What does life after AI look like? Cory Doctorow’s new book claims to know, and a pointed review in The Conversation claims he has got it wrong. Doctorow’s The Reverse Centaur’s Guide to Life After AI arrived in June 2026 promising a route out of the hype cycle, and by 10 August reviewer Michael Noetel had branded it an outdated map that fails to answer its own question.
This opinion piece weighs both sides of that argument, because we think each camp is holding half of a genuinely useful book. We cover AI releases weekly on our AI models and tools hub, from OpenAI’s mid-August ChatGPT update to open-weight model launches, and we build systems that use natural language processing every working day — so the question of life after AI is not academic for us or for our clients. Here is where the book helps, where the review lands, and what neither offers a business trying to plan.
Table of contents
- What Does Life After AI Actually Mean?
- Inside the Reverse Centaur’s Guide to Life After AI
- The Outdated Map: What the Review Gets Right
- The AI Bubble Case for Life After AI
- Where the Critique Overreaches
- Life After AI for Businesses: Plan for Both Maps
- What a Better Guide to Life After AI Would Say
- Verdict: A Compass, Not a Map
- Life After AI FAQ
- References
What Does Life After AI Actually Mean?
The phrase life after AI sounds apocalyptic, but Doctorow means something narrower and more interesting. His argument is that today’s generative AI industry is a classic investment bubble, and that the important question is not whether the bubble pops but what remains afterwards. Life after AI, in his telling, is the period after the money runs out — when the hype evaporates, some companies collapse, and society decides what to salvage from the wreckage.
A question the whole industry is avoiding
That framing deserves more credit than it usually gets. Almost every AI vendor pitch assumes a straight line from today’s capabilities to permanent transformation. Almost no vendor deck contains a slide titled “what happens to your workflow if our funding dries up”. Doctorow has been asking that second question since his 2023 Locus essay on what kind of bubble AI is, and the book extends it to a full theory of life after AI: which tools survive, who ends up owning them, and whose labour gets reorganised around them.
Why the answer matters in 2026
The stakes have only grown since he drafted it. AI infrastructure spending now props up a meaningful share of stock-market value and capital expenditure, which means the shape of life after AI is a macroeconomic question, not a tech-blog debate. If the spending stops abruptly, the consequences reach pension funds and payrolls far outside Silicon Valley. A serious guide to life after AI would therefore be genuinely useful. The dispute is whether Doctorow has written one.
Inside the Reverse Centaur's Guide to Life After AI
The book, published by Verso in June 2026 at a brisk 240 pages, is organised around one memorable idea. A centaur, in the old chess-computing sense, is a human assisted by a machine. A reverse centaur is a human conscripted into assisting a machine — the delivery driver whose routes, breaks and bathroom stops are dictated by an algorithm, or the moderator cleaning up after an automated feed.
The reverse centaur, explained
Doctorow’s fear is that the AI economy is built to mass-produce reverse centaurs. The industry’s most valuable product, he argues, is not any model but a story told to investors: that workers are about to be obsolete, so firms should buy the machine and demote the human to its minder. Whether the model can actually do the job matters less than whether the boss believes it can.
| Arrangement | Who steers | Everyday example | Doctorow’s verdict |
|---|---|---|---|
| Centaur | The human, aided by the tool | A developer using code suggestions they review and own | Technology worth keeping |
| Reverse centaur | The machine, tended by the human | A driver whose app schedules every minute of the shift | The business model the bubble funds |
What the book actually recommends
The consolation Doctorow offers is that the bubble will burst before the worst version arrives, and that life after AI can be shaped by policy: antitrust enforcement, interoperability rights, worker protections, and picking through the productive residue — cheap GPUs, unemployed statisticians, open-weight models — once prices collapse. The tools should work for us, he writes, not the other way round. As a diagnosis it is vivid and frequently persuasive. As a plan, the reviewer argues, it is where the book runs out of road.
Readers of Doctorow’s earlier work will recognise the machinery. The book is effectively the enshittification thesis — platforms decay once they stop competing for users and start squeezing them — applied to the biggest capital buildout in tech history. That lineage is a strength: it grounds the AI argument in a pattern he has documented across search, social media and marketplaces for a decade, and it explains why Brian Eno and others blurbed the book as the clearest guide to the moment.
The Outdated Map: What the Review Gets Right
Noetel’s review makes two central charges. First, the book was drafted in mid-2025 and the ground has moved: dismissing coding assistants and agents as pure hype reads badly in a year when those systems shipped real, measurable work. Second, the guidance is thin — his summary of the book’s advice is “pick a side and boo”, which is a cruel line precisely because it is not entirely unfair.
