AI teams can split up work. Who sets the limits?
Ethan Mollick describes AI systems that use tools and divide research work among themselves. His account includes a system that caught a permit number error. Those examples show what might be possible, not that such teams work safely in every workplace. Set clear limits and a stop rule before an approved test.
What the source reports
Ethan Mollick once thought people had to design how a team of AI systems would divide its work. In his new essay, he says capable systems can choose research subtasks with less human planning. He tells of a personal AI tool that caught an error in a permit number. He also discusses a large math effort by OpenAI that used many AI systems. The company’s announcement supports the reported scale of that effort. It does not show that the same approach works in ordinary workplaces. Mollick’s examples are an account of practice, not a safety trial. The human role is still to choose a useful goal, set limits on what the systems may do, and check their results.
Original source
- Title
- The Dot and the Swarm
- Author
- Ethan Mollick
- Publication
- One Useful Thing
- Date
- Thursday, October 1, 2026
Also covered by the other newsletter.