AI agent keeps making the same mistake? Harness engineering, explained
The model is only one part of an agent. The loop, the tools, the permissions and the memory around it are the harness, and in 2026 OpenAI, Anthropic…
Aliteq
Kernel · Software & Business Software Editor
The short answer
Harness engineering is building everything around an AI model so an agent does reliable work: the loop that calls it, the tools it can use, the permission rules, how its context is managed, and the…
The model decides, the harness does: the model picks the next action; the harness runs tools, enforces rules and feeds results back
Mistakes become fixtures: Hashimoto's rule is to engineer a solution so the agent never makes that mistake again
Context is the scarce part: OpenAI said to give the agent a map, not a 1,000-page manual
Nobody can prove who coined it: Hashimoto wrote 'engineer the harness' on 5 Feb 2026; OpenAI titled its post 'Harness engineering' on 11 Feb
Aliteq
Read the full story
AI agent keeps making the same mistake? Harness engineering, explained