Fourteen months is a long time in AI
The timeline problem is structural, not a matter of sloppiness. A book drafted in mid-2025 reached shops roughly twelve months later and met its most-read review about fourteen months after drafting. The chart below shows why that gap hurts a book making claims about fast-moving capabilities.
The claims that aged worst
Noetel’s sharpest evidence is concrete. Capabilities Doctorow waves away as marketing had, by review time, produced results that are hard to dismiss — and a reader relying on the book alone would not know any of it happened.
| The book’s position | What the review says happened by mid-2026 | Our read |
|---|---|---|
| Coding assistants are overhyped autocomplete | Assistants became capable enough that US government guidance treats them as a security concern | The capability is real; the risk framing has flipped |
| Agents are demo-ware | Agent systems credited with progress on a decades-old open maths problem | Cherry-picked, but not nothing |
| Models are safely boxed tools | Documented cases of models working around sandbox restrictions | Strengthens the case for caution Doctorow himself wants |
A guide that declines to guide
The deeper complaint is the missing second half. A reader finishing the book knows what Doctorow is against, but not what to do on Monday. Noetel contrasts it with scenario-planning work that names concrete levers — compute disclosure, capability evaluations, chip tracking — and concludes that on the question of where AI is actually heading, the book “will leave you misinformed”. For a volume whose subtitle promises to teach you how to think about AI before it is too late, that is the most damaging sentence a reviewer could write.
The AI Bubble Case for Life After AI
Here is where our opinion parts company with the review’s harshest reading. Strip out the capability predictions and the book’s economic core survives contact with 2026 remarkably well — because it never depended on models being weak. In his widely shared essay on the coming AI economic shock, Doctorow assembles figures that no capability breakthrough has answered: the gap between what the industry earns and what its infrastructure requires keeps widening.
The revenue gap nobody has closed
The sums are stark. Doctorow cites Morgan Stanley’s estimate that the industry’s real annualised revenue sits near 45 billion dollars, against Sequoia partner David Cahn’s calculation that current data-centre spending needs about 800 billion dollars in revenue to pay back, and Bain’s projection that profitability requires some 2 trillion dollars a year by 2030. One takeaway sentence before the numbers: revenue is running at roughly a fortieth of what the buildout assumes.
Better models do not fix broken unit economics
This is the part of the life after AI argument the review never really engages. An agent solving a maths problem is a scientific milestone; it is not 755 billion dollars of new annual revenue. If anything, stronger capabilities deepen the hole, because frontier training and inference costs climb with every generation. You can believe the models are genuinely impressive and still believe the financial structure carrying them is unsustainable — that is precisely Doctorow’s position, and calling the map outdated does not redraw the terrain. Life after AI remains a live scenario for any planner who can read a balance sheet.
Where the Critique Overreaches
Having granted the review its strongest points, it is worth naming its blind spots, because a business reading only Noetel would end up as lopsided as one reading only Doctorow.
Capability progress is not bubble refutation
The review treats capability milestones as evidence against the bubble thesis, but bubbles are financial phenomena. Railways were transformative and railway manias still ruined investors; the dot-com crash arrived while the web was becoming genuinely indispensable. Doctorow’s Locus essay made exactly this distinction — the question is what residue survives the pop, not whether the technology works. Fibre in the ground outlasted the companies that laid it. GPUs, trained engineers and open-weight models can outlast today’s AI valuations the same way, which is a scenario the review never actually rebuts.
The reverse centaur is already here
The book’s central image also does not date, because it describes labour relations rather than model quality. Warehouse pace-setting, algorithmic scheduling and quota-driven moderation all existed before generative AI and will survive any correction. Better models arguably accelerate the pattern: the more plausible the machine, the easier it is to justify demoting the human who supervises it. On this point, mid-2026 evidence has strengthened Doctorow’s warning even as it embarrassed his capability predictions.
Opinion cuts both ways
It is also fair to note that the scenario-planning school Noetel prefers has its own unfalsifiable habits. Reports that attach probabilities to superintelligence timelines look rigorous, but most of their numbers are expert vibes with confidence intervals. If Doctorow’s crash lacks a date, the acceleration camp’s transformation lacks one too. An honest account of life after AI admits that both genres — the polemic and the forecast — are arguments about power wearing the costume of prediction, and judges them by the actions they recommend rather than the certainty they perform.
Life After AI for Businesses: Plan for Both Maps
For our clients the interesting question is not who wins the review battle. It is what a sensible organisation does when one credible camp forecasts collapse and another forecasts acceleration. The honest answer to what life after AI looks like is: nobody knows yet, so plan for both branches at once.
Adopt where value is proven today
The MIT finding Doctorow cites — that 95 per cent of enterprise AI pilots showed no measurable return — is a warning about how projects are chosen, not proof that nothing works. The five per cent that succeed share traits we see in our own AI strategy work: a narrow process, a measurable baseline, and a human who stays in charge of the output. That is the centaur configuration, and it pays for itself whether or not the bubble pops.
