Skip to content
Agent YAP

What is it?

Agent harness

The harness is the software wrapper around the model. The LLM provides the thinking and produces a "tool call" (a request to run some action), but the harness is the code that actually runs that action, feeds the result back, manages the message history, and decides when to stop. The model is the brain; the harness is the body and nervous system that lets the brain act on the world. A recurring lesson in these articles is that as models get smarter, you should make the harness simpler, not more complicated. See also the Bitter Lesson.

From the AI agent glossary

Where Agent harness is explained

  1. Building the PromptBuilding Coding Agents and Harnesses · Concept explanation
  2. Tools and Tool ExecutionBuilding Coding Agents and Harnesses · Concept explanation
  3. Sandboxing, Approvals, and CheckpointsBuilding Coding Agents and Harnesses · Concept explanation
  4. Parallelism, the Responses API, and the App ServerBuilding Coding Agents and Harnesses · Parallel tool calls
  5. Skills, MCP, and Deterministic ComputationBuilding Coding Agents and Harnesses · MCP: a standard plug for external tools

Every term, A to Z