Claude and Gemini crush system design and high-level modelling, but the moment you drop into actual function logic, they fall apart. It s like pairing with someone who writes clean diagrams but chaotic code.
And if you ask them to improve it, they vanish into SOLID-theory rabbit holes and over-engineered OOP madness.
These days, since I vibecode more, I have found myself refactoring what the robots are writing ever so often that it has got to a point where I straight up delete function logic and write it myself.

Curious if others are seeing the same?

Hello everyone, recently I am refactoring old AI codes in hot spots. I used both Claude Code and Codex to quickly implement the feature at that time. It works but when I look into the logic today, it introduces too much unnecessary complexity (a lot of helper/manager/try-except). Although I have manipulated CLAUDE.md (emphasizing KISS principle, introduce Linus https://gist.github.com/iiiyu/4c... ), the code agent still try to add entropy on the whole project. I can understand the code LLM is trained to program defensely, but if I do not review carefully and really understand the logic, the project quickly becomes hard to maintain. Now every week I will leave 1 day to write codes without AI to clean the whole project for longer future.
Do you have similar experiences or solutions to share?

Let's make a thread where you guys can share the apps you're making and support each other!

Two months ago, I'd never heard of Product Hunt. When I told people we were launching @AI Context Flow here, they told me to keep my expectations in check.

Fast forward to today: #1 Product of the Day and #1 Productivity Tool of the Week.

The journey was chaotic, humbling, and honestly surreal. If you'd told me this would happen, I wouldn't have believed you.

To everyone who upvoted, commented, and cheered us on: Thank you. Your support means everything and keeps us building.
If you need any tips on how we pulled this off as complete first-timers, ask your specific questions below