Harness²: Google's Open-Source Framework for Self-Improving AI Agents

TL;DR Harness² is an Apache 2.0 framework (v0.1.0) from Google Research that improves an agent's harness at deployment time. Model weights stay frozen. Only the prompts, tools, memory, skills and

TL;DR Harness² is an Apache 2.0 framework (v0.1.0) from Google Research that improves an agent's harness at deployment time. Model weights stay frozen. Only the prompts, tools, memory, skills and orchestration around the model change. Core mechanism: Propose, Probe, Compose. No ground-truth labels or test cases needed. Result: one refinement step lifted GLM-5 by 11.5 points on WorkBuddy-Bench (64.2 to 75.7). 1. What is a harness? A harness is the system around a model that lets it act as an agen