Why You Lose Control When AI Writes Your Code

I wrote about this and I talked about it on my podcast a couple weeks ago like there's something called comprehension debt which is a form of technical debt. It's like before cloud code, before agentic code, we had this theory of our codebase in our minds otherwise it wouldn't exist. Right? That's the first couple months of pods was I knew exactly what Podskan was in my brain and then I kind of coded it into reality by putting it into this codebase. Now that cloud code is working for me, it has its own theory of what the product is and that is the mix of the system prompt and this platform docs and the com the comments and the actual code. that is not necessarily what I have in mind and it could well be slightly different enough that cloud code just does things differently and changes the overall actual theory of the codebase that is in my mind is kind of decynchronized at that moment. So I do this step so I can recynchronize my theory of the codebase that I have in my mind with what the actual theory of the codebase is as expressing code. And if you don't do that then your understanding of what the codebase is drifts away from the actual codebase and you end up with code you don't understand. And the problem is that the agentic coding system does not retain the theory. Like they built the theory up. Like when I start this, when I say what I want and I hit enter, cloud code builds this internal theory of what the codebase is. It changes the code and then it ends. It says all changes are built and ready and it discards the internal theory completely. Whatever it was thinking the codebase was is completely gone. — Yeah. It has to sort of like relearn that. So I need to take this step to rebuild this internally and maybe adjust the comments or maybe adjust my prompt or whatever so that there is like this unified theory of the product that I have which is why I stepped through each of these changes.