What is an agent harness?

An agent harness is the software around a language model that turns a single-step text generator into a working system. The model proposes; the harness does everything else: assembling the prompt, enforcing rules, running code, and deciding what happens next.

A typical harness owns several jobs. It builds and maintains context: system prompt, message history, retrieved documents, and tool results, all within the token budget. It defines the tools the model may call, validates the arguments the model produces, executes the real function, and feeds the result back as text. It runs the loop: send context to the model, act on the response, repeat until the task finishes or a limit is hit. It also enforces what the model cannot do, through permission checks, timeouts, retries, and audit logs.

Anthropic's guide on building agents draws a useful line between workflows, where the sequence of model calls is fixed in code, and agents, where the model dynamically decides the steps. Either way, the harness is that code. Agent frameworks and SDKs ship prebuilt harnesses, but the moving parts are the same: context management, tool dispatch, and loop control. When an agent misbehaves, the harness, not the model, is usually where you look first.