H
Harness²
A plug-and-play framework for recursive agent harnessing that refines prompts, tools, memory, skills, and orchestration around a fixed model.
15.3k GitHub starsApache-2.0 licenseBuilt by Google Research1.4k GitHub forksBenchmarks: JobBench, LAB, WorkBuddy-Bench
Opportunity score
53/100
Harness² is technically impressive and sits in a large market, but it is not yet a business and would need a commercial wrapper to become investable.
Founder verdict
MAYBE
Harness² is a technically compelling open-source research project, not yet a startup; the opportunity is in a focused commercial wrapper.
What is Harness²?
Harness² is an open-source research framework from Google Research for recursively optimizing the 'harness' around an AI model—prompts, tools, memory, skills, guardrails, and orchestration—without changing model weights. It introduces task-level recursion to refine a harness for the current task and harnessing-level recursion to retain adopted harnesses and lessons across tasks. The GitHub repository has strong early traction with 15.3k stars and 1.4k forks, and the README reports benchmark gains on professional tasks such as WorkBuddy-Bench, JobBench, and LAB. However, there is no visible commercial product, pricing, revenue, or startup motion; it currently behaves as a research artifact rather than a business.
Get the full Harness² playbook
The complete reverse-engineering — what to copy, what to avoid, and exactly what to build instead.
- Full opportunity score across 8 dimensions
- The real problem & why customers pay
- Why it's winning — with evidence
- Complete business reverse-engineering
- Competitive advantages & moat analysis
- Weaknesses, risks & what to avoid
- Clone strategy — what to build instead
- Week-by-week MVP roadmap
- Recommended technical stack
- Founder verdict & highest-leverage move