Stop Babysitting Your Coding Agent

So when you are actually creating a ticket, you can automate all the all the stages of etiquette. For example, you file a GitHub issues in GitHub board. Of course, you can automate using GitHub CLI or GitHub MCP. You can draw and you can ask Claude Code, hey, this is my ticket, and you can code it. It will be able to code that. In most cases for small bugs, definitely it will be able to handle it because Claude code is quite good enough to figure things out in terms of what is the context in your in your code base, and it will be able to create that. So, what exactly will be a bottleneck? For a small task? This might work, but when it comes to a larger task, what will happen is it doesn't get the enough context of what exactly you need for your application. For example, if your spec just says, hey, create a landing page, and that landing page need to be look like this and this, and you know, you create that. Now that's just a spec with definition of how the UX need to be there, how the UI needs to be there. But now the coding agent, does the coding agent know whether you need to create multiple tasks for it, or whether you need to create a single task with single PR. It doesn't know. And does the coding agent know whether to pick JavaScript, TypeScript, or whether to pick Python or whatever language is is available. What tools or what frameworks or what are all the context behind it? It doesn't know, and it doesn't have the context in terms of whether to build this like in Apple scale, like Apple or or bigger company scale, or we need to build it like a startup scale with a landing page with a wait list, it doesn't know that context, even though you clearly mention each and every spec of it, it doesn't provide that insights which is needed for that factory. So that is where you need to have, you know, what exactly you are going to build that you need to provide that context to the factory. Then automatically you can start to automate slowly, like 10%, 20%, 30%. You can able to do that. Now, when you are doing your coding manually, what exactly you do is like you will say, okay, hey, Claude Code, this is my spec, and you can provide me the answer. So since you it's your harness and you you keep your memory, like the Claude code will keep memory of what are all the things you have chatted before, then automatically from the memory, it will be able to get some context. Okay, so-and-so person, and he likes TypeScript, he likes this runner, or he likes this way of coding some flavors and all those things. But now also it is able to not able to 100% match what is in your expectation. So you will iterate again and again, you will do the planning, you will iterate again and again, and then you will ask to execute. Once the execution is done, you will once again go and iterate again and again. Now, what exactly happens is like you are managing the process, you are sitting babysitting the process exactly. But when it comes to a factory, you can't you know go and babysit each and every step, but instead you create that as a process, like how you can create that as a process. That is what we we are going to talk about.