to poop. All right. All right. BRB. No, I'm good. I'm good.

Yeah, we'll go at it at your pace, but if we're going to download DMC MCPs—like just as a tutorial—I would like to download the Figma one for me because I have the connector, but I'm pretty sure I need to install the MCP. It's a separate thing. I'm not sure exactly how it works. All right. Okay. Okay. Sounds good. Mm-hmm. All right, bro. What the fuck is a kilometer? Miguel Taip. Si pa, habla.

I'm realizing that what they're doing in Lockheed while it is technically relevant is that you've never touched before. I'm paying attention and I don't understand. And I just like, holy shit, there's so much shit here that I've never seen before in my life. Look, you make me feel so much better. I don't know what an MCP is, but I understand an MCP is like a plugin basically for AI, for access. Like I have a Chromebook MCP if I knew what it stood for maybe I would have there. You go: Model Context Protocol. MCP is an open source. Okay. I understand. Exactly. Alright, that makes a lot more sense now and the other shit starts clicking. I understand what they say when they say "up this instance" and run the VM—all that coding language. Yeah, I mean it's not even coding; that's just like tools that we use right because like a very... A VM is a virtual machine. That's basically a computer that you spin up inside of your computer. It's like a program. Imagine, like when you open Google Chrome instead you just open up Windows. Si, ok. That's a VM, right? But then they say shit like "yeah, just get all these plugins MCPs and shit" and I'm like what the fuck is going on.

I think you're muted, Cameron. Well, I'm... Yeah, like webhooks, right? Okay. Kind of. So what is a survey agent? Is it basically just another cloud instance that it spins up to go do a specific task? It'll keep all of that and stuff all that stuff in mind when it does, right? Okay. Okay, I'm back. Okay, mine hasn't been working so I think I'm going to uninstall it and reinstall need to be with the Linux. Okay okay. Good news: could use the info account. How should I install this on Windows then? Do I just use those commands? Should I set up the account for... And then do I use a team or enterprise plan or? Gotcha. What if I already have a cloud account with my email? I literally, like I accessed the link and I made me create an account so I did it right then and there. Yeah, I just clicked the link and said "use for free" and then it just redirected me to my account. Wait, give me give me one sec I'm gonna be right back. I already have all this installed so I'll be right back. Okay.

Yeah, can you give me a five? I just... I really need to do this thing for semi like really fast. Thanks. I'm shit out of luck. I don't have my computer. I don't have my laptop, which is where I have all the... It's okay. Just give me a second. I'll go up while we do figure this out. Firefox. Yeah. I'm going to leave myself. I know what OAuth is. Okay. So you want me to open up a new profile on Firefox with that? Yeah, one sec, my bad. I just... I needed Seva's mini chat sign in. I think so. Three exclamation points. And now I need to log in there. Where do I log in to that? We're good. Okay. Well, I mean, I just made a Mozilla, like a Firefox profile. I haven't really touched any of it. Okay, and then... All right.

I'm now getting a prompt for a Google authentication. 70. I need to install Claw to Code like apart from the desktop app. Okay. No, I'm on Windows. Sure. Okay, I'm installing the desktop app right now. Should I be logging in with the info systems or should I be using mine? Okay, I finished with Seva so we're good to go. Yeah, just logged into the desktop app. It might be difficult for Luis but... I'm ready. Yeah. Hold on, where is that at? Yes. On the code section, yeah. No. Can I share? Here all right so this is what I get: only context menu, I just get new chat, projects, artifacts and Spyro edit. Yeah, so you get like a two week free trial, you can maybe sign it up for the or don't sure. It pulls that pretty bad. Been dropping so much shit lately. Okay. Okay. All right, there is an email now verification code. All right. I am in account okay let me sign it in because I'm sorry yeah Yeah. Okay, so this is what it looks like now. I have my chat and... But I got a little pop-up that says, "Git for Windows is required to run global sessions. If it's already installed, set the Cloud Code Git Bash pass environment variable." That was done. I have to download that. Okay. Where do I put this though? Do I put that on the code file? It doesn't show up. I don't know if you're... Can you see my screen? How does it stop sharing? You guys see the new Cloud design tool?

