WTF is Harness Engineering?
Improving everything around the model (tools, config, checks, AGENTS.md) so the agent stops making the same mistake.
Also called: agent harness, harness
TL;DR
Agent = model + harness. You can't change the model, but you own the harness: the instructions file, the tools, the linters and tests it runs, the permissions. Harness engineering is the habit of turning every agent mistake into a permanent fix in that layer, so it never happens again.
How it works
Hashimoto's definition: whenever an agent makes a mistake, engineer a solution such that it never makes that mistake again. In practice that means adding rules and context to AGENTS.md or similar files, giving the agent deterministic tools (formatters, type checkers, test runners) instead of relying on judgement, constraining permissions, and designing feedback loops it can act on. Research framings describe the harness as every piece of code, configuration and execution logic that isn't the model itself.
Origin
- Coined / introduced by
- Mitchell Hashimoto
- When
- 5 Feb 2026
- Story
- Named in his essay 'My AI Adoption Journey'; the term was quickly picked up by framework authors and researchers.