The Mythical Man-Month in the Age of AI

Brooks’s Law holds that adding people to a delayed software project makes it later, a lesson Frederick Brooks drew from managing IBM’s OS/360 project and published in 1975. A failed attempt by four people to connect a laptop to a screen prompts a question: does the law still apply when AI agents can be brought up to speed instantly, without onboarding, meetings, or interpersonal friction? Those advantages seem to remove the coordination costs that made adding human workers counterproductive.

The argument is that the law has changed shape rather than disappeared. Some work remains inherently sequential: as the “Baby Problem” illustrates, nine agents cannot complete a nine-month task in one month. Other overhead persists in new forms, because a person must still specify requirements and completion criteria, review code for errors or hallucinations, and integrate agents’ contributions. Without clear orchestration, multiple agents working on one codebase may overwrite or undo one another’s work. The revised lesson is that adding agents to a project with unclear requirements produces the wrong answer faster. AI can accelerate execution, but it cannot make a poorly defined task parallel or remove the need for human judgment and coordination.