Remote Coding Agents Write 90% of Code for Companies Already....

What am I looking at? You're looking at a terminal screen. It shows the output of commands related to Docker Compose, specifically the process of building and starting. Cool. So, I've been getting ready to deploy violian, so I needed to learn some of that. And the Gemini thing is actually pretty helpful with little Linux commands that I don't know. But that's not why you're here. You're here to learn about background agents, remote agents. These agents, I've been talking about them for months on this YouTube channel. Months. I've only been on here for two months, I think. But they're here. They're coming out. We see them in Cursor. We see them with Claude Code. We see them in Augment, which is the one that I'm most bullish on. So, what is going on with them? Why would you use them? There's a lot of different areas of a codebase, right? And I think we should be mapping them to the different job responsibilities that people do. So, this is a massive task list with a bunch of context in it from augment. And I pair that with a bunch of prompts that I have tagged in the GitHub instructions booklet. And if you want to get access to that repo, you just go to the VI withAI repository, click on AISDLC. You can find the prompts there totally free under prompts. Right? So with augment, I've basically made this massive list, right? And it's ticking through here. And I'm doing this manually. I have this as a theme on my channel where if you're trying to automate something, you have to do it manually first because you can't audit the steps. And you should not automate something that maybe has really bad process steps within it. And you also just won't know anything about what you're doing if you're trying to automate something. If I try to automate this YouTube channel right now, it would be a lot better than if I tried to automate it two months ago before I did any of this. So I'm trying to figure out having spent just one day of working with augment on this particular piece of work, how would I split this out into remote agents moving forward. So this is a good example where at the start we had a very different architecture to what we have now. And so what we had before was a bunch of different pieces that were fun to play with but had horrible experience when working with the remote agents. So let's just think about it. Before I had set up Astro as a marketing site and then let's call it I had a Discord bot that was called Val. This is running on node. Just a little simple script. This is your front end, literally just marketing. And then I had an express server that was serving API requests. And then I had Stripe. These are the third parties. Discord. And that's more or less it. So these would be your third parties. This would be your API. This would be a little server. And we had them all in these buckets. And what I quickly realized was that while this was fun to build, it didn't make sense because as soon as I gave this stack to Oh, sorry. I actually missed something here. I have an application shell, which is actually like the app, right? And that was in tan stack start. And the issue with this is while I had a good DX making it, as soon as I gave it to any sort of Agentics stuff, any sort of LLM, it would have no clue what's going on. And so I decided to prioritize what the market likes, which is Nex.js and fast API to swap those out because the market ultimately dictates the popularity of something. The popularity of something dictates the behavior of the LLM. So I went and I watched a bunch of interviews about the guy from Versale. He says, I wrote this down, but I didn't quite write it there, but how can we build the process flows for streamlining from idea to production? And so for me, my systems are agents. So with remote agents, I decided, okay, next is going to be the thin front end because it's going to be great at that. And you also have V0 which has a model that you can use. Then we're going to use fast API as our core API. We'll still have when I say thin, it means we can do API stuff here. So this will be like userf facing and then these will be the kind of AI slash any sort of like processing. So these two and then if I wanted to I could add astro but no. So then what I'm thinking is in this case where I went from this to this what jobs were part of that? So you can start to map them out because the end state is that you have a bunch of remote agents that are trained on different context and have different job responsibilities, different access to rights on certain folders and not others and the ability to just have readonly stuff. So in the case that I just did that migration, there was some fast API, there is some more fast API, there's some more fast API, there was some next work. But the way I'm thinking about this is how does this map to your traditional organization. So on a product team, let's assume the following because they're different everywhere, but you'd have your PM and they'd be technical. You'd have your front end, maybe two of those. You'd have your back end, maybe two of those. And then you might have an architect that likes rotates between teams. Let's say that we have these and then there's a designer that floats between teams. So, what I'm just trying to do is assign remote agents to each one of these. And then that job description will be a part of their prompt. And what's beautiful about this is you can write your own scripts. That is not it. You can write your own startup scripts within the background agent section. I'm just getting into this now. I've been reading through their docs and I'll have a more in-depth video tomorrow on the main channel. That looks nice. But what I can do is if I did new and then new remote agent and I pop that open and let's make this a little bit bigger for someone yells at me. But I would tie the job responsibilities of what I expect them to do to a branch. So branches would no longer be ephemeral in this case. But I would spin that up. It starts on a Docker container. And then that script will also include a prompt what it has access to, what it does not, following the same patterns from Ader. That's where I think this goes. So it's important that you Great. I think I just changed my SSH keys. Maybe that's affecting it. But that's what I'm thinking of doing. So in the future when I do have a big piece of work like this, which is a migration of the full stack, then I can just put it on remote agents to go do it. Now, I didn't want to do this with remote agents now because it just covered so many different things. And I made this analogy on our Discord today where we're having an impromptu chat about augment and remote agents and all this. And my idea is that it's tough to trade off between when you're building the car, which in this case is the SAS product or the client project or the work project that you're working on, and then when to work on the factory, which is how are the cars getting made? So, I'm trying to split 25% of my time to working on the factory, which would be workflow, because if I skew way too far into just the workflow, then you don't actually do anything. You're just building like free dev tools and patterns. But I was amazed at how good this did at taking my prompts and then building out this full migration all the way down to actually deploying it on Docker on side of my VPS. So, that's it for today's video when it comes to the lesson. Now, let's jump into some questions and answers. Let's go into here. Let's go into community. Your channel should have way more subs by now. Keep at it, man. You're putting out great content. I'm no YouTube Algo expert, but sometimes you speak pretty fast. I'm wondering if the algo is having a hard time haring your speech. I personally don't have any trouble understanding you. How? Okay. Yeah, I could agree. But people can slow it down and I just make these videos because I want to make them and I could probably get more views by changing things. But either way, I appreciate this comment. This is great. If you haven't commented, you should go comment right now and I'll answer your question tomorrow. Let's go over to the other channel. This one, the Daily. And this guy is crazy. And we got a bunch of comments on Turbo. Oh, I never released that. Whoopsies. Is there anything else in here? Final learn Python. I guess we're stuck to next in Python. Yeah. Why Astro? Cuz it was fun. Round the corner. Sharpen the edges. You're dropping cones on us. I had to look that up. That was a nice phrase. Great video. Thanks. Can you make one outing a complex feature? Covered that yesterday. If you learned just one thing, make sure you like the video. If you learned two things, make sure you subscribe. If you learned three things, go buy yourself a cake and have a good day. All right, I'll see you tomorrow.