New term: harness engineering

2026-02-05 · Publisher : Mitchell Hashimoto · Domain : Artificial Intelligence

Harness engineering means designing an agent's environment (instructions, tools, tests, checks) so that a fixed mistake never happens again. Mitchell Hashimoto named the practice in February 2026; OpenAI popularized it six days later.

What changes

  • Every agent mistake becomes a permanent fix to its environment.
  • Engineers shift from writing code to designing the system that frames the agent.
  • At OpenAI, a team shipped a product with no hand-written code at all.
  • The term extends context engineering with tools, loops and controls.
  • Thoughtworks formalized it: guides upstream, sensors downstream.

### Definition **Harness engineering** is the discipline of designing an AI agent's whole environment: instruction files, tools, scripts, tests, linters, checks and feedback loops. Its principle: when the agent gets something wrong, you fix its environment so that the mistake becomes structurally impossible. ### Origin On February 5, 2026, Mitchell Hashimoto (co-founder of HashiCorp) published "My AI Adoption Journey". He wrote: "I've grown to calling this 'harness engineering.' It is the idea that anytime you find an agent makes a mistake, you take the time to engineer a solution such that the agent never makes that mistake again." On February 11, 2026, Ryan Lopopolo (OpenAI) published "Harness engineering: leveraging Codex in an agent-first world", summed up as "Humans steer. Agents execute." ### What the term covers - At OpenAI, about a million lines and some 1,500 pull requests in five months, with no hand-written code. - A short AGENTS.md acting as a table of contents to deeper documentation. - Mechanical rules and background maintenance tasks against architectural drift. ### Link with context engineering Context engineering decides what the model sees; the harness encompasses that context, but also tools, memory, the execution loop and controls. Birgitta Böckeler (Thoughtworks) distinguishes "guides", which steer the agent before it acts, from "sensors", which help it self-correct afterwards. ### Criticism and limits Böckeler notes that automated sensors catch structural issues well, but neither they nor guides reliably catch misdiagnosis or overengineering. Some also see it as a new name for long-standing good practices. ### Key takeaway Harness engineering makes the engineer the designer of the system the agent works in. Quality now lives in the scaffolding as much as in the code.

Links

Tags : Vocabulaire, Harness, Agents, Code