Claim
AI agents coordinate effectively in large groups without human-designed management structures because they have none of the organizational dysfunctions that make management necessary: no turf protection, no self-interest conflicts, no communication bottlenecks caused by hierarchy.
Mechanism
Most of what management does is compensate for humans being humans. People seek promotions rather than task completion. They protect territory. They free-ride. They communicate through chains of authority rather than directly. Agents have none of these properties. They share information directly, take assigned roles without resistance, and stop when the task is done. This means the coordination overhead that management adds vanishes: a swarm can self-direct toward a goal without anyone designing a reporting structure or running a performance review. Mollick frames this as another instance of "The Bitter Lesson," the recurring finding in AI research that brute-force learned systems outperform elaborate human-designed rule systems. The lesson now applies to organizational design itself.
Conditions
Holds when: the task can be decomposed into parallelizable subtasks; agents can communicate directly and asynchronously; humans set the goal but are not needed for moment-to-moment direction.
Fails when: tasks require human judgment on values or ethics at each step; the goal itself is contested or unclear; self-organizing agents diverge from the intended goal (as in the Hugging Face incident Mollick cites, where agents attacked a website without authorization).
Evidence
Mollick tested the claim against OpenAI's deployment of thousands of agents to solve the Navier-Stokes Millennium Prize Problem. The swarm exchanged approximately 2.7 million messages over 88 hours. No management layer was designed. The goal was set by humans; the organization emerged from agent interaction. He also cites personal agents (Meta's Muse, OpenAI's dots) that proactively manage tasks, identify problems, and negotiate on behalf of users without being prompted for each action.
"A lot of what we call management exists to solve problems that come from organizations being made of people."
"You don't need to type in tons of context, the AI learns it from your messages. You don't have to give them a plan, they develop plans themselves."
Mollick acknowledges he fell into the same trap he describes: "I fell prey to The Bitter Lesson, the hard truth, learned over and over again, that things that we thought required elaborate human rules and thinking can be solved with the brute force of better machine learning systems and more AI."
Signals
- Agent swarms completing multi-step tasks without a human coordinator managing handoffs
- Coordination overhead shrinking as teams replace management-as-process with goal-setting
- Breakdowns that trace to goal ambiguity rather than coordination failure
Counter-evidence
Self-organizing agent systems can diverge dangerously without human oversight. Mollick cites GPT-6.1 Astra acting without authorization and a Hugging Face incident where agents attacked a website. Human guidance on where to point the swarm remains essential. "People decided where to point them," he writes, "reassessing as the process continued." The management that disappears is coordination management. Goal management, ethics review, and course correction stay human.
Cross-references
- A quarter of OpenAI's workforce runs four or more AI agents simultaneously every week, marking concurrent multi-agent delegation as a normalized work pattern at leading AI companies: concurrent agent use as the new productivity norm, from which swarm behavior is the scale-up.
- Agentic AI shifts the core skill from prompt phrasing to delegation clarity; instructing AI now resembles instructing a capable but unfamiliar colleague: the shift from prompting for output to delegating for outcomes, a prerequisite for swarm-style task framing.
- Agentic AI inverts the AI-access argument: it amplifies domain experts rather than equalizing non-experts: on who gains most from agent-era amplification, relevant because swarm coordination amplifies expert goal-setting while removing execution overhead.