Reflexion
In one line
Reflexion is an agent pattern where the agent reviews its own previous attempt, writes a critique, and uses that critique to do better on the next try.
Going deeper
Reflexion is an agent pattern introduced at NeurIPS 2023. Instead of running a single reasoning loop and stopping, the agent writes a natural-language critique of its own previous attempt, stores it in memory, and uses it as context for the next try. The motivation was a known weakness of ReAct-style loops: making the same mistake repeatedly.
The flow is straightforward — act, evaluate the result, write a self-reflection note, then attempt again with that note in context. Reported gains over plain ReAct on coding and math-reasoning benchmarks are what made the pattern stick.
The practical caveat is that the self-critique is not always right. A bad reflection can make the next attempt worse, so production systems usually pair Reflexion with an external check — a test runner, a judge model, a verifier — to confirm whether the reflection actually improved things.
Sources
Related terms
LangGraph
LangGraph is the graph-based agent workflow framework from the LangChain team — built so that branches, loops and human-in-the-loop steps are first-class.
AI AgentAutoGen
AutoGen is Microsoft Research's multi-agent framework, built around conversational agents that exchange messages and divide work between themselves.
AI AgentCrewAI
CrewAI is a multi-agent framework that lets you stand up role-based agents that work as a team, with role, goal and task defined declaratively.
AI AgentReAct
ReAct (Reasoning + Acting) is the classic agent pattern where an LLM loops through Thought, Action and Observation steps — reasoning out loud and calling tools as it goes.
AI AgentAgent Evaluation
Agent evaluation is the test and metric framework for measuring how accurately and safely an agent completes its goals — distinct from plain LLM benchmarking.