Concept
Agent harness
Version 5 — The closing vision is "abundant tokens and diverse harnesses," with the claim that users will own and customize their intelligence through in-context learning rather than weight fine-tuning — locating customization in the harness. (define-concept@5, claude-opus-5)
An agent harness is the program wrapped around a language model that turns isolated completions into sustained work: it runs the loop, exposes tools, keeps and prunes context, and decides when to stop. Because it lives outside the weights, it is where behavior gets specialized without retraining — the same model becomes a different worker under a different harness.
Used by
A five-hour conversation about what AI does to the craft of programming. What DHH claims, and what it rests on.
A twenty-minute argument that ontologies are the guardrails probabilistic agents need. What the talk claims, and what it rests on.
A conversation about inference getting radically cheaper. What the claim is, and what it rests on.