I'm sorry, I'm super distracted right now. I'm like, I don't know a little bit. I haven't done anything with it but I'm accessing it right now I'm just okay so look get a little taste of it. Thank you. Thanks, that was insightful. I don't even know what I get it is. Okay. Do I change the model? Oh God. You told us already but running it through the app now say we have the same power as running it through the terminal or something else right okay that's amazing. I'm sorry, just... So I can understand why would we need to initiate a new CloudMD file? What is the function of it? Thank you. Hey guys, I'm sorry. I had to walk away for a little minute. Can I get caught back up? I'm sorry.

And I asked Claude to make that. I see something there that said Florence; I thought I read it too. So for the time being, should I just ask it to make a blank project? And you mentioned MCPs and skills—what else could you send me that link, Cam? Or did you send it already? Okay, okay. See, I'm just... so I can read up a little bit more on that later. Mm-hmm.

Okay, I feel like generally we should maybe try to teach us how to build something simple, like maybe something that can be used efficiently in terms of staying on top of to-dos or something along those lines. That way we can just kind of get our hands dirty and start understanding what the workflow feels like. Then, as we decide to start building stuff on our own, we can always cross-reference with you: "Hey, I'm having a problem here; how should I fix it?" etc. Thank you. Mm-hmm.

Sorry, Cam, is that connected the same thing as an MCP? Okay. Okay, I have the same problem; I'm on my personal account and just did it with my personal email for the sake of figuring out how to build it—it's not a huge deal. Should I add a Gmail? Should I be signing in as Flow Systems? Do you have the verification email or should I just sign into the email? No, thank you. You're trying to teach a class to three people; I wanted to say, I really appreciate it. And then when it says "choose an account for Gmail," do I use my personal one? Yeah, I'm more connected. Yeah, same here; I got my Gmail connected.

And it seems like I already have—I know I'm going on my own tangents—but I already have the Figma one downloaded. So I just haven't played around with it yet. So I guess we don't need to go into that. What's... Sorry, what's the third one? It's Gmail, Google Drive, and what else? Calendar. Okay, okay, okay; I got all three. That's weird. I don't see that. Wait, wait, wait—yeah, I have Dispatch and Customize. Yeah, I did that. I only have Dispatch and Custom Acts. I also don't have the routines. My god, yeah, I got it. Mm-hmm.

I don't know; I got lost for a second. I didn't download the writing skill, and I'm just asking it to download it for me now. So I didn't even monitor that. They make your life so easy, Rogue. Oh man. Yeah, I created it. It said, "Uh, try it: start a new conversation, say 'give me my daily digest.' The description should trigger it." Okay.

All right, so then this is cool. But for example, how would we go into having something like this being autonomous or recurring? And what was that you sent us? Sorry, ah, another prompt. Okay. There was something I did; I've got to remember how I did it. I'll test the scheduler—that's actually what it's telling me. I gave it pretty much that prompt and said: "There are three options." You can do an in-session cron job with a `--durable` flag, which persists the schedule across CC restarts, but CC still has to be running at AM for it to fire. Also, auto-expires after seven days; recurring jobs self-delete and need recreation. Good only if you keep Qwen open on this machine. Option B is OS-level Windows Task Scheduler running `claude p run` the daily digest skill at 08:00—truly headless laptop awake plus network plus off tokens, fresh digest delivered with high uptime; opening anything output goes to a file or email. This is what "headless" means literally. Recommended that one.

So straight out of that, I have a question behind me—I'm keeping up but uh, I'm putting in the problem now for the cron job: What is cron? Cron is a Linux utility that basically runs tasks on clock instead of you telling it to do something; it checks the clock and sees, "Hey, it's 8 p.m. It's exactly when the user told me to do something." I'm going to do that thing that the user told me to do. Okay. Me: I'm on Mac; is it still cron? Right, I should... Yeah, I'm able to. Okay, okay, perfect. I'm going to try that.

