AI raises the bar for everyone while the rewards flow upward to the same place they always have. There's an old factory story: a worker figures out how to assemble parts faster, feels clever, tells the line manager — and the answer is "great, this is now the minimum standard for everyone." Nothing changes for the worker except the expectation. AI is doing this at civilization scale. Everyone will get paid the same, but the output demanded is now 10x higher. The clever worker's reward is a heavier baseline.
Unless you own the means of production, you don't capture the value. But even owners fight other owners who are also running at 10x, using the same tools, accessing the same models, burning through the same APIs. The competitive advantage collapses into whoever can think clearest and move fastest — and even that edge is temporary because everyone's learning curve is compressing toward the same ceiling.
Building AI tools is an easy way to get wiped. The frenzied AI-native startups are an army of hopeful founders, each sprinting on the same treadmill, building wrappers around the same three APIs, praying their UX moat holds for six months before the foundation model just absorbs their feature. It's a race that ends with everyone collapsing in exhaustion without anyone winning.
The deeper irony: everyone is using the same 3-4 giant mega datacenters. Every startup, every solo builder, every enterprise — all flowing through the same pipes. The data pouring into Anthropic, OpenAI, and Google's servers 24/7 is orders of magnitude beyond what Google Search ever captured. Not just queries — entire therapy sessions, personal finances, business strategies, full startup codebases, architectural decisions, private conversations. Everything. All captured. All processed. All stored.
We are writing apps for centralized megacomputers. Each startup is treated by the megacomputer as just another app — another tenant on someone else's infrastructure, another revenue line in someone else's quarterly report. The architecture has come full circle: mainframes in the 1950s, terminals accessing centralized compute, dumb clients depending on smart servers. That's exactly where we are again. The terminals are prettier now — iTerm and Ghostty instead of VT100, ChatGPT instead of TSO — but the topology is identical. Centralized intelligence, distributed dependence.
The 1950s mainframe operators controlled who computed what. Today's megacomputer operators control who thinks what, builds what, ships what. The illusion of independence — "I'm a solo founder with AI" — dissolves when you realize your entire operation runs on three companies' permission and pricing. They can rate-limit you, deprecate your model, change their terms, or simply raise prices. You own nothing except the terminal.