Twenty People Can Now Put a Billion Dollars to Work
Casado argues AI turned software into a problem money can solve.
Invest Like the Best with Patrick O'Shaughnessy · Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498]To show the old limit, Casado poses a hypothetical: five founders ask him for a billion dollars. He says the money would sit there, or the founders would detonate by hiring a crowd of engineers and be worse off.
Casado is an investor who previously ran his own software startup for a decade. He argues that artificial intelligence has broken one of technology's oldest limits.
The reason is an observation about software teams called the mythical man month, the title of a 1975 book by Fred Brooks. It says adding people to a software project can make it slower.
Adding programmers did not speed the work. It slowed the work, because every new engineer added communication and coordination. Output grew much more slowly than headcount.
Casado says he lived that tradeoff. Hiring a crowd of engineers to move faster leaves a team really in trouble.
Casado argues AI changes the equation because digital work is no longer capped the same way by human coordination. He says a team of 20 can now put a billion dollars to productive use, where before engineering was the bottleneck.
Whether that spending pays is still unknown. He has said he has no idea whether a billion dollars deployed this way returns $500 million or $2 billion.
But the ability to absorb it is new, and large. Casado says Anthropic and OpenAI have probably raised more than the rest of the downstream ecosystem combined, and sees them as ways to turn heavy spending into expansion.
Questions people ask
Why does adding engineers slow software down?
Each new programmer adds relationships the team must maintain. With 10 people there are dozens of possible conversations, with 100 people there are thousands, so time shifts from writing code to staying aligned. That is why Brooks warned that adding staff to a late project can make it later.
What does a small AI team actually buy with a billion dollars?
Mostly it buys computing and the work around it. Teams rent large clusters of chips, pay for data and electricity, and run long training and testing cycles. Unlike salaries, those purchases do not require the same constant coordination.
Does spending more guarantee better results?
No. More spending can produce more capable systems, but customers may not pay enough to cover the cost. Casado puts the uncertainty plainly in terms of whether a billion dollars returns $500 million or $2 billion.