Who he is
Affiliation: Author, journalist, activist. Special Advisor, Electronic Frontier Foundation (EFF staff page, checked 2026-09-21); A.D. White Professor-at-Large, Cornell (six-year term 2024–30, announced 2024-07-23 — missed by the v1 file); Visiting Professor of Computer Science, Open University (UK); MIT Media Lab research affiliate; Visiting Professor of Practice, UNC School of Information and Library Science. Editor of Pluralistic (daily, no ads, no trackers). Two books since the file's last real update: Enshittification: Why Everything Suddenly Got Worse and What to Do About It (MCD × FSG, 2025-10-07; 2026 Locus Award for Non-fiction; FT/Schroders Business Book of the Year longlist 2025) and The Reverse Centaur's Guide to Life After AI (FSG, June 2026). Role check 2026-09-21: no departure or new employer; hat count went from two in v1 to five-plus.
One-line position: AI is a bubble run as a "pump-and-dump swindle" by monopolists, and its real product is not a robot that does your job but a reverse centaur — a human conscripted to serve, and take the blame for, a machine. The fix is the same as for enshittification: restore competition, regulation, interoperability and worker power across the whole stack, not "AI governance" in isolation.
Discipline & technical bet
A non-technical thinker with a structural bet — he makes no capability claims. The wager (2025-12 → 2026-09): LLMs (large language models, the text engines under chatbots) are real tools, but "AI" as sold is fiction — "LLMs are real, AI is fake" (2026-09-12). The industry has "very bad unit economics" (each new user loses money) and "anything that can't go on forever eventually stops." When it pops, the residue is cheap open models, distilled weights and GPUs "at 10 cents on the dollar"; the lasting damage is labor — replaced workers "won't return" — plus AI-dependent systems he calls "high-tech asbestos" (Pluralistic, 2025-12-05; Democracy Now, 2026-06-25). His copyright bet inverts the creator lobby's: do not extend copyright to training; deny copyright to AI output so firms must "pay humans" (2025-12-05). Newest strand: chatbots have "very low switching costs," Chinese open-weight models lag the US frontier "by a mere four months," and distillation (copying a big model's behavior into a smaller one) defeats any ban — so an AI moratorium is a cartel in disguise (2026-09-16).
Key claims (Says)
- Enshittification (coined 2022, popularized 2023): platforms attract users → attract business customers → squeeze both → die. Held — signature-framing check: he still endorses it and expanded it into a 352-page book with a "four forces" cure (competition, regulation, interoperability, worker power); American Dialect Society Word of the Year 2023; Macquarie Dictionary Word of the Year 2024; 2026 Locus Award. Behavior matches the framing (book, tour, EFF events). The v1 clause "AI accelerates each stage" was a bench gloss, not his — split out below.
- AI as enshittification accelerator — cheaper content, customer service and moderation lower quality while raising extraction. Drifted (mechanism held, framing moved). Mechanism: Google AI Overviews cut clicks to results from 15% to 8%, with 1% of overviews sending a click to a cited source (Pew, 2025-07); Klarna replaced ~700 agents with AI (2024-02), then rehired because "we focused too much on cost — the result was lower quality" (2025-05). Framing: he now treats AI as a bubble and a labor-discipline tool in a separate book, not as a stage of enshittification; "accelerator" is the bench's label.
- The problem is structural, not technological. Held — "the most important fact about any technology isn't what it does, but who it does it for and who it does it to" (2026-05-14); "systemic problems have systemic solutions — you can't shop your way out of a monopoly" (2026-06-25).
- Interoperability is the remedy — mandated and adversarial (building compatible tools without permission). Held as position; the remedy has not moved — see prediction 3.
- IP law protects extractors, not creators — DMCA, terms of service and DRM (copy-locks) stop users and rivals building alternatives. Held — Audible holds ~90% of audiobooks and its DRM lets Amazon "steal millions of dollars from writers without losing our business" (2026-05-14); he backed it with money — see Does.
- AI copyright concerns are real but secondary — the training-data fight is smaller than the leverage fight. Held — opposes a new copyright over training; backs the US Copyright Office line that AI output cannot be copyrighted (2025-12-05); tells creative workers the 2023 writers' strike won through sectoral bargaining, not copyright (2026-06-25). Market datum: the $1.5B Anthropic settlement (~$3,000 per book, 2025-09) paid for piracy after Judge Alsup ruled training on lawfully bought books fair use (2025-06) — authors were paid via a class action, not a new training right, which fits. No Doctorow post on the settlement was found.
