Curated reference of agentic AI patterns for coding agents. Use when designing workflows, skills, or agent behavior.
Source: nibzard/awesome-agentic-patterns (Apache-2.0)
Split reasoning into two phases:
- Plan phase — Generate a fixed sequence of steps before seeing untrusted data.
- Execution phase — Run that exact sequence. Outputs may shape parameters but cannot change which steps run.
When to use: Complex tasks where the action set is known but parameters vary. Can 2-3x success rates by aligning on approach first.
Cursor/Claude Code: Use "plan mode" or brainstorming → writing-plans → execution flow.
Spawn focused sub-agents with isolated contexts for shardable subtasks. Aggregate results when done.
Key practices:
- Clear, specific task subject for each subagent (traceability)
- Launch independent tasks simultaneously (parallel, not sequential)
- Limit to 2-4 subagents for coordination
- Plan synthesis upfront
When to use: Large multi-file tasks, context window management, I/O-bound workflows, parallel code migrations.
Design/spec drives implementation; tests verify spec. Spec and tests form a feedback loop.
When to use: Spec-driven development (e.g., no-vibe-coding workflow).
Transform imperative tasks into verifiable goals. Define success criteria. Loop until verified.
When to use: All implementation work. See Karpathy guidelines.
| Category | Examples |
|---|---|
| Context & Memory | Curated context window, progressive disclosure, working memory via TodoWrite |
| Feedback Loops | CI feedback, reflection loop, spec-as-test |
| Orchestration | Plan-then-execute, sub-agent spawning, parallel tool execution |
| Reliability | Structured output, schema validation retry, workflow evals |
| Tool Use | Code-then-execute, CLI-first skill design, dual-use tool design |
| UX & Collaboration | Human-in-the-loop, spectrum of control, proactive trigger vocabulary |
For the complete pattern catalogue with 100+ patterns, see:
- Website: https://agentic-patterns.com
- llms.txt: https://agentic-patterns.com/llms.txt (machine-readable for RAG/LLM context)
- Repo: https://github.com/nibzard/awesome-agentic-patterns