Evaluator Agent | AI Agent Types | Agentic Swarm

The Evaluator Agent critiques outputs and improves them through feedback loops. It checks work against the brief and either refines it or sends it back.

Why it matters

It keeps quality high as volume grows, and the lessons it produces are what the learning layer stores. It sits in the Memory and Recursive Learning Layer because evaluation is how the system learns what good looks like.

Use cases

• Improving copy and refining translations
• QA on generated content
• Scoring outputs against a rubric
• Flagging claims for human review

Inputs

• Execution outputs
• Success criteria and standards
• Past quality signals
• Brand and policy rules

Outputs

• Pass, fail, or improve decisions
• Refined outputs
• Quality scores and notes
• Lessons for memory

Failure modes

• Approving work that misses the brief
• Endless revision loops
• Checking style but not substance
• Discarding its own lessons