Brooks knew this from experience. Working on IBM’s 360 mainframe operating system project, he watched software organizations collapse under their own complexity. Every new worker contributed exponentially to communication costs. New people needed training, and ramp-up time means they are slow to produce. Existing workers had to stop what they were doing to train the newcomers — a double whammy that compounded with every new hire.
For 50 years, no one found a way around it. Of the 66 unicorns (startups worth over $1 billion) that were flush with cash in 2021, 30 haven’t raised funds since, and 11 have raised at lower valuations. Although other factors were undoubtedly at play, this is yet another data point that illustrates productivity can not be bought simply by hiring more engineers.
Then, in 2022, something changed.
Since 2023 a new set of laws have begun to govern how capital gets deployed, ones that more or less render the Mythical Man-Month* irrelevant. This is apparent if you look at companies pouring capital into AI models and seeing immediate returns in research and model capability. Model companies have managed to deploy more capital with smaller teams and produced outsized revenue growth as a result. In fact our internal data show that the larger AI companies have nearly three times the revenue run rate per full-time employee as non-AI software and tech companies.
The reason runs deeper than tooling or workflow efficiency. Modern AI approaches have evolved to draw from large amounts of compute rather than complex engineering, which means the old coordination problem resulting from complexity largely disappears. Rich Sutton famously captured this in his 2019 “Bitter Lesson” essay, arguing that simple algorithms leveraging powerful computers consistently outperform clever algorithms built on “domain-specific” human knowledge. When Sutton wrote the essay in 2019, there was no ChatGPT, and no hundred-million-dollar training runs for developing advanced models. The subsequent rise of frontier AI has since validated his argument more dramatically than perhaps anyone expected.
Brook’s long-standing observation applies to building traditional software. Developing AI has turned out to be quite different. Rather than requiring large teams across multiple subsystems that need to coordinate, AI models are developed by smaller teams whose output increases in quality as a function of the data and compute thrown at them. The upshot is something Brooks would have found almost unimaginable: capital can finally be deployed at speed, and the relationship between investment and output is far more direct. To wit, now you can throw money at software engineering in order to get more output. That is, if you’re building AI models that end up doing the work we’d otherwise use traditional software for.
This is playing out in private markets as companies raise historic amounts of money, with historically small teams, and enjoy historic growth. OpenAI, Anthropic and Cursor have grown from a few million dollars in revenue to billions in under two years.
There’s also a change in what it takes to win. For a long time, the answer to the Mythical Man-Month* was better leadership and stronger organizational culture. Better-managed teams got the better of rivals by executing faster and more efficiently with the same amount of capital. But recently, AI moved the bottleneck from people to compute, and managing great teams at scale now matters less than it used to.
Brooks’s constraint was always on the supply side: you simply couldn’t build great software companies fast enough to meet demand for them. The same logic extended to venture capital: funding was abundant, but the great companies that could absorb it were not. This pattern has been observable across cycles, with returns concentrated in a small number of outliers and no amount of fundraising changing how many of those outliers exist in any given era. But the outlier scarcity was never about ideas or capital. It was about what Brooks had seen: you couldn’t scale companies on demand. Change that, and you change the scarcity.
What Comes After The Mythical Man Month
The implications of this shift will be profound. Returns will accrue to those who can deploy capital quickly and efficiently, and not just those with the most consumer insight, hustle or leadership prowess. For customers and investors this might mean a lot more opportunities to build generation-defining companies without facing fundamental limits to scaling.
The rate of change across the software industry and everything it touches will only continue to accelerate. And the range of things that software can touch will increase as well. Before AI, engineering was rate-limited by the diminishing returns of throwing more programmers at a problem. With AI, the world has figured out how to get around that.
Brooks identified the trap that the software industry spent over fifty years navigating. The Mythical Man Month was thought to be insurmountable. But in the age of AI, it might simply be about a large enough compute budget and a small enough team that knows when and how to use it.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。