Been a bit quiet lately because I’ve been focused on job applications and interview prep, plus trying to clear out a load of stuff before my car lease ends.
On the side though, I’ve been doing some agentic coding on other projects. One thing that’s become really clear: once you have a decent codebase, AI agents can easily introduce regressions when you ask them to make changes.
To help prevent that, I’ve started putting specific instructions in my AGENTS.md file. The two main ones are:
1. Use TDD: Write failing unit tests first (red), then write the code to make them pass (green). Forces the AI to think through the details properly before touching the code.
2. Follow SOLID principles (especially Open-Closed), so the AI designs for extensibility from the start and uses proper patterns instead of just patching things.
It’s made the output noticeably more reliable. Thought it might be useful for anyone else using AI agents to build stuff.
#buildinpublic #indiehacker #AIcoding #Claude