I recently wrote about optimizing repos for AI, but I’ve not quite settled on the question of, should you optimize repos for AI? That is, is there really substantial daylight between what an AI dev needs and what a human needs?

Issues with optimizing repos for AI​

Nearly everything I put in agent docs is useful for humans​

Granted, without AI agents I wouldn’t go through the trouble of documenting a lot of this. But that doesn’t mean the content isn’t useful. Even documentation to discourage AI-specific antipatterns is generally useful information for a human dev too.

AI agent doc organization is fragmented​

Each coding agent uses its own configuration file pattern for repo-specific instructions:

This creates a hassle just keeping guidelines between agents consistent, much less making information available for humans.

Solution: Make it easy for humans​

A better way is to organize information for humans and link to relevant artifacts in agent-specific files1:

This makes keeping information consistent across agent docs much easier, and is much more future-proof for updates to agent file schemes. It also keeps humans in the loop!

For commands/skills, automation can help - for example, I wrote the just-claude utility for automatically synchronizing Just recipes with Claude Code Skills.

This also sidesteps dev angst about being left behind. There’s really no difference between the goal of economical token use for AI and reducing cognitive overhead for humans - organizing for humans first makes everyone happy.

Footnotes​

I wrote about what content I put in these files in the (ironically titled) post Optimizing repos for AI ↩