It still shipped three years late, and bloated when it arrived.
The project began in 2001 and was scheduled to end in 2003. Features kept being added, each tangled into the last, until the codebase could barely be compiled. By 2004, Microsoft did the only thing left. It stopped, scrapped much of the work, and cut the scope roughly in half. Even then, Vista did not ship until 2007.
The instinct behind this is older than the company.
In 1975, Fred Brooks, who had run one of the largest software projects of its era, published The Mythical Man-Month. His law is one line: adding people to a late project makes it later.
The reason is arithmetic. Output rises by one with each person you add. The lines of communication between them rise with the square of the number. Twenty people are not twice the coordination of ten. They are four times it. And every newcomer has to be taught by someone who was, until that moment, doing the actual work.
There is a limit to this. Work you can cleanly split survives more hands. Ten people pick cotton faster than one. But nine people cannot deliver a baby in a month, and most real projects sit closer to the baby than the cotton.
McKinsey and Oxford studied 5,400 large IT projects. On average, they delivered 56% less value than promised.
The 2026 version wears a new costume: add more AI agents and ship faster. The agents do write faster. They do not shrink the cost of making all that output agree.
A late project is rarely short of hands. It is short of a smaller plan.
When a deadline slips, is your first instinct to add people or to cut scope?
