Self-Evolving Agents

Link : https://cookbook.openai.com/examples/partners/self_evolving_agents/autonomous_agent_retraining

---

Notes

• repeatable retraining loop for agentic system (llm-ensemble), a self-healing workflow
  • human review
  • llm-as-a-judge evals
  • iterative prompt refinement

for loop

• Baseline Agent : your homeboy
• Human Feedback (or LLM-as-judge) : do a alignment of human feedback and llm-as-judge feedback when getting started, important
• Evals & Aggregated Score : threshold or max_retry on loop for new prompt(s)
• Update Baseline Agent : loop achieves targeted performance, replace original baseline agent