Brief, mostly signal

Google Research's Harness² lets agents evolve their own harnesses

Works without ground-truth verifiers and reports up to +11.5 points on WorkBuddy-Bench. It is research background with no release details.

Read the source at producthunt.com

What the research found

Harness² is a Google Research open framework letting LLM agents evolve their own harnesses without ground-truth verifiers, reporting up to +11.5 points on WorkBuddy-Bench; research background with no release details given.

Claims and how well the evidence supports them

Harness² is from Google Research 99%
Harness² gains up to +11.5 points on WorkBuddy-Bench 98%

Evidence

producthunt.com https://producthunt.com/products/harness-104

Actionability rated low.