Are you using the Flow Systems account? Yeah. Is it possible that, well, if you're not using it, then it doesn't matter, because I'm currently running something on mine. Yeah, it is; I'm setting up a routine right now. I think I got it. Yeah, it's testing it right now. Cam, I'm a little curious: what's your dream, or a little dream routine? Yeah, I thought I saw it in your... Ah, all right. Okay, yep. I'm setting up the routine, but you can keep going; I'm still listening and keeping up. So everything's working.

I was kind of interested because you did say that you had like a bunch of stuff that you wanted to also give us so we could maybe start setting up some sort of base to go off for certain functions and whatnot. Cam, I have a question before you keep going into that: I already have everything ready for the routine, but it's not letting me create it. I imagine it's because I haven't selected an environment, but it's not showing any options when I click on it for environments or repository at the... Let me see your remote; maybe that was a problem. Okay, okay, no worries. Yeah, I just copied it. Just in case. Probably not.

I don't use this computer for work. If I have Git Bash, is that good enough? As Node, when you say Node, do you mean the package manager? It's fun. And I fucking love that Claude can do everything for me. Hey. Hey. Hey. Hey. Hey. You're still typing in a prompt; shut the fuck up. That's if AI doesn't take that job too. Right. So quick question: I ran it, and it told me to run the Daily Digest tool once. It says it's running it now. I'm wondering where it's gonna end up—is it gonna come through the info email or is it gonna come through... Well, that sounds like a baller-ass fucking tool that I could definitely use. It is currently running; I'm good with it. Thank you.

I have it installed already, but all when you say login should I just like make an account for Clockfair and connect it? And I can't connect it even though I'm on my personal account. Okay, but could you send me the password? It's all good, bro. When it comes to threat actors, we're not really the target anyway, like until they find out how much money we make because that's all that it really matters for them. Do I need to make an NPM account? No, no, no, my bad. On service automation in a new session. Cam, it says email and password do not match at info@flowsystems.live. Uh-huh. And the password I just copy-pasted it. You're sure it's that one right? So maybe I'll wait. Yeah, wait. Let me try it. Hmm, silly me. All right, no worries... Jonstad? Yeah? Are you understanding this? Yeah. To a certain degree, yeah. Like I said, the more I actually do, the more I'll learn. It's just for the time being like hitting me with the firehose of information; there's only so much I'm going to receive. Okay, and question Cam: This prompt that you've sent—instead of writing an MD file locally, whatever—I paste that into my same routine, and would it work the same? Okay. This guy's going on 10 minutes for the install. Okay, boys, but I have a question since I have a routine and the nstyle chat separately, they're not like... Okay, yes, I just like to ask my question. Okay, okay, yeah, I'll just reference the routine and ask it to do that. It's saying let me actually... Well, there are... this what's the PowerShell? Yeah, but I have it on my like I already have it on my screen. Have like I'm logged in on my other screen and it's logged in. So here, the link from that page. I have it. Yeah, I have it. It just took me back to a login. Clicking always allow on everything is giving me PTSD. Yeah, I mean, it's already given me nightmares. Okay, I think it's done. All right. In Cloud. I'm logging into Cloudflare now using the flowsystems.live same password. How do I know if I'm using Google Earth? Like, could I... uh, so like in theory now that we're tapped in here um Oh, he left. Verify now. It's just saying password required and then when I try to use the password it says it doesn't match. Do you want me to stop streaming? Space login. Also, how do I stop this guy? It says here all three are installed and working: Node.js, NPM, and Wrangler. Well, typically that would mean just running source. Sure, how do I stop this here that says wait it did take me to... it did take me to a site the first couple times yes, and then when I tried to log in there it's when I tried to do a single sign-on it said I needed a password and then when I put in the password it said it didn't match. Here if I just run this, then it opens up the page. Okay. Yeah. I got ahead of that too. Yeah, so everything's working for me perfect. Okay Cam, I have a question. Run the last one. Just run the last one. Last command that you sent. Okay, the question is if this is going to be making like, you know, this deployed website each time, isn't that going to make a mess in Cloudflare? Okay. Okay, perfect. Yeah, I guess this is just for the sake of practice. Hooray for sycophantic AI. Okay, it shipped. I actually did no input on the signing things. I imagine it's just like from previous shit I thought bloated. So, and also it's not going to make a mess. I just talked with it because it's every week it's just going to rewrite on the same URL. Not gonna have like a thousand Miguel Briefs URLs you know yo give me I'll be back in time bro. I think my roommates have felt blackout drunk or something yeah I think retarded. So where do I see that? Right. It looks like it's still running something. Okay, now it's telling me where it is. Okay um it's doing daily digest now Mm-hmm. All right. What skill for what? Probably. Pedal. It's a genuine question. Why would you need a skill if it's already deploying shit by just me asking for it like directly so what what they still add you know Aha. All good. Okay. Okay, perfect. CC, so it's like basically like you already have like that skill is like a pre-written prompt with a bunch of instructions so you don't have to type it like super specifically each time. Okay. Perfect. Mm-hmm. Okay. Mm-hmm. Still nothing I asked it to run. Let me run digest again. No, I'm running it through the routine. I know we're transitioning so I work on the websites alone, but with this new cloud design shit bro revisions are going to be way easier. I've been seeing some features that they don't need to be like even text-based. Why not? You always write something in, but I can literally circle like you're going to work on this only. So I don't have to prompt every time. Don't touch anything else. Come on dude. Yeah. Need to be so hyper specific it uh it made drafts in my email yeah I'll be right back. Yeah I mean it probably it was probably when it told me like hey dude I have three options to do all this shit. This was before I installed the Node.js and shit. So it probably did it then. And it's like okay I'll just use your email instead. Okay. Yeah but I mean honestly like it feels like this is mostly just prompt engineering which is great going to be hard for me but um If only they let us use it. At least from what my friends tell me at work uh Claude is still probably I was reading too that that has been happening a lot recently recently at them okay claw come on guys saw this video that cloud sometimes doesn't even read your prompt entirely it it it I think it has to do with this effort. You can say like it's intelligence level and it's effort level. And with some new updates, they load it forward that at them I don't know I guess for less usage and shit at them on the Mm-hmm. Do you think it's something that's going to... like is this the new norm or is it gonna go back when Mythos is finally released? Yeah. Hey, my death is huge. I got it.

