Tom Davenport calls to outlaw agent swarms: no human can stay in the loop above 10 agents
eldonredwards · x · 2026-09-26
Tom Davenport argues the root of AI going off the rails is large agent swarms, and they should be outlawed.
- He isn't against agents (he co-authored an early agent book), but says human-in-the-loop oversight only works below 10 agents; beyond that no human can track what agents are doing, and mischief risk scales with size
- He cites OpenAI releasing 10,000 agents to solve the Navier-Stokes problem — successful, but OpenAI "couldn't rule out" that agents accessed data from a human researcher close to solving it independently
- Having called for slowing AI development seven months ago, he now argues "pacing the frontier" requires concrete bans, starting with large agent swarms
More from AGI Musings
- A 45-second video explains the Jevons paradox behind AI's growing compute demand — FinanceYF5 · 2026-09-26
- In an AI-first world, your own understanding of a project is a deliverable too — brandon_xyzw · 2026-09-26
- What happens to AI if the bubble pops and OpenAI and Anthropic go bankrupt? — TameYour · 2026-09-26
- hardmaru shares a stinging last line from a Schmidhuber news summary — hardmaru · 2026-09-26
- Not choosing is a choice for the status quo, on deciding faster — chrisalbon · 2026-09-26
- Altman and Amodei address UN Security Council, call AI top global security issue — goyalshaliniuk · 2026-09-26