The Ledger That Bet Wrong on the Right Line

Brynjolfsson and McAfee draw a boundary in Chapter 12 that reads, in 2026, like a warranty that expired early. Ideation, they write — "coming up with new ideas or concepts" — is the one thing computers cannot touch. "We've never seen a truly creative machine, or an entrepreneurial one, or an innovative one." Software can produce lines of English that rhyme, they concede, but "none that could write a true poem." Attempts at software writing software, they add, have been "abject failures." This is not a hedge buried in a footnote. It is the load-bearing claim of the book's most optimistic chapter, the one arguing that humans and machines will race together rather than machines simply winning. Twelve years on, code-completion tools write large fractions of production software at companies that employ actual engineers to supervise them, and language models generate verse indistinguishable from competent MFA output on command. The authors weren't naive — they hedge everywhere else, citing Moravec's paradox, insisting nothing should be "treated as gospel" — but on this one axis they staked the whole argument for human economic relevance on a line that moved faster than any other prediction in the book.

Elsewhere the ledger balances better than it has any right to. The chapter on superstar economics — Instagram's fifteen employees, the "fractal-like quality" of income concentration among the 0.01 percent, network effects turning platforms into winner-take-all arenas — reads now less like forecast than like an early field guide to a world that has since fully arrived. What the book called a "canyon" between first and second place has become the entire architecture of the creator economy, where a YouTuber with ten million subscribers and one with fifty thousand occupy different economic universes despite doing structurally identical work. The book's instinct that digitization would produce a "fractal" distribution rather than a simple gap has proven more accurate than most of its labor-market forecasts, because it correctly identified that the mechanism was mathematical (low marginal costs, network effects) rather than merely cultural.

The policy chapters are where hindsight turns least kind, though not for the reason you'd expect. The authors consider and reject a universal basic income, preferring Milton Friedman's negative income tax because it preserves "the incentive to work" — treating labor-force attachment as a good in itself, worth protecting even at some efficiency cost. In 2014 this was a defensible technocratic compromise. It reads differently now that the loudest advocates for basic income experiments are the AI lab executives building the systems the book worried about — Altman funding UBI pilots, DeepMind researchers signing open letters about post-labor economics. The book treated basic income as a fringe idea "not part of mainstream policy discussions," worth a respectful paragraph of intellectual history running from Thomas Paine through MLK. It is now a live pilot program funded by the very industry the book was trying to reassure us about. The people closest to the automation the authors describe have quietly concluded that the negative income tax's faith in the dignity of continued labor-force participation may not survive contact with what they are building.

Where the book sits in its own lineage is instructive: it shares a shelf with Bastani's automated-luxury-communism optimism and Markoff's augmentation-versus-automation history, both of which also treated the human-machine relationship as something that could be designed rather than something that would simply happen to everyone at once, at scale, without much design input from anybody. The "second machine age" as a coinage never really stuck — we don't talk about mental power the way the Victorians talked about steam, we talk about "AI," a term that flattens the book's careful distinctions between routine and non-routine cognition into a single undifferentiated anxiety. That flattening is arguably the book's most important unintended lesson: it spent three hundred pages building a taxonomy of what machines could and couldn't do, and the taxonomy didn't survive a single scaling law. The question the book leaves open, and cannot answer from where it's standing, is whether its own negative income tax — its whole policy architecture of subsidizing labor to keep people economically legible — still makes sense once the "new idea" clause, the last exemption the authors were willing to write into the human contract, turns out not to hold either.