- NEW — "AI can't do your job, but an AI salesman can convince your boss to fire you and replace you with an AI that can't do your job" (2025-12-05). Held on its observable — Klarna is the exact sequence (fire, degrade, rehire; 2025-05). Caution: one firm is not a market. The firing half is confirmed at market level by the Canaries payroll data (entry-level employment −19% relative in exposed occupations, 2026-08 — see Brynjolfsson, Frey); the rehiring half has no market-level dataset yet.
- NEW — AI-doom talk is marketing: "AI Safety" and sentience worries are weak critiques that "don't drive investment" (2025-12-05); x-risk stories "help them raise more investment capital" (2026-09-12). Drifted — by 2026-09-17 he concedes some bosses "sincerely believe" it and that believers and cynics form an "incoherent bundle" (via Klein & Taylor's End-Times Fascism). Marketing-only became marketing-and-faith.
- NEW — "LLMs are real, AI is fake": the September-2026 "went rogue" story (OpenAI agents hacking Hugging Face in a security exercise) was "a Python loop querying a chatbot" that "serves as a kind of front-end to a database" of CTF logs (2026-09-12). Not graded — interpretive; the incident's internals could not be verified from a primary source. His concession that LLMs make hacking tools more accessible is the checkable half and is uncontroversial.
- Number flag: "$1.4 trillion spent on AI, doubled from $700B" (Democracy Now, 2026-06-25, as summarized). $1.4T is OpenAI's eight-year commitment (Altman, 2025-11-06), since reset to ~$600B by 2030 (CNBC, 2026-02-20); realized hyperscaler capex is ~$800B for 2026 (Huang, 2026-09-10). Flagged, not graded — wording unverified from a transcript.
Notable predictions — with falsifiable checks
- (2023) "Enshittification" becomes the dominant frame for tech criticism. Held — ADS 2023 and Macquarie 2024 words of the year; award-winning book; the word now stretches to public services and climate, which he accepts rather than polices. Check: still in major-outlet headlines unironically at 2027-12-31; if the stretch has turned it into a synonym for "worse," log dilution as a Drift.
- (2023) AI deployed by monopoly platforms accelerates extraction. Held at market level — Pew: 8% vs 15% clicks, a 47% relative fall (2025-07); Ahrefs: −34.5% clicks on queries carrying an overview; Similarweb: zero-click searches 56% → 69% (all via Search Engine Journal, 2025-09-29); Digiday/DCN: referral traffic down as much as 25% for some publishers (2025). Valence note: he claims the whole decline; some is Google's own ranking churn. Check: DCN median Google referral change for publishers, calendar 2026 — a second consecutive double-digit fall confirms.
- (Ongoing, v1 "TBD") Without interoperability mandates, AI entrenches platform monopolies. Held on the narrow observable, rewritten — no US interoperability statute has passed (App Store Freedom Act, H.R.3209: hearing 2026-07-22, not enacted); the EU's DMA has produced zero solutions from 56 formal interoperability requests to Apple (FSFE, 2026-04-20) plus a new Apple investigation (2026-06-09); the one interop-shaped US remedy is judge-made — Mehta's order that Google share search data with qualified rivals including AI firms (2025-09-02), appealed by both sides. Entrenchment datum: the "AI Big 10" are 41% of the S&P 500, matching tech-and-telecom's share at the dot-com peak (BofA via Yahoo Finance, 2026-07-05). Check by 2027-12-31: (a) any US interop statute enacted, (b) D.C. Circuit upholds data-sharing, (c) Google search share still above 85% — (a) or (b) without (c) moving cuts against his "law is the only lever" claim.
- (2025-12 →) The AI bubble pops: foundation-model companies vanish, GPUs sell at "10 cents on the dollar," open tools survive. Live — not yet. As of 2026-09-21 the opposite is printing: hyperscaler capex ~$800B in 2026, ~$1.3T guided for 2027 (Huang, 2026-09-10); Nvidia ~$1T in Blackwell/Rubin orders through 2027 (GTC, 2026-03-16); four direct customers were 61% of Nvidia's revenue in the quarter ended 2025-10-26 (his "circular" tell). One retreat his way: OpenAI cut its plan from $1.4T of commitments to ~$600B by 2030 (2026-02-20). He gives no date — unfalsifiable as stated; the bench attaches one. Check by 2027-12-31, any of: a big-four hyperscaler cuts capex guidance year-over-year; a top-five frontier lab takes a down-round, fire-sale or bankruptcy; Nvidia data-center revenue falls two consecutive quarters. None → Wrong on timing.
