Frontier AI Agents Fail Open-Ended Research: $3,000 and 6 Days Yield Rejected Papers
gerardsans · x · 2026-08-02
A recent study tested whether frontier AI agents can conduct open-ended AI research. Agents were given six days, a $3,000 budget, full virtual sandboxes, and web access to tackle central research questions from two unpublished NeurIPS 2026 papers.
The results indicate that while agents completed the full execution loop with zero human intervention and handled many engineering tasks, they failed at core research objectives, and recursive self-improvement (RSI) never materialized. Original authors acting as expert reviewers rejected the outputs with low scores of 1/6 and 2/6. The study identifies five recurring failure modes, most notably poor judgment, such as misjudging task difficulty.
Related event: AI Agent Autonomous Research Fails: Papers Rejected(2 posts)→
More from coding & agent
- Dev builds MCP layer so Claude Code can work on servers without ever holding SSH keys — NETMESAFE · 2026-10-16
- 'Refusing to code with AI in 2026 is like refusing Google in 2005' — prasenx · 2026-10-16
- Claude Code adds idle compaction to save 1h prompt cache on idle sessions — EricBuess · 2026-10-16
- Handling agents that crash mid-task: the hard part is the unknown outcome across the network — Responsible-Dare5139 · 2026-10-16
- OpenAI DevDay Community Hackathon heads to Berlin with developer, founder and GTM tracks — paw_lean · 2026-10-16
- Data scientist career path is shifting: ML skills no longer enough, AI engineering is the new bar — mdancho84 · 2026-10-16