Reflection Agents: Self-Critiquing LLM Systems

Reflection agents enable LLMs to self-critique and iteratively revise outputs until quality thresholds are met, a powerful agentic pattern with documented risks. Seven failure modes are catalogued, including hallucination during critique compliance, making this critical for teams deploying autonomous LLM pipelines in production.