What Is Loop EngineeringAre you using your coding agent wrong Full episode httpswwwyoutubecomwatchvM7hTxVvmEJY Channel httpsyoutubecompaddyborajam ai coding vibecoding

If you're still prompting your coding agent, you're doing it wrong. You should be doing loops instead. Loop engineering. So this is a new thing. You define what you want to achieve, and then you let the team of agents go to work to produce the final outcome. And you don't have to kind of get involved in the middle steps to keep prompting. So the way I do it, I use a concept called a spec kit. So this is a GitHub has a repo for spec kit. So, you know, people who are watching and who also write code or build apps, you should use this for important things. Spec kit allows you to plan what you want to build in phases. And then you kind of describe, you just like turn on WhisperFlow, describe to your either Cloud Code or Codex what you want to achieve, like for a big feature or like a series of features, what we call an epic. So if you want to build an epic. And then a spec kit is used to plan it out. So this uses way more tokens. So I don't use it for everything. I don't use spec kit for everything, like smaller things, because then it just like, it seems like it's a fairly simple thing to do. We don't need to just like plan a lot. But I like it for big, important and consequential things where, you know, code might break. A lot of code needs to be written. It needs to be tight. It needs to be verified and checked. It needs, you know, so, and then I tell the agent to first create a discussion in GitHub, there's a discussion section in GitHub. And I have the Cloud Code CTO and I have the Codex architect and I have all the engineers. And I have all of them go in there, create a discussion and just comment. So they comment back and forth several times. I let four or five different components. So like MCP engineer, web engineer, model engineer, computer engineer, and then CTO and the Codex architect, all of them are in there discussing. They're looking at the different parts of the code, discussing, and then once they come to an agreement, like I'm not weighing in. And then once they do it, then I say, all right, now you spec it to plan it out in phases. It creates issues and tasks and who's gonna do what, what's next, phases, phase one, phase two, phase three. And once it's done, you say, all right, go, start it. At this point, this may take several hours of work. And I explained last week, I think you remember, I set up email inboxes for all of these engineers and architects. So they email each other. They check their emails every five minutes. They email back and forth, they notify each other, hey, like I finished something, check this out, check this out. Sometimes several engineers check it out at the same time. And these engineers are of course AI. People are AI engineer, right? Engineer who's an AI or AI who's an engineer. So, and then in the end, they finish the work and what's left for me to do is to verify. They actually write the tests, make sure that it works, but then I also verify that it works the way that I want it. So like I'm involved in the beginning and then I'm involved in the end. I'm not involved in any of those things that happen in the middle. And this is loop engineering.