ModelRefs / Self-Refine Agent — Agent Pattern
Self-Refine Agent — Agent Pattern
Agent drafts, critiques, and revises its own output in a bounded loop before returning the result. Cheaper than multi-agent debate for long-form output.
Overview
Single-agent variant of debate: produce → critique against rubric → revise. Stops after K rounds or when the critique returns no new issues. Cheaper than multi-agent debate for long-form output.
When to use it: You need higher quality on long-form output and can afford 2-3× latency.
Pattern details
| Pattern class | reflection |
|---|---|
| Difficulty | intermediate |
| Autonomy | guided |
| Also known as | self-critique agent, draft-critique-revise |
| Last reviewed | 2026-06-07 |
Known failure modes
- Loop converges identical text — Critique surfaces nothing new. Mitigation: Inject explicit rubric and forbid generic feedback.
When not to use it
- Refining past 3 rounds (diminishing returns).
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Self-Refine Agent — Agent Pattern.
Frequently asked questions
When should I use the Self-Refine Agent agent pattern?
You need higher quality on long-form output and can afford 2-3× latency.
What are common failure modes of Self-Refine Agent?
Loop converges identical text
Is Self-Refine Agent production-ready?
Yes when paired with the safety controls and observability hooks documented on the pattern page.