# Agent Reflection

> When an AI agent reviews its own work and fixes what it finds.

**Field:** Artificial Intelligence  
**Category:** AI Agents  
**Also known as:** Self-Critique (AI), Self-Reflection (AI), Self-Correction (AI)

## Think of it like

It is like checking your work before you hand it in. You catch your own mistakes.

## Definition

Agent reflection is when an AI agent checks its own work and fixes it before it moves on. It writes a draft or takes a step, then reads the result with fresh eyes. It asks what is wrong or missing, and writes a better version. Real feedback helps most, such as a failed test or an error message. Without that, a model can miss its own mistakes, so reflection works best when it can check against facts.

## Example

Jack's agent writes a function, runs the tests and sees two fail. It reads the errors, finds its own mistake and fixes it before it reports back.

## Why it matters

One try is often not good enough. A quick self-check catches many mistakes before a person sees them.

## Common mistake

Trusting reflection with nothing to check against. A model grading itself by feel can call a wrong answer right.

## Three principles

- **Draft, check, fix.** A second pass on its own work.
- **Real feedback wins.** Tests and errors beat opinion.
- **Same model.** No second judge is needed.

## Related terms

- [Evaluator-Optimizer Loop](https://learn.benibix.com/term/ai/evaluator-optimizer)
- [Agent Loop](https://learn.benibix.com/term/ai/agent-loop)
- [Reasoning Model](https://learn.benibix.com/term/ai/reasoning-model)
- [AI Agent](https://learn.benibix.com/term/ai/ai-agent)
- [Hallucination (AI)](https://learn.benibix.com/term/ai/hallucination)

## Told apart from

- [Agent Reflection vs Evaluator-Optimizer Loop](https://learn.benibix.com/compare/agent-reflection-vs-evaluator-optimizer): People think reflection and an evaluator-optimizer loop are the same thing, and they are not. In reflection, one model reads its own work and fixes it. In the loop, the job is split in two. One model writes, and a second model judges it by clear rules. A second judge can catch what the writer keeps missing.

![Three panels. Left: one job splits into three agents side by side, and their answers merge into one. Middle: a writer model passes a draft to a checker model, which sends notes back until it says pass. Right: one model writes a draft, reads it again, and fixes it.](https://learn.benibix.com/diagrams/ai7-agent-patterns.svg)

---

Source: https://learn.benibix.com/term/ai/agent-reflection
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-19
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