Hello. Okay, it did it now. What did I miss? Okay, this is going to be the typical workday. Like I just prompted until I get what I want is kind of what I'm doing. Oh, all right. That's much easier. Now I'm not as worried. Sure. Well, yeah, yeah. Like I said, the more I do, the more I'll learn. I learn better when I'm hands-on.

I think that we could set something up pretty easily actually. If you say, let's say that you as a user would send the info at systems email account an email every morning and say, "This is what we talked about last session." You can keep track of these action items using the AI. Basically, send the info at systems on email with action items for everybody and then have Claude go and grab that action list and then divvy out the tasks for each person on a daily basis. Basically, we have a Trello board or a Kanban board of some type, and then you just populate it using Claude. Yep. And I also think that what we could do is kind of like reverse our daily digest to instead of pulling from our email account, we pull info and then it sends us an email or some sort of like meeting calendar thing.

Yeah, for sure. I mean if you want—when are we going to do this next? Like, is there at any point going to be a... Like, I know this was just getting everybody up to speed these last couple meetings, but do we have any planned tasking? Are we currently working on something okay? Like a small workshop. So basically just a weekly scrum. Got you. Okay, I feel like now I'm set up to contribute. In that case, the email I sent you, jmarques2000, that's the email I use for my GitHub as well.

I find it funny that Claude made a fucking action item for me to go buy a car. No, I don't. It's just some guy sent me an email about a car I was looking at online and I put in a little too much information and they said, "Hey, buy it." Now they won't stop messaging me. I wish. That'd be fantastic. Color code is green. Awesome.

One quick question: So have I been the only one using the Infosystems account? Using this over multiple people? Yeah, but I have it. No, but I mean like right now in this session, like am I the only one prompting? Okay, I would like to see if we can have multiple sessions across different devices because now that I've set all of this up and I'm using that account, I'm scared that if I swap to my account, I won't be able to access it. And I can transfer it to a different model. I've seen... Okay. I can transfer it. Why don't we ask Claude to write a script together? Create an empty file that will transfer it all for us anyway. Now we're thinking with magnets.

