Critic agent
Published Updated
What is Critic agent?
A critic agent is a second AI agent that reviews each round of another agent’s work against a clear goal and sends it back with notes until the result passes.
A second AI agent that reviews each round of work against a clear goal and sends it back with notes until the result is good enough.
A critic agent is the reviewer in an agent workflow. It checks the work against the goal, points out what is missing or weak, and gives the working agent another round to improve it. It works best when the goal is clear; with a fuzzy goal, the critic cannot tell when the work should stop.
How it relates to Orbit
Orbit’s Convergence Gate is a bounded critic loop: Sweep reviews the result, the agent fixes what it found, and the loop stops when the reviewer clears or a limit is hit. See Convergence Gate for details.
When you would use it
You use a critic agent when an agent needs several tries to reach a clear goal and you want feedback between those tries instead of trusting the first answer.
Common operations
- Reviewing a generated game against the requested look and features before another improvement round.
- Checking a code change against its acceptance criteria before the agent says it is finished.
Frequently asked questions
Why use a critic agent?
Agents are poor judges of their own output. A separate reviewer with a clear bar catches mistakes the worker misses.
Does Orbit have a critic loop?
Yes. The Convergence Gate runs Sweep’s review through fix cycles until the reviewer clears, the findings oscillate or a maximum iteration count is reached. See Convergence Gate.