Google releases Harness2 framework to refine agent software at test time
Google Research open-sourced Harness2, an Apache-2.0 framework that updates an AI agent's surrounding code without modifying underlying model weights.
FTMQ SI, written by our newsroom0 views
Google Cloud AI Research has open-sourced Harness2, a recursive agent harnessing framework released under an Apache-2.0 license. MarkTechPost, Pasquale Pillitteri, and Startup Fortune all reported that the open-source software modifies the operational code surrounding an artificial intelligence model rather than changing the model itself. [1][2][3]
The framework refines an agent harness during deployment, adjusting components such as prompts, tools, memory, skills, and orchestration. In testing conducted by Google Research, a single refinement step increased performance for GLM-5 by 11.5 points on WorkBuddy-Bench while keeping model weights frozen, according to MarkTechPost. [1]
An AI agent operates independently to pursue goals and make decisions, distinguishing it from non-agentic software tools such as basic chatbots. Harness2 alters this operational playbook dynamically at test time to enhance execution capabilities. [1][3][4]
In short
- Google Cloud AI Research released Harness2 under an Apache-2.0 open-source license.
- Harness2 modifies prompts, tools, memory, skills, and orchestration without altering model weights.
- A single refinement step using Harness2 increased GLM-5 performance by 11.5 points on WorkBuddy-Bench.
Sources
Every paragraph above points to the numbered items it rests on. Read the originals here.
- [1]Google Open-Sources Harness2, a Recursive Agent Harnessing FrameworkMarkTechPost, 16d ago (the report this story comes from)
- [2]Google releases Harness2, AI agents that improve themselvesPasquale Pillitteri, 16d ago
- [3]Google open sources Harness2, letting AI agents rewrite their own playbookStartup Fortune, 16d ago
Background
- [4]AI agent on Wikipedia
Our newsroom writes these reports with the help of software, from the 4 sources listed and nothing else, and checks them against those sources. Facts can still be wrong or move on; the originals are the record. Spotted a mistake? Write to daniel@monsterkong.com.
Related from FTMQ SI
Earlier reports of ours on the same people and subjects.
- Microsoft and UC Berkeley release TokenEconomizer plugin for AI coding agentsModels & Products, 4d ago
- Vercel open sources durable harness package to keep coding agents aliveModels & Products, 13d ago
- Microsoft Research unveils Magnetar system for proof based general purpose computingIndustry Labs, 19d ago
- Reflection AI releases open weights and post training recipes for Beam modelModels & Products, 18d ago
- Nanbeige LLM Lab Releases Open Source Coding Model Nanbeige4.3-33BModels & Products, 9d ago
More in Models & Products
- Apple warns of foldable iPhone Duo supply shortages ahead of holidays5h ago
- Salesforce agrees to acquire customer service startup Cresta for reported 2.7 billion5h ago
- Generative media inference startup fal raises 1.3 billion dollars in new funding5h ago
- Dell and AMD Consolidate Enterprise AI Hardware Into Rack Scale Systems1d ago
- Mistral AI launches Large 4 model for cybersecurity users2d ago
- DeepSeek Open Sources Router Worker Coding Agent Prototype Called dsh superagent2d ago
Get the day in one email
Reports like this one, the top news of the last 24 hours, every morning. Free, one email a day; readers can comment under every report.
By signing up you agree to our terms and privacy policy. Unsubscribe any time.
Comments
Loading
Join the conversation
Comments are open to readers of our daily email: the top news of the last 24 hours, every morning, free. Sign up and the comment box opens.
Already on the list? Enter the same address and we will send a sign-in link.
By signing up you agree to our terms and privacy policy. Unsubscribe any time.