I Spent 3 Months Fighting Messy AI Code

How are things going with your web coding?

Do you do a lot of refactoring after the first AI implementation of a task?

I've been actively implementing AI coding into my work processes for three months now, both at work and in my pet projects, and I've realized that even when you give the AI a reference, it often misses the mark - somewhere it doesn't catch connections between components, somewhere it decides its approach will be better :) As a result, you get such a mess and have to double-check and constantly fix things more.

I'd like to hear about your experience, advice, and recommendations. And share my developments for solving these problems with you.

I'm developing a system of Claude Code commands and agents that you can tell your idea to in your own words, and they'll clarify details with you, research your code, conduct research on dev resources on the internet if needed, and create a clear technical PRP plan for implementing the task.

After that, you just need to start executing this plan and go drink coffee :) This system will be effective for more or less complex tasks, not simple ones like "add a button."

I'd like to share what I've built with you. Hopefully it'll be useful for everyone

https://github.com/croffasia/claude-code-prp-generator

P.S. I'm not a supporter of the OpenAI concept :) A one-prompt application isn't a working story in real projects.RetryClaude can make mistakes. Please double-check responses.