12-Factor Agents: Patterns of reliable LLM applications
summary
Argues for 12 engineering factors (repo: humanlayer/12-factor-agents) that make LLM apps reliable: convert language to JSON and treat tools as structured outputs, own prompts and context windows, control flow and pause/resume via simple APIs, and unify execution and business state. Cites talks with 100+ founders, 4,000 GitHub stars and 200k social impressions, says prototypes hit 70–80% then fail, recommends Small, focused agents with 3–10 step loops (example: HumanLayer deploy bot uses LLM for deploy decisions) and keeps agents stateless while owning state externally.
AI-written summary · about
chapters (11)
- 0:00The agent journey
- 1:34Principles of reliable apps
- 2:20The 12-factor agent approach
- 3:01Rethinking tool use
- 4:25Controlling the flow
- 6:17Context management and state
- 7:12Managing execution state
- 8:35Context engineering
- 10:21Error handling and human interaction
- 12:24Small focused agents
- 14:24Summary and future outlook
related talks
38mNobody Pages the LLM: Engineering Rigour for Vibe Coding
01h 00mFrom LLM Theory to Practical Agentic Implementations
41mBuilding LLM Agents – Challenges and Learnings
50mHow to Test LLM Agents
38mMaking LLMs reliable - A practical framework
19mLLM codegen fails and how to stop 'em
27mThe Lean LLM development playbook
26mContext Engineering: Practical Techniques for Improving Agent Quality Today