The Reverse Centaur's Guide to Life After AI

Cory Doctorow opens with a story that encapsulates the book's central argument. A Chicago freelance writer named Marco Buscaglia became an internet laughingstock after his syndicated summer reading list, produced for King Features Syndicate (a division of Hearst Communications), turned out to contain ten nonexistent books generated by a chatbot. Tech reporter Jason Koebler of 404 Media investigated and found that Buscaglia had been tasked with writing the bulk of a sixty-four-page supplement, work that would ordinarily require teams of writers and fact-checkers. Doctorow argues Buscaglia was set up to fail: His real job was to absorb blame for the chatbot's mistakes.

Doctorow contrasts this with his own use of Whisper, an open-source speech transcription model he downloaded to transcribe dozens of hours of podcasts on his laptop. The tool worked efficiently and gave his computer a useful permanent capability. This contrast introduces the book's organizing framework, drawn from automation theory: a "centaur" is a person assisted by a machine they choose to use, while a "reverse centaur" is a person conscripted into serving as a machine's assistant. Doctorow was a centaur; Buscaglia was a reverse centaur. The same distinction applies to Amazon warehouse workers monitored by AI cameras that penalize them for "time off-task," and to delivery drivers scored by algorithms that dock points for braking to avoid hazards. The key question about any technology is not what it does but who it does it for and who it does it to.

Doctorow positions the book against "inevitabilism," the claim by tech executives like Mark Zuckerberg and Apple CEO Tim Cook that their abusive practices are simply how things must be. Science fiction, he contends, is an anti-inevitabilist literature: By imagining many possible futures, it demonstrates that the current arrangement is a choice, not a law of nature.

In the first section, "Before," Doctorow introduces "criti-hype," a term coined by Virginia Tech scholar Lee Vinsel for criticism that inadvertently amplifies hype. When critics breathlessly repeat a tech company's exaggerated claims, they help sell the products they oppose. He traces this pattern from the 1950s moral panic over subliminal advertising through present-day surveillance advertising, arguing that the same companies that profited from ad tech are now driving the AI bubble. Using the history of the cash register, Doctorow illustrates that technology's social consequences depend on power, not on the technology itself. In a unionized workplace, a labor-saving tool benefits workers; without such protections, bosses use it to cut wages and lay off staff.

The chapter "Really Weird Math" forms the book's economic core. Doctorow explains how Microsoft and OpenAI inflate their AI revenue figures through a circular accounting arrangement involving data-center access tokens. He explains why tech companies obsess over growth: a "growth" company's price-to-earnings ratio is far higher than a "mature" company's, granting advantages that disappear the moment the market perceives the company as mature. He cites Meta's $230 billion single-day stock loss in 2022 and Nvidia's $600 billion loss after the Chinese AI model DeepSeek's release in 2025. To maintain the appearance of growth, companies manipulate Key Performance Indicators (KPIs), a tendency captured by Goodhart's Law, which holds that when a measure becomes a target, it ceases to be a good measure. Doctorow details Facebook's "pivot to video," in which the company allegedly inflated view statistics and convinced the media industry to invest billions in video production, only to stop pushing video and bankrupt media companies. The same logic drives AI: a June 2025 study in The Journal of Hospitality Marketing & Management found that over 90 percent of consumers are less likely to use a product advertised as AI-based, yet companies manufacture engagement numbers by tricking users into triggering AI features. Seven giant AI firms account for 35 percent of U.S. stock market value, making the entire economy dependent on the AI growth story.

The second section, "Bubble," argues that AI's business case rests on displacing workers' wages. Doctorow distinguishes between a "centaur" arrangement for AI radiology, where the AI offers a second opinion and the radiologist checks disagreements, and a "reverse centaur" arrangement, where most radiologists are fired and survivors review AI output at superhuman speed. The latter fails because of "automation blindness": when humans review output that is usually correct, they lose the ability to spot errors. He also describes "slopsquatting," a security threat in which attackers name malicious code libraries to match fabricated names that chatbots commonly generate.

On AI-generated art, Doctorow defines art as a communicative act in which an artist infuses a medium with millions of tiny intentional decisions. AI output is limited because the only human intent is in the prompt, diluted across millions of meaningless pixel-level decisions. He invokes cultural theorist Mark Fisher's concept of "the eerie" to describe AI art as having the appearance of intention without an intender, an effect that fades over time into perceived soullessness.

Doctorow argues that expanding copyright law will not protect creative workers from AI. Four decades of copyright expansion have coincided with media companies growing larger while the share of income going to creative workers has fallen, because concentrated industries extract any new rights through nonnegotiable contracts. Instead, he advocates labor organizing. He recounts the Writers Guild of America's 2023 strike against studios over AI, in which writers won every demand. They secured the right to use AI as they chose, without wage reductions: the right to be centaurs, not reverse centaurs. Doctorow also identifies one existing copyright rule that genuinely benefits workers: Only works of human creativity receive copyright protection, so AI-generated outputs enter the public domain. If companies replace workers with AI, they lose copyright over their products.

In "Actual, Existing AI," Doctorow dismisses fears of superintelligent AI and turns to real-world harms. He reports an MIT study finding that 95 percent of commercial AI deployments fail with no measurable profit impact. Klarna, a fintech company, fired its customer service staff and replaced them with chatbots, only to launch a rehiring spree the following year. Australia's Commonwealth Bank declared its chatbot replacement of human representatives "an error" within a month. Doctorow details the "garbage in, garbage out" problem with predictive policing, citing a study by the Human Rights Data Analysis Group (HRDAG) showing that a predictive algorithm amplified existing racial biases in Oakland drug policing, and that feeding biased outputs back into the model caused bias to snowball. He identifies surveillance pricing, where companies use personal data to charge each customer the maximum they will pay, as a genuinely profitable and harmful application of AI.

Doctorow documents the widespread practice of disguising human labor as AI, tracing the phenomenon from the early 1800s Mechanical Turk, a chess-playing "automaton" secretly controlled by a hidden person, through modern examples. Amazon's "Just Walk Out" grocery stores were staffed by Indian call-center workers clicking their best guesses, and Builder.ai, a company valued at over $1 billion, claimed to build apps with AI but secretly employed hundreds of low-waged Indian programmers. Doctorow argues that "agentic AI," a term for AI systems designed to perform complex multi-step tasks autonomously, and self-driving cars both rest on the impossible premise that the world will rearrange itself to accommodate AI's limitations.

The final section, "After," distinguishes between bubbles that leave nothing behind, like Enron and cryptocurrency, and those with productive residues, like the dot-com bubble, which trained millions of technologists. Doctorow assesses AI's unit economics as disastrous: Each generation of models costs vastly more to train and operate. The productive residues he identifies are stand-alone, open-source models that run on personal computers, useful for transcription, image processing, translation, and interactive journaling without the costs or privacy risks of commercial services. He presents the HRDAG's use of AI to process police records, flag false-arrest correlates, quantify racial discrimination in parole decisions, and analyze decades of extrajudicial killings during Colombia's civil war as a model for beneficial AI use. He warns that the bubble's collapse will cause severe economic damage, leaving lasting harms: carbon emissions, workers fired because AI salesmen convinced bosses to replace them, and degraded services. The job of a good AI critic, Doctorow concludes, is to pop the bubble quickly by targeting its financial basis. The future is not inevitable, and the most important fact about any technology remains who it does it for and who it does it to.