Fragmented - AI Developer Podcast 306 - Keeping your agent instructions in sync and effective
17 snips
Mar 10, 2026 They tackle how AGENTS.md emerged as a common language for AI coding tools and the messy reality of keeping repo, personal, and tool rules synchronized. They compare different sync workflows like RuleSync, dotfiles, and symlink scripts. They also review recent research on auto-generated versus human-written AGENTS.md and when explicit repo-level instructions actually help.
AI Snips
Chapters
Transcript
Episode notes
Agents.md Emerged To Fix A Fragmented Ecosystem
- The AI agent instruction ecosystem was fragmented with many harness-specific files before agents.md emerged as a converging standard.
- Kaushik and Iury note agents.md was donated to the AAIF and OpenAI helped rally adoption, but some vendors like Anthropic keep their own formats (cloud.md).
Iury Uses A Dotfiles Repo And RuleSync
- Iury keeps agent configurations in a dedicated dotfiles-style repo and uses RuleSync to import/export harness configs between tools.
- He separates global user rules from project rules and runs RuleSync to regenerate per-harness files instead of manually copying.
Automate Agent File Setup With A Shell Script
- Kaushik automates setup with a shell script that symlinks a maintained .agents folder into each tool to keep files consistent.
- Run the script when switching machines so your project and global agent configs install identically.