Doctorow at the Berkman Klein Center: "enshittification" as a three-step account of platform decay, plus warnings about "reverse centaur" AI work and seven firms holding 35 percent of S&P 500 value

Doctorow uses "enshittification" to capture a three-step decay of big tech platforms: they lure users with promises such as respecting the privacy of their data, betray those users to business clients, and then betray those business clients as well in service of their own bottom lines. The term turns scattered complaints about worsening platform experience into a nameable process, which helped it travel: the American Dialect Society named it Word of the Year in 2023, and it has since been added to the Oxford English Dictionary. This is a conceptual argument advanced in the talk; its diffusion is evidenced by the 2023 Word of the Year designation and OED inclusion, with no quantitative data or controlled study reported in the text.

He offers a political economy of Silicon Valley: a mature public company grows slowly, pays regular dividends, and supports stock prices of 10 to 20 times annual earnings per share, whereas tech companies as archetypal growth businesses sustain price-to-earnings ratios twice that high, often on the promise of conquering or absorbing an existing profitable sector such as taxis, trucking, bookstores, or newspapers, and when much of their top talent is at least partially paid in company stock, it becomes vitally important that those promises are believed. It shifts the explanation of platform behavior from product logic to capital markets and equity compensation, explaining why firms are incentivized to sell the public on Next Big Things rather than do one thing well. This is an analytical frame presented in the talk; the specific figures given are the 10-to-20-times earnings multiple for mature firms and twice that for tech firms, with no sample or statistical test provided.

He distinguishes centaur from reverse-centaur experiences: the former endows a human user with superhuman strength, expediting work and improving life, as with corrective lenses, while the latter occurs when employers aiming to lower labor costs force people to use the technology on far-from-human terms, for example a hospital laying off all but a few human radiologists who are left "marking the AI's homework," hunting errors that are devilishly hard to spot and taking the blame whenever mistakes are made. It moves the AI discussion from whether the technology can change the world to how labor relations are organized, showing that the same technology can produce opposite humanistic outcomes under different power arrangements. The hospital radiology scenario is explicitly framed in the text as Doctorow's speculation, an illustrative hypothetical rather than an empirical finding.

He downplays existential AI risk, arguing that LLMs are at bottom "word-guessing machines" that still cannot research themselves, remain relentlessly error-prone, and are "nowhere near recursive self-improvement"; his own worry is the firms' growing centrality in the unequal global economy, since the seven companies making big AI bets now represent 35 percent of the total value of the S&P 500, value that could be "vaporized overnight, when investors lose confidence," followed by financial crisis, austerity, and a further lurch into hard-right politics. It relocates risk from model capability to financial concentration and political consequence, and draws on science fiction as an "anti-inevitabilist" literature to argue that the current arrangement is a choice rather than an iron law of history. The 35 percent figure is Doctorow's statement in the talk; the subsequent chain to financial crisis, austerity, and political shift is his inference, with no empirical test reported in the text.

This report is aimed at readers following platform governance, antitrust, and the labor effects of AI, and it is meant to supply "enshittification" and "reverse centaur" as vocabulary for discussing platform decay and the power structure of AI deployment. It offers the views and frames of a live talk, useful as a starting point for classrooms, policy discussion, and product-ethics debate rather than as a citable empirical conclusion.

The text offers no systematic verification of the three-step enshittification process and no quantitative support from user-satisfaction or platform data; the 35 percent market-value share and the chain to overnight vaporization are statements and inferences from the talk, so readers making investment or policy judgments should still consult original market data and fuller research. The hospital radiology scenario is explicitly labeled speculation and should not be treated as an industry fact that has already occurred.