Nvidia research puts the agent harness, not the model, at the centre of the result

Reported on 21 August 2026: Nvidia research finds that agents can perform well and stay on task through fine-tuning of the scaffolding around them — the harness — even when the underlying model is not especially strong at the task itself.

What it means If the harness carries the outcome, agent quality is engineering you own rather than a model you rent, and it is the half you can actually iterate on.