13/01/2026
Here’s the thing about working with AI on codebases:
You explain an architectural decision carefully. The model seems to understand. Then a few prompts later, it suggests something that quietly breaks the system.
The instinct is to prompt better. Longer explanations. More reminders.
That’s not the fix.
The real problem is that the decision only lived in chat. And chat disappears.
What works: anchoring decisions in code. Module names that signal responsibility. Function names that encode behavior. Tests that assert boundaries.
When intent lives in structure, you stop babysitting the model and start collaborating with it.
Read the blog - link in bio!