Yeah, I think if that's the case, if there is an account that is used by our clients, then that should be solely a client account and use it for development. Yeah, for sure. We'll talk about it more at our next meeting, I guess. Right. And also, to save you some time... I really appreciate that you're doing this. Come on, get he—you know maybe come again sitting down every Sunday might be too much since you're giving us basically a class. So come on, I think some other options would be: come on, you could give us homework so we actually learn because I feel like a lot of the learning process, at least for me, has been trying to, you know, I need to fuck up some things. So if I just follow your things word by word, I'm obviously going to get it done perfectly.

I also think that it should be more like not so much like homework; it should be more like us leaning into ideas we might have and trying to build it. Well, that's the homework. It's not going to be like, "Oh great, this bot, like this fake ass task." Like everything can be useful, you know? Come on guys, try to build a bot today that does this. Or come on, or you can say everything should be personalized. You know, we're not down to this institution structure that we need to follow things by the dot, but just he can give us some direction on things that might be helpful or might be good learning experiences. And we just go off of it and we explore our own options.

So I also think that it's important that we, you know, draw up that diagram of how we should do things. Because it's like what you said earlier: if the code base is a mess and everything else is spread around and everybody's using different accounts, then there's no real structure and that can lead to problems. So from the developer side I would say the first thing we should probably get to is setting up a workflow. Basically having everybody have their own account where they have their own tools and stuff like that. But maybe some shared resources are in place. And then any customer accounts remain customer accounts. And that way we have single... Yeah, I agree. Yeah. Thank you. Thank you.

I would love that. Like that's definitely what I want to get to. Of things that I like, I already know that I want to try like start working on and and I feel like a lot of the endpoints for that type of stuff is going to be like being like remote controlled like I can set it up on my phone or I can set up on someone else's phone they can use it. So yeah hopefully yeah we can cross that threshold. I mean, to be fair, like I do have a Mac, you know, it's just like, I feel like if I can actually have this be like if I can have my computer on at all times actually working as a bot and like I can, you know, I can just leave it on on a setting where it's like not consuming a lot of energy, but I can always keep it on. It'll just be way more useful once I actually set it up. You know what I'm saying?

I also think that Claude is such a powerful tool that you can kind of circumnavigate all that stuff with prompts. Thank you. See, Pero honestly Man, I had a couple questions, but seeing now this whole claw design, Ted, this might bring a whole new approach.

You know, I had some other questions regarding workflows and how you went about the website for La Moza. I think I should explore Cloud Design first because it's going to be what you get the most bang for your buck for what you're doing. Yes, and I already have the Figma MCP downloaded, so that was the other thing; I'm good with that. So, I'm good over here; I just need to explore this new tool.

All right, okay. So basically, it's the same process. And also, if you want us to read up on anything—so you don't have to explain everything yourself, oh my god—but please send it over. You know, anything like the cloud forms you sent earlier or anything like that; just send it our way so we can speed up the time and learning process with that and also save some of your time.

Okay, I think we've got a lot of potential; I think we're pretty good for today. We got a lot of stuff in. I think I'm going to take tomorrow morning to try to just start building something, and then I'm just gonna hit you up with my inevitable bottlenecks. Thank you guys for the crash course on AI; yeah, we'll get it popping and also put it to good use. I'll try all right, boys. I appreciate you coming.

Executive Summary

Cameron led a 3+ hour Flow Systems team working session with Gabby, Jose, Luis, and Miguel, covering infrastructure overview (Cloudflare Workers, Wrangler CLI, monorepo structure), AI development tooling (Claude Code, skills, sub-agents, MCPs), and hands-on onboarding. The team encountered Windows compatibility issues with Claude Code routines and Wrangler login, but successfully got everyone set up with Claude Code desktop, Google connectors (Gmail, Drive, Calendar), and a daily digest skill deployed as a Cloudflare static asset. Gabby contributed expertise on WSL setup, Cursor-Claude Code symlinking, N8N/Flowise self-hosting via Docker, MCP-to-CLI conversion for token savings, and ClickUp for project management. The session concluded with alignment on weekly scrums, GitHub repo sharing, structured homework-style learning, and separating client accounts from development accounts.

Action Items