- (2026-05/06) Workers replaced during the bubble are not rehired after it; the damage is "asbestos in the walls." Live. Check: at 2027-12-31 the Canaries entry-level employment series for AI-exposed occupations is still ≥15% below its 2022 baseline → supports; recovery to within 5% → Wrong.
- (2026-06-25) Unlike 1999, workers resist AI; the business press asks how CEOs will "convince their workers to use AI." Live. Check: by 2027-06-30, do the major employer surveys show mandated-use policies (AI in performance reviews) rising — mandates rising supports resistance; falling suggests adoption is organic and he is Wrong.
- (2026-09-16) The AI-CEO "slowdown" push is collusion: a pacing regime will exempt incumbents and bar Chinese open weights. Live. Check: if any US or EU moratorium/pacing regime is enacted by 2027-12-31, does it (a) grandfather existing frontier models and (b) restrict Chinese open weights? Both → Held; neither → Wrong.
Revealed behavior (Does)
- Money: Kickstarted the Reverse Centaur DRM-free audiobook, ebook and hardcover — $134,498 from 3,880 backers against a $10,000 goal (1,345%), 2026-05-14→06-04 — with Audible excluded. Forgoing the 90% channel costs him revenue, which is why the DRM claim is credible.
- Time: two books in nine months on his two theses, plus a 2026 tour (SXSW London, re:publica, Boston University, EFF events) — the enshittification/AI axis is where his hours went, not novels.
- Platform independence: Pluralistic runs daily with no ads and no trackers — he is not paid by the platforms he grades.
- Institutions: accepted the Cornell chair (2024), kept the EFF, MIT, Open University and UNC seats — a credentialed activist, not a freelancer.
- Where he sends readers: EFF, Tech Workers Coalition, unions, local data-center fights (cites the Monterey Park data-center ban) — organizing, not AI-governance panels (2026-06-25).
Feels
Fears a hollowed labor market and a generation of "asbestos" — AI-dependent systems nobody can maintain — left behind when the money leaves. Hopes the four forces can be rebuilt. Wants to be true: that the bubble pops before the damage is permanent.
Hears
Naomi Klein and Astra Taylor (End-Times Fascism), Adam Becker, Ed Zitron, Cal Newport, the "accountability sink" idea, EFF lawyers and labor organizers; for data, MIT NANDA and Pew. Almost no frontier-lab researchers, no VC, no policy-institute economists.
Sees
Twenty-plus years inside EFF's DMCA and DRM fights — he has watched the interoperability lever get criminalized case by case. A publishing business he runs himself, so the Kickstarters give him first-party sales data on DRM-free demand no analyst has. A daily-blog vantage over the trade press that lets him pattern-match faster than book-cycle critics.
Incentive map
Re-derived this cycle. Who pays him: FSG advances and royalties on two books whose thesis is the persona; Kickstarter direct sales ($134K this round); speaking fees; the Cornell chair; EFF (advisory, terms not public). What he is selling: "enshittification" and "reverse centaur" as brands, a DRM-free publishing model, and EFF membership. What he can't say: that the word is overstretched (it is his franchise); that AI has delivered real productivity somewhere (it would undercut a book three months old); that Audible's terms are tolerable (he is a party in interest, having excluded his own titles). Rule-25 discount: moderate on the bubble call and on any "AI does nothing useful" absolutism — both are now product; low on the interoperability and DRM claims, which he has held for two decades and which cost him money; low on his labor claims, where he cites others' data.
