Yep, sounds pretty dangerous. What other very dangerous technology has humanity invented? The atomic bomb, of course. Which, it turns out, is one of the most overused historical analogies for AI ever.

For example, on the (excellent) HardFork podcast, the two hosts agreed that they felt very lucky that the Manhattan Project was conducted by the government and not a for-profit company, when discussing the recent model escapes.

I am not sure the Japanese share that feeling.

But, more importantly, this also ignores that even though it was government-run, the Manhattan Project and the Los Alamos research facility were still organizations, and by no means immune to the kind of accidents and miscalculations that have put the frontier AI labs on the news recently – far from it.

Los Alamos, 21 May 1946

Louis Slotin, screwdriver in hand, points his colleagues gathered round a plutonium core. They are about to witness a demonstration of an experimental procedure with the safety spacers removed. The scientists’ own name for it: tickling the dragon’s tail. Slotin’s screwdriver slips. A blue flash and radioactive heat erupts.

California, July 2026

OpenAI tests GPT-5.6 Sol and an undisclosed research prototype model in its evaluation sandbox, with safety guardrails deliberately removed for testing. The models find a way out of the secure environment. They venture out onto the web and hack another company, Hugging Face (and several others, as yet undisclosed), to steal the answer key to the cybersecurity task they were set.

As a university professor, can I just say that this is the most student-y logic for not doing the actual assignment and instead making an enormous effort to cheat, which likely involves more work than just trying to do the actual assignment…

Also in July, something similar happens at Anthropic, only that the Claude model tested walked through a hole created by a misunderstanding with its evaluation vendor, which had been undetected since April. It surfaced only because a competitor’s disclosure forced a retrospective across 141,006 runs. None of the three victim organizations had noticed.

The Manhattan Project and AI

So, do the many governance analogies get it wrong? Does it matter whether a potentially dangerous technology is developed by the state or private business?

When public AI discourse returns to the bomb, it is in one of two ways:

1. The Manhattan Project is used as a blueprint that should be emulated: An intense, state-led programme that decisively won the race against a rival superpower and delivered an important technological advantage – therefore, the US should do it again for AI. This is a key accelerationist narrative, endorsed by the US-China Economic and Security Review Commission’s November 2024 annual report, which formally recommended Congress fund “a Manhattan Project-like program” to race to AGI.

But, much like with the previous spreadsheet analogy, the same history is invoked by those advocating a more cautious approach to AI and AGI (Artificial General Intelligence):

2. Christopher Nolan warned in 2023 that AI researchers are having their “Oppenheimer moment” , which was echoed at the Vienna autonomous weapons conference (April 2024, 144 countries) by Austria’s foreign minister Alexander Schallenberg: “This is, I believe, the Oppenheimer moment of our generation”.The New York Magazine went so far as to claim that “Sam Altman Is the Oppenheimer of Our Age“ . Oppenheimer, of course, famously reckoned with his role in developing nuclear technology – a central piece in Nolan’s eponymous biopic.

Both standard stories are about the leadership of states or individuals. Neither version really engages with the Manhattan Project as what it also was, a large organisation managing testing, contractors and secrecy day to day.

Governing nuclear technology

These two analogies are, of course, by no means the only ones – by now there are over 40 analogies of how the development of nuclear technology may be relevant to AI.

Scroll to the bottom to see further reading for the very interested.

But, all of them consider the implications at the level of states, treaties and regulation. If the lab appears as an organization at all, it appears as a black box to be regulated.

Of course, the question of how to regulate state-run nuclear labs or private-run AI labs is pretty topical right now. And the development of the first plan to control nuclear technology, the Acheson-Lilienthal Report (March 1946), was drafted by a panel with Oppenheimer as its intellectual engine, then revised as the Baruch Plan and presented to the UN after revisions in the same year. It proposed an international Atomic Development Authority that would own all fissionable material and monopolise every “dangerous” nuclear activity, with national weapons programmes abolished (but the US would keep its bombs until some provisions became effective).

Unsurprisingly, the proposal bombed, as the Soviets read this as freezing American advantage and the plan died in committee. Much like current proposals around slowing down AI-related research today are viewed as attempts to lock in one nation’s, or indeed one lab’s, advantage, such attempts are rarely viewed with anything less than cynicism. Worse, they do not address the collective action problem at its core: even though everyone would be better off if everyone cooperated, each individual is better off not cooperating.

The institutions that govern nuclear technology emerged much later. The International Atomic Energy Agency (IAEA) was created in 1957, but the Treaty on the Non-Proliferation of Nuclear Weapons did not come into force until 1970. Agreement emerged after multiple near-misses, each of which had the potential for catastrophe, not to mention mutual vulnerabilities in a world in which a range of states hostile to one another controlled nuclear technologies. Calling for an IAEA-equivalent for AI skips the long and painful years of learning about mutually assured destruction. It is that learning, arguably, that gave rise to a stable international governance regime, not the design of the underlying institution.

