Intermediate · Systems

Agent Harness

Visual diagram · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. Everything surrounding a model that makes it useful in real workflows — the tools, permissions, memory, context, and orchestration around the model itself.

Technical Definition

Everything surrounding a model that makes it useful in real workflows — the tools, permissions, memory, context, and orchestration around the model itself.

How it works

Popularized by GitHub's September 2026 agent glossary, the harness is the software layer that turns a raw model into a working agent. The model only generates text; the harness connects it to codebases, editors, terminals, and pull requests, supplies tools and permissions, maintains memory and state across calls, and decides when the task is done. This explains why the same model can shine in one product and flounder in another — the difference is the harness. The now-standard systems equation is Agent = Loop(LLM + Harness): the model reasons, the harness keeps resources available, invokes tools, preserves state, accepts feedback, and keeps operating as the task changes. Harness engineering — designing this layer deliberately — became a first-class discipline in 2026.

Visual Explanation (flowchart)

User goal → Harness (tools, memory, permissions, orchestration) → Model call → Tool execution → State update → next model call

Related Concepts

• Loop Engineering — The practice of designing repeatable systems around agents, instead of manually prompting them for one task at a time.
• Hill Climbing (Agent Improvement) — The ongoing work of incrementally improving an agent and its harness once both are already running in production.
• Agent orchestration — Agent orchestration is the management of multiple AI agents or LLM calls to efficiently handle complex tasks.
• Model Context Protocol (MCP) — An open standard from Anthropic that lets LLM apps connect to external tools, data sources, and prompts through a uniform client–server interface.