Reflexion
ELI5 - The Vibe Check
Reflexion is an agent re-reading its own exam answers before the next attempt and catching its own mistake. It doesn't get its weights updated or go back to school, it just writes itself a note about what went wrong and keeps that note around for the next try. Cheap, weirdly effective, and a little bit humbling to watch an AI do to itself.
Real Talk
Reflexion is an agent pattern where the model critiques its own output or trajectory after a failure, generates a natural-language reflection on what went wrong, and feeds that reflection back into its context on the next attempt. It improves task success rates without any weight updates, functioning as a form of in-context learning from the model's own mistakes.
When You'll Hear This
"The agent reflected on why the first patch broke the build, then fixed it on attempt two." / "We added a reflexion step and the retry success rate doubled."
Related Terms
Agent Loop
The agent loop is the heartbeat of any AI coding agent: think, act, look, repeat.
Chain of Thought (CoT)
Chain of Thought is when you tell the AI 'show your work' like a math teacher.
LLM-as-Judge
Instead of paying a human to grade a thousand AI-generated essays, you hire a second AI as the strict teacher who reads the first AI's homework and hands b...
Self-Correcting Agent
A self-correcting agent notices its own mistakes and fixes them without being told. It runs tests, sees failures, debugs itself, retries. Sounds magical.