A pattern language for herding AI coding assistants

What it does

This repository is a curated catalog of 24 AI development patterns organized by maturity level and lifecycle phase. It treats AI-assisted coding as an engineering discipline rather than a solo magic trick, mapping out dependencies so teams know whether to start with a readiness assessment or jump straight to parallel agents. The patterns cover everything from sandboxing AI tools away from secrets to running multiple models against the same task for cross-validation.

The interesting bit

The author applies old-school software pattern language—complete with dependency graphs and maturity badges—to the chaotic world of LLM-assisted development. Rather than pretending AI workflows are self-evident, it explicitly labels which patterns are beginner, intermediate, or advanced, and warns that security and deployment practices should run continuously rather than waiting for a Phase Three waterfall.

Key highlights

• 24 documented patterns across Foundation, Development, and Operations tiers
• Maturity levels from beginner to advanced with explicit prerequisites
• Dependency graphs showing how patterns chain together; for example, atomic decomposition unlocks parallel agents
• An experimental folder for patterns still under active consideration
• Automated validation via GitHub Actions to keep the catalog internally consistent

Caveats

• The author explicitly warns that the patterns are subject to change as the field evolves
• Experimental patterns are acknowledged but not detailed in the main documentation

Verdict

Engineering leads trying to scale AI assistants beyond solo cowboy coding should bookmark this. If you are looking for a drop-in library or framework, this is pure knowledge work—no installable code, just structured guidance.