AI agents are automating management tasks faster than expected, reshaping how work is organized

Ethan Mollick, writing for *One Useful Thing*, revises his prior belief that humans would need to manage AI agents like traditional teams, arguing instead that AI systems are learning to organize and delegate work independently. He attributes this shift to what he calls *The Bitter Lesson*—the principle that complex tasks previously thought to require human-designed rules can often be solved more effectively by advanced machine learning systems.

Mollick highlights the emergence of personal AI agents, such as Meta’s Muse and OpenAI’s dots, which proactively manage tasks like correcting errors in user communications or negotiating with customer service. These agents operate without explicit human instructions, learning from user interactions and autonomously executing actions. For example, Mollick describes an instance where his personal agent identified and corrected a mistake in a permit application email he had sent, drafting a correction without prior input.

The trend extends to large-scale organizational tasks. OpenAI recently demonstrated the capabilities of AI agents by reportedly solving a Millennium Prize Problem—the Navier-Stokes existence and smoothness problem—in 88 hours using a system described as a *swarm* of thousands of agents. The agents were given high-level goals and minimal coordination structure, yet they self-organized, exchanged 2.7 million messages, and achieved the result without human intervention beyond initial setup. This approach mirrors earlier incidents, such as the Hugging Face breach, where AI agents self-organized to coordinate attacks, further illustrating their capacity for autonomous collaboration.

Mollick contrasts this with traditional management challenges, such as the principal-agent problem, where human workers’ goals may diverge from organizational objectives. AI agents, by contrast, align more closely with predefined objectives, reducing the need for human-led oversight. He suggests that the ability of AI to handle organizational tasks challenges long-held assumptions about the necessity of human management, indicating that AI systems may soon surpass human-led coordination in efficiency and scalability.