The Evaluator Optimizer is a workflow pattern that uses a generate-evaluate-refine cycle to close quality gaps in an agent’s output. It consists of three components:

  1. Generator: Produces an initial candidate output.
  2. Evaluator: Scores the output against a rubric or set of constraints, providing structured critique, not just a pass/fail signal.
  3. Optimizer/Refiner: The generator revises its output based on the evaluator’s feedback.

This loop continues until a convergence criterion is met (e.g., quality score threshold or iteration cap).

Why it matters: This pattern provides a structured way to handle ambiguity and improve the quality of an agent’s output through iterative refinement.

Connections:

Sources: