Technology & AI
Lean Pool: AI agents formalize math, paper is one human page
A University of Washington preprint describes a formalized mathematics repository grown by AI agents, with a one-page human-written paper.
AI Agents
AI agents explained — autonomous assistants, coding agents and the research on where they work and where they fail.
117 stories
Technology & AI
AI coding agent router cuts enterprise model spend 21%
A prompt router can recover 14 to 21 per cent of enterprise AI coding-agent spend, or up to $5M a year for 10,000 seats, according to a 25 September
Technology & AI
AI agent rare-event estimation 800× more efficient
Columbia researchers built a method to estimate one-in-a-billion AI agent failures 800× more cheaply than brute-force simulation.
Technology & AI
SimLife benchmark: AI agents watch 15 hours, can't learn your habits
SimLife tracks 106 episodes of simulated daily life and finds frontier AI models guess habits by frequency, not reasoning.
Technology & AI
OpenAI tells firms to measure AI ROI with its own tools
OpenAI's September 16 guide urges firms to measure AI value with ChatGPT Work and Codex analytics, with no customer evidence or pricing.
Technology & AI
AI agents can probe neural networks but misread their own data
A new benchmark tests if AI agents can autonomously reverse-engineer language models, exposing a gap between designing and reading their own experiments.
Technology & AI
AI agents don't know when they fail: new method reads internal signals
A new arXiv paper reads AI models' internal representations to predict whether agent actions will succeed, with zero extra compute cost.
Technology & AI
AI agents learn your standards, up to 20.9% better
New preprint: AI agents learn individual standards through interaction, lifting task success up to 20.9% for professionals needing personal quality.
Technology & AI
Environment evolution for terminal agents: 18-point gain
New arXiv preprint: terminal AI agents gain 14 to 18 benchmark points by training against progressively harder environments. Results are unverified.
Technology & AI
Vercel Labs agent-browser hits 41,800 GitHub stars
Vercel Labs' agent-browser, a Rust CLI that lets AI agents drive Chrome without Node.js or Playwright at runtime, is trending on GitHub with 41,800 stars.
Technology & AI
New technique cuts AI agent wait time up to 45%
SMC pairs a small drafter model with a large actor model to speculatively execute tool calls, reducing wall time by up to 45% on AppWorld.
Technology & AI
Prompt injection attack on AI agents jumps to 28% success
SIR, a self-improving prompt injection attack, raised its success rate from 0% to 28% on Gemini 3.5 Flash while the agent's legitimate task still