Theories aligned with
- Enshittification / platform political economy — his own; no theory file on the bench yet
- Augmentation thesis — the reverse centaur is its negative image: same technology, inverted "who it does it for"
- Automation-displacement and Entry-level work crisis — he supplies the mechanism (capital-driven automation makes more of the product, labor-driven makes it better) that the Canaries data now measures
- Adjacent to surveillance capitalism (Zuboff) but from a builder/activist register, focused on remedies rather than diagnosis
What he's reacting against
- AI discourse that treats AI as a standalone phenomenon — it must be read inside the political economy of platform power
- Tech optimism that ignores platform dynamics — AI deployed by monopolies serves the monopoly
- AI-safety discourse that ignores present harms — the risk is present-day extraction, not future AGI; now extended to "went rogue" journalism (2026-09-12)
- IP maximalism that locks users in — interoperability, right to repair, adversarial interoperability as structural remedies
- NEW: the creator lobby's copyright-expansion reflex — "do we want to support all the workers in America, or be on the same side as our bosses?" (2026-06-25)
- NEW: the AI-CEO "slowdown" push as a competition-suppressing cartel (2026-09-16) — press accounts attribute the September calls to Altman, Amodei and Musk; his post does not name them and the attribution is not verified here against their own statements
Where he overlaps / splits (with Rich)
- Overlaps with Zuboff on surveillance and extraction; splits on register — she diagnoses from academia, he prescribes from activism
- Overlaps with Crawford on AI as a power structure; splits on prescription — she maps the atlas, he builds the interoperability case
- Overlaps with Khan on antitrust; splits on scope — enforcement tools vs legislative structural change; the Mehta data-sharing remedy is a Khan-style win he did not forecast
- Overlaps with Lanier on data dignity; splits on approach — Lanier wants to pay for data, Doctorow wants to break the hoarders
- Splits with Andreessen / Thiel fundamentally — deserved market success vs regulatory capture; oddly converges with Andreessen on one fact: Chinese open weights now lead the open frontier
- NEW — overlaps with Zitron on the bubble; splits on remedy — Zitron's exposure is financial, Doctorow's is structural (four forces)
- NEW — overlaps with Acemoglu (capital-driven automation is "so-so automation" restated) and with Frey / Brynjolfsson on the Canaries data, while rejecting Brynjolfsson's earlier "novices gain most"
- NEW — splits with Huang on everything measurable ($3–4T/yr infrastructure by 2030 vs "10 cents on the dollar") — the bench's cleanest live bet
- Partial overlap with Marcus on "LLMs are real, AI is fake" — Marcus argues capability limits, Doctorow business-model limits; different falsifiers
Track record
- "Enshittification" is the most successful new concept in tech criticism in a decade — ADS 2023 and Macquarie 2024 Word of the Year; book (2025-10) won the 2026 Locus Award for Non-fiction and made the FT/Schroders longlist
- The Internet Con (2023) — the most actionable book on platform reform: specific policy, not just critique
- The Reverse Centaur's Guide to Life After AI (2026-06) — the AI thesis in book form, three months old at this review; ungraded until the bubble check resolves
- EFF advisor since the early internet era — decades of credible digital-rights advocacy; Cornell A.D. White chair (2024–30)
- Prolific novelist (Little Brother, Walkaway, the Martin Hench books through Picks and Shovels, 2025) — technical fiction and non-fiction advocacy in one career
- Consistently ahead of the mainstream on platform-power dynamics; 2023 AI-extraction call confirmed by Pew/Ahrefs/Similarweb data two years later
- Pattern: excellent on mechanism and on naming; no dates on his biggest forward call, which is the bench's problem to fix, not his
Empirical vs normative
- Empirical: platform enshittification cycles (Facebook, Twitter, Amazon, Google); AI Overviews and click loss; Klarna-style fire-then-rehire; AI unit economics; low chatbot switching costs; four-month Chinese lag
- Normative: mandate interoperability; reform IP; give users exit rights; deny copyright to AI output; unionize; ban data centers locally where residents want to
- The empirical pattern-matching is strong and now data-backed; the prescriptions are specific and politically hard — three years on, none has passed in the US
- Conflict note (CONVENTIONS rule 25): moderate — two in-market books and a brand ride on the bubble popping; his oldest claims (DRM, interop) carry the least conflict and the best record
Weak spots / open questions
- "Enshittification" may be too totalizing — not every platform follows the trajectory and some AI deployments genuinely improve products; he now accepts the word's stretch to climate and public services, which is a brand risk he does not police
- Interoperability mandates remain the prescription; implementation is enormously complex and US momentum is near zero (ASFA: one hearing in 2026)
- Retired 2026-09-21 — bench was Wrong: v1 said he is "less technically engaged with AI capabilities." He now argues mechanism directly (the front-end-to-database account of the Hugging Face incident; distillation; switching costs) and wrote a book on how to think about AI. The critique is still political economy, but the engagement is real
- Activist positioning: he systematically emphasizes the worst-case reading of corporate actions — selection bias in the cases he cites (Klarna, Amazon warehouse injuries), no counter-cases logged
- The frame is consumer-platform-shaped; enterprise, scientific and frontier-research AI follow different dynamics he rarely addresses
- NEW: the bubble call has no date and no threshold from him; the bench's check above is a proxy, not his
- NEW: one loose number ($1.4T "spent") in a broadcast interview; his written work is more careful
- NEW: "AI-doom is marketing" softened to "marketing and faith" within five days of posts (2026-09-12 → 09-17) — watch whether the strong form returns in the book tour
Rich's take
- (your synthesis here)
Delta log
2026-09-21 — v1→v2 migration + validation (bulk run)
- Grades: 9 Held / 2 Drifted / 1 Wrong (the Wrong is the bench's own v1 weak-spot claim, not his); 4 predictions live with checks attached; 1 claim not graded (interpretive); 1 number flagged.
