Awesome Agentic Patterns

Curated catalogue of 169+ repeatable, production-ready patterns for autonomous and semi-autonomous AI agents. "Tutorials show toy demos. Real products hide the messy bits."

Pattern Categories (8 Major Areas)

1. Context & Memory — 20 patterns: window management, episodic memory, state externalization
2. Feedback Loops — 14 patterns: CI integration, self-critique, iterative refinement
3. Learning & Adaptation — 7 patterns: reinforcement fine-tuning, skill libraries
4. Orchestration & Control — 51 patterns: task decomposition, multi-agent coordination
5. Reliability & Eval — 21 patterns: guardrails, observability, reproducibility
6. Security & Safety — 15 patterns: sandboxing, access control, audit trails
7. Tool Use & Environment — 25 patterns: API design, shell integration, sandbox design
8. UX & Collaboration — 16 patterns: human-in-the-loop workflows, transparency

Pattern Criteria

• Repeatable (multiple teams using it)
• Agent-centric (improves sensing, reasoning, or action)
• Traceable (backed by public references)

Discovery Resources

• Pattern Explorer: searchable, filterable
• Compare Tool: side-by-side analysis
• Decision Explorer: interactive use-case guidance
• Graph Visualization: relationship mapping
• Pattern Packs: curated collections
• llms.txt for AI assistants

Origin

Grew from "What Sourcegraph learned building AI coding agents" and the "Raising an Agent" series.

4,500+ stars, 401 forks, Apache 2.0.