Arbor: open-source framework for AI agents doing autonomous long-horizon research
burkov · x · 2026-09-30
Andriy Burkov introduces Arbor, an open-source framework for autonomous research under a setting called Autonomous Optimization: an AI agent takes an initial research artifact (code, models, data pipelines) and iteratively improves it toward a defined objective without step-level human supervision.
- Motivation: agents can now handle long coding tasks, but sustaining genuine scientific research across long horizons remains open — discovery requires exploring competing directions, learning from costly failures, and refining future attempts
- Existing systems fall short: most treat experiments as isolated sequential trials or lean heavily on human intervention, losing structural continuity and overfitting to development feedback
- Arbor is designed to preserve structural continuity across the experiment-reflect-improve loop
More from coding & agent
- 1,000 AI agents discover new CRISPR-like system in virus DNA within 24 hours — CurieuxExplorer · 2026-09-30
- SEAD: SAGE defender cuts tool-agent attack success from 48% to 4% against DART attacks — Xinjie Shen · 2026-09-30
- IBM's Q&D trains proactive agents to ask better questions, beating a 15x larger model — ibm · 2026-09-30
- One owner per decision: managing coding-agent project memory across eight repos with Git — oliver-zehentleitner · 2026-09-30
- Hister: open-source private search engine turns your pages and files into an MCP-searchable agent index — bibryam · 2026-09-30
- Monid, 'OpenRouter for Agent Tools,' Unifies 2,000+ Tools Across 72+ Providers — aigclink · 2026-09-30