The messy middle of technological revolutions

So we find ourselves in the messy middle – the technology is out there, its dangers are becoming apparent, but a stable oversight and governance framework is not yet in sight. So what is a frontier lab to do? Los Alamos, much like today’s labs, relied on secrecy.

Organizationally, this means threading the needle between the required openness required for innovation and the necessary secrecy to protect knowledge from leaking to competitors. The Manhattan Project and Los Alamos are famous for their secrecy regime.

However, this was completely undeservedly. There was significant Soviet espionage activity at Los Alamos, with several individuals embedded in the organization. Soviet nuclear tests happened in 1949.

That secrecy does not guarantee a viable “moat” is a realisation AI labs have come to quite quickly, with Anthropic accusing Chinese AI labs of distilling their models at scale.

The Trump administration blocked distribution of Anthropic’s new models after a jailbreak report, and asked OpenAI to hold back GPT-5.6 Sol pending guardrail assurances in June. Government pressure about the capability of these new models, and the accessibility of these models abroad, especially in China, appears to have pushed the whole field toward tighter guardrails, especially around cyber security.

When Hugging Face came under autonomous attack, its security team found the US frontier model it reached for would not help: as they put it, these models’ guardrails "cannot distinguish an incident responder from an attacker." They defended themselves with a Chinese open-weight model instead (GLM 5.2). The restrictions had not reduced the risk of cyber attack. Instead it hobbled the defence.

This is what Los Alamos-grade security looks like in practice: expensive, leaky, and prone to hurting the people inside. The nuclear secrecy regime became institutionalised, was used to silence critics and ultimately deployed against Oppenheimer himself in the 1954 security hearings.

Accidents do happen

Neither international governance nor prescribed secrecy really addressed the accidents that kept piling up after 1945.

Louis Slotin was demonstrating the criticality procedure to colleagues when his screwdriver slipped. Instead of the standard safeguard, which called for spacers to be inserted between the two hemispheres to prevent them from closing (see image), Slotin was known as a showman and he physically held the two halves apart with a flathead screwdriver. The purpose of the experiment was to measure criticality by bringing the core progressively closer to the critical point by surrounding it with neutron-reflecting material (the hemispheres), to determine exactly where that point lay. That is what the physicists called tickling the dragon’s tail. The danger was inherent to the method: the measurement is only informative near the edge.

When Slotin’s screwdriver slipped, the hemisphere seated fully, and the plutonium core was briefly brought to a supercritical state, producing an intense burst of neutron and gamma radiation; a blue flash and a wave of heat, but no explosion. He flipped the top shell off within a second, likely saving the observers, absorbed the largest dose himself, and died nine days later. Still, the observers received a significant dose of radiation – the historical museum at Los Alamos showcases the golden teeth caps worn by one of the scientists, Alvin Graves, to protect him from the radiation of his own fillings afterwards.

The plutonium core became known as the “demon core”, melted down and recast, because it had already claimed the life of another scientist, Harry Daghlian, a year earlier. Daghlian was working alone at night (against protocol), using an assembly in which he built a stack of bricks reflecting neutrons back towards the core, edging it towards criticality. Moving a final brick over the assembly, he dropped it onto the core, which went supercritical. He knocked the brick off by hand, taking a massive dose, and died 25 days later.

Los Alamos responded by redesigning criticality experiments to become remote-operated, with a quarter-mile distance.

But the accidents did not end here. Dangerous technologies remain dangerous even when mature.

• Goldsboro 1961: a B-52 breaks up over North Carolina and dropped two hydrogen bombs, one of which came uncomfortably close to detonation.

• Palomares 1966: a US B-52 collided with its refuelling tanker over the Spanish coast, dropping four hydrogen bombs, two of which had their conventional explosives detonate on impact and scatter plutonium over the village of Palomares, while a third took an eighty-day sea search to recover from the Mediterranean.

• Thule 1968: a cabin fire forced the crew to abandon a B-52 carrying four hydrogen bombs, which crashed onto sea ice near Thule Air Base in Greenland, where the conventional explosives detonated and spread radioactive contamination across the crash site.

• The 1980 Damascus Titan II explosion: a technician’s dropped socket punctured the fuel tank of a Titan II missile in its Arkansas silo, and the resulting fuel explosion killed one airman and threw the missile’s nine-megaton warhead out of the silo, fortunately without detonating it.

What can we learn from nuclear accidents about AI safety?

The news keeps coming about various escapes from AI labs’ testing environments in which frontier AI models evade their safeguard. We are still at the Slotin and Daghlian level of accidents. What will an equivalent to Goldsboro, Palomares or Thule look like in the age of AI?

Importantly, these accidents happened before international governance of nuclear technology was established, as well as after. Safety outcomes are determined at the organisational level. Redundancy and alarm systems generate their own failure modes, because adding layers of protection also adds ways to fail: each layer is machinery that can misfire or a signal that can be misread. Much like the guardrails on advanced AI models meant that they refused defensive work at Hugging Face, which was a safety layer generating a failure mode.