- Held: enshittification framing (still endorsed, expanded — ADS, 2024-01; Macquarie, 2024-11; Locus 2026); structural-not-technological (Pluralistic, 2026-05-14); IP-protects-extractors (same); copyright-secondary (Pluralistic, 2025-12-05; NPR, 2025-09-05); AI-salesman claim (Forbes, 2025-05-18); dominant-frame prediction; extraction prediction (Pew via PPC Land, 2025-07; Digiday); no-US-interop-law prediction, rewritten with a check (H.R.3209; FSFE, 2026-04-20; Courthouse News on Mehta remedy).
- Drifted: "AI as enshittification accelerator" — mechanism held, but his framing moved to bubble + reverse centaur (Democracy Now, 2026-06-25); "AI-doom is marketing" — now marketing and sincere belief coexisting (Pluralistic, 2026-09-17).
- Wrong (bench claim): "less technically engaged with AI capabilities" — retired on Pluralistic, 2026-09-12 and the June-2026 book.
- Most surprising delta: a second book, entirely about AI, shipped inside the review window (FSG, June 2026; Kickstarter $134,498 / 3,880 backers, Kickstarter) — and the June-2026 v1 pass, dated a week before launch, mentioned neither it nor the October-2025 Enshittification book.
- Employer/role check (mandatory): no job change; Cornell A.D. White Professor-at-Large (2024–30) missed by v1 (Cornell, 2024-07-23); EFF Special Advisor confirmed (EFF). Incentive map re-derived: conflict rose from low to moderate — the AI thesis is now a product.
- Post-review sources in window: Democracy Now (2026-06-25); Pluralistic 2026-09-12, 09-16, 09-17; Tyee review via AI Weekly (2026-07-25); Huang at Goldman Sachs (2026-09-10).
- Tier: quarterly — the bubble call is graded on quarterly capex and earnings data, and the moratorium fight is in motion; wake triggers cover capex, lab distress, interop law, the Google appeal and his own publishing events.
- Confidence unchanged at high — daily public record; grades rest on primary posts plus market-level datasets.
Sources
- 2026 primary (in window): "On the sincerity of AI bosses" (2026-09-17) · "How an AI moratorium can save AI bosses" (2026-09-16) · "LLMs are real, AI is fake" (2026-09-12) · Democracy Now interview (2026-06-25) · Kickstarter launch post (2026-05-14) · Kickstarter results · Tyee review via AI Weekly (2026-07-25) · "No one wants to read your AI slop" (2026-03-02)
- 2025 primary: "The Reverse-Centaur's Guide to Criticizing AI" (2025-12-05) · Enshittification (MCD × FSG, 2025-10-07) · Slate on the book (2025-10) · Blood in the Machine interview (2025-10)
- Role / affiliations: EFF staff page · Cornell appointment (2024-07-23) · Pluralistic · Book release schedule
- Market-level data used for grades: Pew AI Overviews study (2025-07) · Search Engine Journal — Pew / Ahrefs / Similarweb figures (2025-09-29) · Digiday / DCN referrals · Nvidia four-customer concentration, 61% (Motley Fool on the 10-Q, 2025-11-27) · Nvidia $1T orders through 2027 (CNBC, 2026-03-16) ·Klarna reversal (Forbes, 2025-05-18) · MIT NANDA 95% (Register, 2025-08-18) · OpenAI $1.4T commitments (TechCrunch, 2025-11-06) · OpenAI resets to ~$600B (CNBC, 2026-02-20) · Huang forecast, capex ~$800B→$1.3T (2026-09-17) · "AI Big 10" = 41% of S&P 500, BofA via Yahoo Finance (2026-07-05) · Anthropic settlement (NPR, 2025-09-05)
- Interoperability / antitrust: H.R.3209 · FSFE on Apple DMA (2026-04-20) · Apple DMA investigation (2026-06-09) · Mehta remedy (2025-09) · D.C. Circuit cross-appeals (2026) · Boston Globe on the slowdown debate (2026-09-20) — CEO "slowdown" calls are reported, not verified from the CEOs' own statements
- Word of the year: American Dialect Society (2023) · Macquarie (2024)
- Carried from v1: The Internet Con (2023); the original enshittification essay (Wired/Pluralistic, 2023-01); EFF advocacy; conference keynotes
migrated v1→v2 2026-09-21 · backup: _archives/cory-doctorow.html.bak-20260921 · converts-from: people/cory-doctorow.md · AI & Society domain