Adding people to a late project slows it down: communication grows quadratically, and onboarding takes away the most knowledgeable people. Second-system effect: the second version is the most dangerous — everything that “didn't fit” into the first gets dumped into it — a direct violation of Gall's Law (41) from within the team that wrote the first.
Both come from Fred Brooks, The Mythical Man-Month (1975), drawn from his management of IBM OS/360. The law, in his words: “Adding manpower to a late software project makes it later.” The mechanism has three parts. Work that cannot be partitioned does not speed up with more people at all. New people need training, paid for by the people who already know the system. And if each pair must coordinate, n people have n(n−1)/2 channels: 5 people have 10, 10 have 45. Brooks introduced the law with the words “oversimplifying outrageously”; it is an observation from experience with a plausible mechanism, not a measured constant.
The second-system effect is the chapter of the same name: an architect’s first system is spare because they are unsure; the second carries every idea deferred from the first, and generalizes the parts that worked. In practice it is the rewrite pitched as “v2 with everything we learned”: a new data platform that will also unify batch and streaming, add multi-tenancy and replace the scheduler — and that has to reach parity with the old one before it delivers anything.
Limits. Adding people early, to work that partitions well, with low ramp-up, does help — the law is about late projects and tightly coupled work. Team boundaries that cut communication paths (15) change the quadratic term. Not expressible in code; expressed instead as a decision to “cut scope, don't hire” and as a requirement that rewrite proposals go through a strangler fig (73) rather than “this time we'll do it right”.