Google Research open-sources Harness² to evolve Agent harnesses automatically
11/03/2026 — 11/03, 02:18·1 sources (all)·1 reports (all)
Key takeaways
- Google Research has open-sourced Harness², a framework that automatically evolves an Agent’s harness—including system prompts, tool descript…
Background & analysis
According to an AI article posted on DEV Community on November 3, 2026, at 01:44, Google Research has open-sourced Harness², a framework aimed at a different lever than model upgrades. When an agent disappoints, the usual instinct is to swap in a stronger model. Yet the same model can shift from disappointing to convincing depending on its system prompt, how its tools are described to it, and how it rereads its own work before delivering a result. That whole operating environment is called the harness. Harness² evolves it automatically without touching the model's weights, targeting the system prompt, tool descriptions, and pre-delivery self-review routines. The article says the accompanying paper and project site detail the method, the data, and the reported numbers, though this material does not include any of those specifics. For now, this single report is the only available account, and there is no further coverage to confirm what has happened since.
AI-generated from 1 reports · updated 3 days ago
- Heat index
- 25
- All sources
- 1
- All reports
- 1
- First seen
- 3 days ago
Reports on this story headlines open the original
Google Research has open-sourced Harness², a framework that automatically evolves an agent's harness — its system prompt, tool descriptions and self-review loop — without touching model weights. The accompanying paper and project site lay out the method, the numbers and the results.
DEV Community · AIAI score 59
Other stories people are talking about
- Heat index 90NewGoogle Research uses smartwatch sensors to detect hypertension1 sources
- Heat index 48Google Research proposes Agile Deliberation for subjective classification1 sources
- Heat index 65Developer open-sources Growth-Harness, an agent framework without a hard turn cap1 sources
- Heat index 840OpenAI ships Images 2.6 with GIF and SVG output15 sources
- Heat index 675OpenAI says its cyber model pipeline confirmed 1,724 Linux kernel flaws and generated 948 privilege-escalation exploits11 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion