deathspiralorgoldenage.org

Further reading

A collection of books, papers, links & references.

The sales pitch

  1. OpenAI (2024), customer story: Klarna's AI assistant does the work of 700 full-time agents. openai.com, klarna.com The pitch that sold AI to boardrooms. Klarna publicly rehired humans in 2025.

The golden age

  1. Iain M. Banks (1987–2012), The Culture series. Post-scarcity, people and AI Minds coexisting, work optional. B taken all the way, and this site's inspiration.
  2. Autor, Chin, Salomons & Seegmiller (2024), "New Frontiers: The Origins and Content of New Work, 1940–2018," Quarterly Journal of Economics 139(3): 1399–1465. academic.oup.com Most of today's jobs didn't exist in 1940. New work is the norm.
  3. Acemoglu & Restrepo (2018), "The Race between Man and Machine," American Economic Review 108(6): 1488–1542. aeaweb.org Automation takes tasks away from people, but new tasks for people get invented too. Whether there ends up being more or less work depends on the balance between the two.
  4. Romer (1990), "Endogenous Technological Change," Journal of Political Economy 98(5): S71–S102. journals.uchicago.edu An idea, once discovered, can be used by everyone at once, so an economy that keeps producing new ideas grows faster and faster.
  5. Weitzman (1998), "Recombinant Growth," Quarterly Journal of Economics 113(2): 331–360. academic.oup.com New things are reformulations of old things, so knowledge can build upon itself and accelerate. The limiting factor is likely the ability to process new ideas into usable form.
  6. Babina, Fedyk, He & Hodson (2024), "Artificial Intelligence, Firm Growth, and Product Innovation," Journal of Financial Economics 151. sciencedirect.com AI-investing firms grew by adding products, not cutting staff. Correlational, old-school (pre-generative) AI.

The death spiral

  1. Keynes (1936), The General Theory of Employment, Interest and Money. marxists.org When firms pay less in wages, people have less to spend, so firms sell less, and on it goes.
  2. Cooper & John (1988), "Coordinating Coordination Failures in Keynesian Models," Quarterly Journal of Economics 103(3): 441–463. academic.oup.com Each firm's customers are other firms' employees, so what is best for one firm depends on what the others do. The economy can get stuck with every firm cutting, even though all of them would do better if none did.
  3. Bessen (2019), "Automation and Jobs: When Technology Boosts Employment," Economic Policy 34(100): 589–626. academic.oup.com Automation adds jobs until demand is satisfied, then removes them. The best case against this website.
  4. Acemoglu & Restrepo (2019), "Automation and New Tasks," Journal of Economic Perspectives 33(2): 3–30. aeaweb.org Measures the balance between tasks lost and tasks invented? in the US: for the last forty years automation has removed tasks faster than new ones have appeared. In the model's terms, A has been winning since before generative AI.
  5. Karabarbounis & Neiman (2014), "The Global Decline of the Labor Share," Quarterly Journal of Economics 129(1): 61–103. academic.oup.com Labour's share of income has been falling worldwide since the 1980s. Also the source of the model's 0.60 spend on wages, a historical level rather than today's.
  6. Banerjee & Wang (2026), "The Reverse Big Push: Generative AI and Self-Fulfilling Automation," arXiv:2608.25602. arxiv.org Argues that firms automate because they expect other firms to automate and shrink their customers' spending, so the expectation makes itself true. The death spiral, formalised for generative AI. A new preprint.
  7. Hemenway Falk & Tsoukalas (2026), "The AI Layoff Trap," arXiv:2603.20617. arxiv.org Each firm keeps the whole saving from automating but suffers only a fraction of the lost customer spending; the rest lands on its rivals. So firms automate past the point that is good for any of them, and the paper finds that nothing short of a tax on automation stops it. The why page's "nudge" is the alternative to that tax.

Cut or create

  1. Brynjolfsson (2022), "The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence," Daedalus 151(2): 272–287. amacad.org Firms can build AI to replace people or to help them, and the incentives favour replacing. The model's two firms.
  2. March (1991), "Exploration and Exploitation in Organizational Learning," Organization Science 2(1): 71–87. pubsonline.informs.org Organisations can exploit what they already do well or explore new things. Exploiting pays sooner and more reliably, so over time they explore less, even when exploring is what keeps them alive. A is exploiting; B is exploring.
  3. Raynor & Ahmed (2013), The Three Rules: How Exceptional Companies Think. penguinrandomhouse.com Studied thousands of companies over forty-five years and found the exceptional ones followed two rules: better before cheaper, and revenue before cost. That is strategy B.
  4. Hoque, Davenport & Scade (2026), "AI Transformation Requires Redesigning Work, Not Cutting Roles," Harvard Business Review, 28 August 2026. hbr.org Commentary on surveys of HR leaders who made AI-driven layoffs: 8.4% said the cuts delivered as promised, 55% found AI needed more oversight than expected, one in three lost critical skills. Strategy A, tried and found wanting.
  5. Ma (2026), "Layoffs tied to AI hurt worker productivity – and the reason may surprise managers," The Conversation, 12 August 2026, summarising a working paper (SSRN 6652799). theconversation.com Analysis of employee reviews, financial reports and AI layoff announcements at US public companies over five years. AI only raises productivity where staff are positive about it. Layoffs blamed on AI, turn staff against it, so the cuts undercut any gains.

The model's inputs

  1. Brynjolfsson, Li & Raymond (2025), "Generative AI at Work," Quarterly Journal of Economics 140(2): 889–942. academic.oup.com The source of the default ×1.15 multiplier in the model: about 15% more productive.
  2. Schatsky (2026), "The AI Value Levers," Harvard Kennedy School Belfer Center. belfercenter.org The source of the default 90:10 company split in the model.