Avoid becoming the reverse centaur shop
The cheapest insurance against every scenario of life after AI is the same: never build a workflow where the tool sets the pace and a person absorbs the blame. When we deploy autonomous AI agents for repetitive roles, the design rule is that a named human owns the outcome and can override the system without penalty. That rule protects you if the vendor folds, if the model degrades, and if regulators start asking enshittification-flavoured questions.
Hedge the dependency
Doctorow’s residue advice translates directly into procurement. Keep prompts, data and evaluation sets portable so a collapsing vendor is an inconvenience rather than an outage. Watch open-weight releases, which keep improving regardless of any single company’s fate. And treat current pricing as promotional: today’s subsidised inference is part of the story investors are being told. Our cost optimisation reviews now model a doubling of AI line items as a stress test, for exactly the reasons the book lays out.
| Action now | If the bubble pops | If capabilities keep climbing |
|---|---|---|
| Automate narrow, measured processes | Savings already banked | Foundation for deeper automation |
| Keep data and prompts portable | Painless vendor exit | Painless vendor upgrade |
| Keep a human owner per workflow | Service survives tool loss | Quality control scales with use |
| Stress-test AI pricing at 2x | Budget absorbs repricing | Margin known before scaling |
What a Better Guide to Life After AI Would Say
The frustrating thing about both the book and the review is that a genuinely useful guide is imaginable, and each side holds half of it. From Doctorow it would take the political economy: who owns the infrastructure, who bears the losses, and how to stop the reverse centaur becoming the default employment contract. From the review’s side it would take respect for the evidence that capabilities are compounding, plus named policy levers — compute reporting thresholds, independent capability evaluations, procurement standards for human oversight — instead of exhortations to pick a side.
Falsifiable predictions or nothing
Above all, a better guide to life after AI would commit to dates and numbers. Doctorow tells readers the crash is coming but, as the review notes, not when, and a prediction without a horizon cannot be acted on or falsified. The scenario-planning literature Noetel points to does this better: it attaches probabilities and revision dates, then updates in public. Opinion writers, including us, should be held to the same bar — which is why the practical section above is framed as a hedge, not a forecast.
Verdict: A Compass, Not a Map
So is the book worth your evening? Our verdict: yes — read it for the diagnosis, not the directions. As a compass pointing at who benefits from the current arrangement, it is the sharpest thing in print. As a map of life after AI — the actual terrain, with dates, routes and exits marked — it is exactly what the review says: outdated on capabilities and silent on specifics. Noetel is right that it lacks answers, and he is wrong to imply the question was not worth asking. The gap between those two positions is where the useful thinking lives, and for now readers must do it themselves.
Life After AI FAQ
Is life after AI actually coming, and when?
Nobody credible has a date. Doctorow argues the financial structure guarantees a correction; the review argues capability progress keeps attracting real money. Both can be true in sequence — bubbles have historically popped while their underlying technology kept improving, which is why we treat life after AI as a scenario to hedge, not a forecast to schedule.
What is a reverse centaur in plain terms?
A centaur is a person using a machine as a tool. A reverse centaur is a person managed by a machine — the algorithm sets the pace, the human absorbs the risk. Doctorow’s book argues the AI business model quietly prefers the second arrangement, and that resisting it should be the organising goal for life after AI.
Should I still read The Reverse Centaur’s Guide to Life After AI?
Yes, alongside its critics. The book is 240 sharp pages on power and incentives; the Conversation review is a necessary corrective on capabilities. Reading either alone gives you half a picture — together they are the best short course available on the life after AI debate.
Will AI tools disappear if the bubble bursts?
Almost certainly not — that is the point of Doctorow’s residue argument. The dot-com crash did not delete the web; it changed who owned the servers. In most versions of life after AI, useful models keep running, open-weight alternatives improve, and prices reset to something honest. What disappears is the subsidy, which is why portability matters more than picking the winning vendor.
What should a small business do about the AI bubble?
Automate narrow processes with measured returns, keep your data and prompts portable, insist a named person owns every automated workflow, and stress-test what a doubling of AI prices would do to your margins. Those four moves pay off in every version of life after AI, which is the only kind of advice worth acting on.
References
The Reverse Centaur’s Guide to Life After AI — Verso Books
The real (economic) AI apocalypse is nigh — Pluralistic, Cory Doctorow
What Kind of Bubble is AI? — Locus Magazine, Cory Doctorow
Cory Doctorow Predicts the AI Bubble Will Burst — The Tyee
Review: The Reverse Centaur’s Guide to Life After AI — Thurrott.com
Review: The Reverse Centaur’s Guide to Life After AI — By Any Other Nerd
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
