What's an agent harness?

This is episode 9 of a series of shorter blog posts answering questions I received during the course of my work. They discuss common misconceptions and doubts about various generative AI technologies. You can find the whole series here: Practical Questions. The term "agent harness", or even simply "harness", seemingly appeared out of nowhere at some point at the start of this year, and everybody immediately started using it as if it had always had an obvious and commonly understood definition. If this caught you off guard, you're not alone: the term has existed since before LLMs, but it was a rather niche AI research term that only went mainstream in early 2026. Definition In short, an agent harness is all the software scaffolding that sits around an LLM. The harness is the software that manages how the LLM uses tools, remembers context, stays within its guardrails, etc. To use a simple metaphor, the LLM is the brain of the AI agent: it understands, reasons, makes decisions, etc. The…