Glossary term
Glossary term
Architecture
Agent that reviews another agent output for quality, policy, or completeness.
Constitutional AI (Anthropic) uses a critic agent to evaluate model responses against a set of principles, identifying harmful content and generating self-critique that the generator uses to produce revised outputs.
Microsoft AutoGen's code-review agent acts as a critic, evaluating generated Python code for correctness, security vulnerabilities, and PEP 8 compliance before the code is handed off to the executor.
A regulatory-filing agent uses a critic agent to cross-check each drafted disclosure section against SEC plain-English guidelines, flagging overly complex language or missing required disclosures before submission.