Technology transitions create more of the 'dying' thing, not less

PCs were supposed to kill mainframes — both grew massively. Amazon was supposed to kill physical retail — both Amazon and Walmart became trillion-dollar companies. Digital was supposed to kill traditional media — there’s vastly more media today. The pattern repeats: old thing grows larger, new thing adds to ecosystem, transition takes a generation (not five years), and there are multiple winners.

Sinofsky argues AI will follow the same pattern: more software than ever (we’re nowhere near meeting demand), AI moves up the stack as part of everything, domain expertise becomes more important not less. Nader Dabit provides the mechanism: Software abundance unlocks entire categories of applications that never existed because software has always been more expensive than we can afford to build, and when costs drop 10-20x, the implementation gap collapsed at the individual level scales to the entire industry.

The timeline tension is real: Shumer and Dabit say it’s happening NOW, Sinofsky says transitions take a generation. Both are right — capability is here, but adoption is uneven. This matters for the middleware-versus-infrastructure question — the middleware isn’t dead yet, but the direction is clear. The “more” takes a specific form in vertical software: LLM competition fragments markets from 3 incumbents to 300 — you don’t go from 3 incumbents to 4, you go from 3 to 300, as lowered engineering barriers create more competitors, not fewer. The data labeling market is a live example: Commodity work's terminal value is zero but structured expert judgment compounds indefinitely — commodity annotation is dying (Appen collapsed from $4.5B to $140M), but expert alignment infrastructure grew to $29B (Scale AI), exactly the “old thing shrinks, new thing grows larger” pattern.