About Agent Patterns¶

A practitioner's catalogue of patterns and primitives for engineers building with AI coding assistants.

Agent Patterns is a reference site for experienced developers using AI coding assistants daily — Claude Code, GitHub Copilot, Cursor, and the surrounding ecosystem. The catalogue is tool-agnostic in its principles and tool-specific in its examples. Pages aim for a sub-60-second read with one concept per page.

What this site is¶

A practitioner resource — patterns, anti-patterns, paper summaries, and tool features — written grok-fast with no academic ceremony. Browsing the tree is the fallback; the primary access pattern is LLM-first discovery and aggressive cross-linking.

How it's maintained¶

The site runs as a content pipeline. Feedback and suggestions land through GitHub issues against the public mirror at agentpatterns-ai/website. Every page goes through research, drafting, review, and lint gates before publication — the inclusion criteria below are what every page is reviewed against.

What earns a page¶

A topic earns a page when it meets three tests:

Practitioner-relevant: it changes a decision an experienced engineer makes while shipping with AI assistants today.
Fast to grok: it can be communicated as a single concept inside the length target for its category.
Sourceable: every technical claim links to a primary source. Unsourceable claims are rewritten weaker, removed, or filed as a research issue — never hedged.

What doesn't¶

These do not earn pages:

Runnable libraries or SDKs — the site documents patterns and trade-offs, not code packages.
Framework tutorials — readers already use AI assistants daily; no step-by-step onboarding.
Generic LLM theory — the audience is engineers shipping with AI coding assistants, not researchers surveying the field.
Opinion essays or hot takes — patterns are presented with trade-offs, not ranked or endorsed.

How pages stay current¶

Pages carry optional currency frontmatter — last_reviewed, applies_to, and status — that surface as reader-facing badges (current, aging, archived). Data-driven triage runs quarterly against Google Search Console traffic and AI-citation signals, marking underperformers for consolidation, archival, or rewrite.

How to suggest a page¶

Open an issue on the public mirror with the concept, context, and any references. Ideas enter a research → drafting → review pipeline before publication.

Key Takeaways¶

The site is a practitioner reference, not a tutorial or a manifesto.
Inclusion criteria are explicit and public — what earns a page and what doesn't.
Every claim is sourced; the unsourced ones get rewritten, removed, or queued as research.
Currency and triage are policy-driven, not gut-driven.

Tags — browse content by topic
Concept Map — all content grouped by theme