ModelRefs / Reflexion — Agent Pattern
Reflexion — Agent Pattern
Agent reflects on failures, writes a self-correcting note, and retries with the note in context. On retry, the lesson is prepended to the prompt.
Overview
After each failed attempt, the agent writes a structured 'lesson' about what went wrong. On retry, the lesson is prepended to the prompt. Improves multi-step accuracy without finetuning.
When to use it: Your agent fails the same way repeatedly across runs and you need it to learn within a session.
Pattern details
| Pattern class | reflection |
|---|---|
| Difficulty | advanced |
| Autonomy | autonomous |
| Also known as | self-reflection agent |
| Last reviewed | 2026-06-07 |
Known failure modes
- Vague lesson — Reflector emits generic advice with no signal. Mitigation: Require concrete observations and a falsifiable change.
When not to use it
- Treating lessons as durable memory across users.
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Reflexion — Agent Pattern.
Frequently asked questions
When should I use the Reflexion agent pattern?
Your agent fails the same way repeatedly across runs and you need it to learn within a session.
What are common failure modes of Reflexion?
Vague lesson
Is Reflexion production-ready?
Yes when paired with the safety controls and observability hooks documented on the pattern page.