Reflection Agent
An agent pattern where the LLM critically evaluates its own outputs and iteratively improves them – like an internal code review.
Reflection agents improve AI outputs through self-critique and iteration – like an internal reviewer optimizing the first draft.
Explanation
Reflection agents first generate a draft, then evaluate it against defined criteria, identify weaknesses, and produce an improved version. This self-refinement can iterate multiple times.
Marketing Relevance
Significantly improves output quality: code review agents, content quality checks, fact-checking loops, and strategic analysis with counter-perspectives.
Common Pitfalls
High token costs from multiple iterations. Can lead to over-engineering. Some models evaluate their own outputs too uncritically.
Origin & History
Reflexion (Shinn et al., 2023) formalized self-reflection for LLM agents. The concept builds on self-consistency (Wang et al., 2023) and Constitutional AI (Anthropic).
Comparisons & Differences
Reflection Agent vs. Self-Consistency
Self-consistency samples multiple answers and picks the most common. Reflection critiques and improves one answer iteratively.