In 1975, computer scientist Frederick P. Brooks Jr. formulated 'Brooks's Law' in his seminal book, The Mythical Man-Month. The law states that 'Adding manpower to a late software project makes it later,' a concept often illustrated by the 'Baby Problem'—the idea that nine women cannot deliver a baby in one month. This principle stems from the inherent overhead of human collaboration: the time required for onboarding new team members, the friction of constant meetings, and the complexities of office politics. In traditional software engineering, these factors create a ceiling on how much adding personnel can actually accelerate a project.
The rise of AI agents and Large Language Models (LLMs) appears to challenge this law by removing these human-centric bottlenecks. An AI agent can be 'onboarded' in milliseconds, does not require coffee breaks, and can process information without the social friction of a human team. However, Jan van den Berg argues that the law has not vanished; it has mutated. While the overhead of managing people has decreased, the overhead of managing agents has emerged. The bottleneck has shifted from human coordination to the necessity of defining 'done,' reviewing code for hallucinations, and stitching disparate pieces of output into a cohesive system.
This shift introduces a new risk: the 'high-speed loop.' Without crystal-clear orchestration and requirements, multiple autonomous agents can work on a single codebase simultaneously, potentially rewriting each other's work in a rapid cycle of conflicting changes. Van den Berg suggests a 2026 update to the law: 'Adding agents to a project with unclear requirements gets you the wrong answer faster.' While agents excel at 'grindy' tasks like reverse engineering or performance optimization, they struggle with complex architectural choices. The primary challenge for modern engineers is no longer just technical execution, but the ability to provide the precise orchestration and requirement clarity needed to prevent AI-driven velocity from becoming a liability.
Furthermore, the role of the individual engineer is changing as a result of these tools. Rather than working on a single ticket to completion, engineers are increasingly able to manage multiple sequential tasks in parallel by leveraging AI to handle the heavy lifting of context switching. This creates a new distinction between median and great engineers: the ability to maintain a high-level architectural view while managing the vast amount of context generated by AI-assisted workflows. The ultimate constraint remains the clarity of the initial problem; without it, the speed of AI only serves to accelerate the accumulation of technical debt or incorrect implementations.
Ultimately, while AI agents can drastically shorten the time required to build specific components, they do not solve the fundamental problem of project scope and requirement definition. The 'Mythical Man-Month' remains relevant because the core difficulty of software—deciding what to build and how it should behave—cannot be automated away. The speed of the tools has increased, but the necessity of human oversight to ensure those tools are moving in the right direction remains the primary bottleneck of the modern era.
The primary source establishes that the 'Baby Problem' remains because sequential tasks are still sequential, and the management of agents requires a human to catch hallucinations and define the final product. It highlights that the 2026 update to Brooks's Law focuses on the speed of producing incorrect answers. Discussion on Hacker News adds that agents are particularly useful for 'grindy' tasks like reverse engineering but are less effective for complex architectural problem spaces. It also notes that engineers are now working on many sequential tasks in parallel, with the ability to context switch without burnout becoming a key differentiator for high-performing developers.
The source material provides a theoretical and anecdotal framework for these shifts rather than empirical project data. While it identifies a clear transition from human-to-human management to human-to-agent orchestration, the specific 'speed' of this new workflow is not quantified. The transition from 'man-months' to 'half-days' for certain applications is presented as a possibility of the technology, but the actual trade-offs in project success rates when using agent swarms versus human teams remain an open question for the industry to explore.
The primary source establishes that the 'Baby Problem' remains because sequential tasks are still sequential, and the management of agents requires a human to catch hallucinations and define the final product. It highlights that the 2026 update to Brooks's Law focuses on the speed of producing incorrect answers. Discussion on Hacker News adds that agents are particularly useful for 'grindy' tasks like reverse engineering but are less effective for complex architectural problem spaces. It also notes that engineers are now working on many sequential tasks in parallel, with the ability to context switch without burnout becoming a key differentiator for high-performing developers.
The source material provides a theoretical and anecdotal framework for these shifts rather than empirical project data. While it identifies a clear transition from human-to-human management to human-to-agent orchestration, the specific 'speed' of this new workflow is not quantified. The transition from 'man-months' to 'half-days' for certain applications is presented as a possibility of the technology, but the actual trade-offs in project success rates when using agent swarms versus human teams remain an open question for the industry to explore.