Classification

Taxonomic Criteria of Harness Optimization Paradigms

Agent harness optimization paradigms can be classified across their construction mode and four fundamental functional capabilities:

  1. Construction Mode: Differentiates whether an operational harness is obtained via ahead-of-time (AOT) search over experience streams, an AOT scaffold adjusted post hoc through test-time feedback editing, or dynamically synthesized just in time (JIT) for each task instance.
  2. Instance Synthesis: Measures whether the optimization technique directly synthesizes an instance-specific operational scaffold tailored to the incoming task rather than applying a precompiled durable template.
  3. Harness Model: Assesses whether the system trains a dedicated generative meta-agent model to synthesize or configure executable harnesses.
  4. Learned Repair: Identifies whether the method explicitly trains a generator to recover from compilation, interface, or runtime execution failures using structured diagnostic feedback from failed trajectories.
  5. Online Evolution: Evaluates whether the system continues improving and refining candidate harnesses after deployment without retraining base generator weights.

0

1

Updated 2026-09-29

Tags

Prep Sessions

Autonomous Agent Control Planes: State Scaffolding and Resilient Execution @ University of Michigan - Ann Arbor

Ch.3 Harness Synthesis and Factory Patterns - Autonomous Agent Control Planes: State Scaffolding and Resilient Execution @ University of Michigan - Ann Arbor

Ahead-of-Time versus Just-in-Time Harness Paradigms - Autonomous Agent Control Planes: State Scaffolding and Resilient Execution @ University of Michigan - Ann Arbor