Ethan Mollick: I underestimated AI's ability to self-organize, agents beat elaborate orchestration
emollick · x · 2026-10-01
Wharton professor Ethan Mollick admits his biggest AI misjudgment: he assumed humans would need to manage agent swarms like managers building a company, with careful delegation and org design. Nope.
Key points:
- The Bitter Lesson keeps reasserting itself: elaborate systems built to feed AIs context at the right time, and intricate prompt-template chains, are outclassed as newer models learn to seek information and plan steps themselves. His own research shows human-planned steps add far less value now.
- Demo: with one prompt, Claude wrote lyrics and submitted them to Suno, while Opus 5.5 produced the entire music video using only code — no image generation, zero human feedback.
- Implication for agents like Muse and Dots: let the swarm self-organize instead of over-engineering orchestration.
More from AGI Musings
- Executive prediction: AI won't kill meetings, it will expose which ones never should have existed — ingliguori · 2026-10-01
- The Drift collects anonymous dispatches on how writers, doctors and academics use—or resist—AI — xuanalogue · 2026-10-01
- Joyce Carol Oates: formulaic fiction doesn't need a human author, and most pop culture is already slop — ShakeelHashim · 2026-10-01
- Anthropic Wants $2T for Week-Long Agents; Skeptic Says It's Just Excel Waiting to Be Fed — gerardsans · 2026-10-01
- AI tutor matches expert humans at 1/918th the cost per learning gain — RachelVT42 · 2026-10-01
- A 40-year-old replicator doomsday math, with 'AI' swapped in, mocks today's AI fears — connoraxiotes · 2026-10-01