Theo on what an agent harness is
Theo Browne's one-line definition of a harness: the layer that gives an LLM tools, context and capabilities, like Claude Code and Codex.
Harness = thing that turns an LLM into an agent
Provides tools, context, capabilities etc.
Claude Code and Codex are harnesses. Pi is a harness. The thing the LLM is interfaced through is the harness.
Theo Browne (@theo), on X, October 22, 2026.
Context: Theo posted this in a reply thread the same evening he wrote that many devs “straight up have no idea what a harness is”, and followed up with a longer X article, “What is a ‘harness’?”. The definition is worth keeping handy this week: Moonshot evaluated Kimi K3.1 Pro on Terminal-Bench 4.0 “with the Claude-Code harness”, which means the model’s headline score was measured inside Anthropic’s harness.
— Theo Browne