Google makes Harness2 AI agent framework publicly available
The tool lets AI agents adjust their own setup while they work, without a fixed answer key. The code carries an Apache-2.0 license and has drawn 4,740 GitHub stars.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Google has open-sourced Harness2, a framework that lets AI agents tune their setup as they run. It lifted GLM-5 by up to 11.5 points on WorkBuddy-Bench while its weights stayed frozen. The Apache-2.0 code has since gained 4,740 GitHub stars.
Why it matters
The framework is presented as a way for AI agents to improve themselves during operation.
Timeline
- Oct 18, 4:15 AM ET: Google open-sourced Harness2, a recursive agent harnessing framework.
- Oct 18, 1:39 PM ET: The Apache-2.0 code had reached 4,740 GitHub stars, with gains of up to 11.5 points on WorkBuddy-Bench.
Coverage
- Google releases Harness2, AI agents that improve themselves (pasqualepillitteri.it)
- Google Open-Sources Harness2, a Recursive Agent Harnessing Framework (marktechpost.com)
More from Technology & AI
- Eight states sue OpenAI, alleging misleading AI safety claims
- OpenAI announces solution to Navier-Stokes math problem
- Datadog bets AI agents can handle testing of AI-written code
- Anthropic and OpenAI sketch wealth-sharing plans before IPOs
- Mamdani administration proposes AI customer service rules
- Florida engineers study how AI behaves in high-risk situations