AI agents are learning to organize work without human-built workflows
Ethan Mollick argues that agents may not need managers to design their workflows: newer systems can learn to organize work themselves. He points to Meta’s Muse and OpenAI’s dots, which connect to a computer and accounts, analyze data in real time, and let users interact through messages while the agents also reach out proactively. The shift undercuts elaborate prompts and hand-built coordination systems.