How do you keep control of your codebase when relying heavily on AI coders?
by•
Building apps with AI models in minutes feels amazing but hitting a weird bug in code you didn't actually write is a nightmare. It’s way too easy to just ask the AI to fix its own mistake creating layers of patches that no one fully understands.
While it definitely speeds up shipping dealing with maintenance or scaling later down the line gets tricky fast.
Do you manually inspect everything the AI generates or do you have a specific rule to make sure you still understand how your app works under the hood?
11 views
Replies
I do not read every line, but I decide the shape, and that has kept me in control. My current app copies the architecture of an earlier app of mine that I knew well, and the rules are written in a project file the agent reads every session: every screen is a pair of files, a view and a view model, every third party SDK is imported in exactly one service and nowhere else, all database access goes through one module. The point is not elegance, it is that when a weird bug shows up I know which file it lives in before I open anything, even if the agent wrote that file. The other rule: when the agent learns something non obvious, a vendor quirk or the reason for a decision, it writes it into a notes file for that area. Months later the notes are what I actually read, not the code.
I don't inspect the code myself. What I keep control of is the rules. Before the agent builds or changes a feature, it has to read that project's business logic doc, and if a rule changes, the doc gets updated before the code does. And when it makes the same mistake twice, I don't just tell it again, it becomes a hook. I've got 69 of those now, and they run whether the agent remembers or not.