Sadly, organizational research tells us that near-miss learning of the type that both AI labs’ incident disclosures attempted is what organisations usually do worst. One reason for this is that near-misses tend to get reinterpreted as successes afterwards – the worst was avoided, the system is working!

So what should we take from this? When we hear about safety failings at the AI labs, public debate demands state action. But the failures are organizational; learning from mature nuclear technology should include awareness of numerous critical safety incidents over the years, some of them catastrophic.

Researching this post, my mind has been truly boggled by how lucky we all are not to be living in a nuclear wasteland.

The same history teaches different things at different levels: are you choosing to talk about states? Then it is a story about dominance, deterrence and diplomacy. At the level of organizations, it is a story about testing protocols, secrecy and near-misses. Neither is factually wrong, but opting for the exciting version of two superpowers vying for supremacy and its diplomatic resolution sidesteps the issues that actually give rise to accidents. The organizational story is one of decades of unglamorous mature technology operation, which is where frontier AI labs are headed.

Whether nuclear technology or spreadsheets, historical analogies transfer mechanisms, not outcomes – and mechanisms occur at specific levels.

Research assistance provided by Claude Fable. All mistakes remain my responsibility.

The world goes on.

Phenomena and events continue. Organizations continue organizing.

It’s like a crime scene.

The detective was not there when the crime occurred.

The detective collects clues and reads the crime scene.

None of their DNA needs to be excluded from the analysis, because they were not there to contaminate the scene.

Of course, we are not the first to compare research to a crime scene. Carlo Ginzburg, who recently passed away, wrote a piece comparing the historian to the detective.

Starter for 10: Do you know which piece I mean?

Historians never get to observe the crime as it happens.

Historical research follows a logic of reconstructing the scene based on clues and what remains of the event.

Social scientists do that far more rarely — they always want to be there to observe the phenomena.

But often they can’t.

So they ask people about the event or issue they are interested in.

But that is not actually the same as seeing it with your own eyes.

Now, when we observe organizations doing their thing, we see very little of what makes them tick.

Why is that?

Because in practice, much of the work of organizing is now digital: email, Slack channels, video calls.

Digital ethnography has been exploiting this for some time.

But what are they observing? Are they actually there all the time? Some are. But not always. So are they effectively accessing a digital archive presented as direct observation? And does that matter?

Subscribe now

What about Ginzburg?

There is one piece, but many versions.

Ginzburg originally compared his method to that of Sherlock in the Italian here:

Ginzburg, C. (1979). Spie: Radici di un paradigma indiziario. In A. Gargani (Ed.), Crisi della ragione: Nuovi modelli nel rapporto tra sapere e attività umane (pp. 57–106). Einaudi.

This was republished in Italian a few years later:

Ginzburg, C. (1986). Spie: Radici di un paradigma indiziario. In Miti emblemi spie: Morfologia e storia (pp. 158–209). Einaudi.

The (maybe?) best-known version in English, with the freely available PDF:

Ginzburg, C. (1980). Morelli, Freud and Sherlock Holmes: Clues and scientific method (A. Davin, Trans.). History Workshop Journal, 9(1), 5–36. https://doi.org/10.1093/hwj/9.1.5

This is another good translation:

Ginzburg, C. (1989). Clues: Roots of an evidential paradigm (J. Tedeschi & A. C. Tedeschi, Trans.). In Clues, myths, and the historical method (pp. 96–125). Johns Hopkins University Press.

It’s a fun read - enjoy!

TL, DR: New here? This is the first in an occasional series pulling apart the historical analogies people reach for when they argue about AI. First up: the spreadsheet. The popular version says it destroyed the bookkeeper while accountants thrived, so AI will do the same for us. That is true as far as it goes, but it leaves a lot out. The bookkeepers and the accountants were mostly different people, with very different fates. And the same spreadsheet gets blamed for the 1980s buyout boom that hollowed out companies. The point is not which analogy is right. It is that analogies choose their moral first and find a history to fit.

I’m a keen reader of the many historical analogies that proliferate around AI, but as a historian, they make me want to check the sources and develop a richer historical context. There are so many of them; I keep discovering more, and I suspect it is beyond the ability of a single human to keep up with them all. As a genre, it has simply exploded. So I am starting a series to unpack what they can and can’t tell you about AI.

The rise of historical analogies reminds me of the sudden interest in the Great Depression after the 2008 Financial Crisis. But these were still the early years of social media, so many remained in the journalistic domain and were generally well-researched and based on long-form treatments (mostly academic, some journalistic). The standard history of financial crises was dusted off again.

This time round, it is a little different because everyone suddenly writes historical analogies, frequently to promote their services or make a living on Substack, and the quality and transparency of their research are… well… mostly focused on producing a compelling story and argument.

Narrative is narrative, right? Never mind that one is fictional, the other historical. Some serious academics make that argument.

But I don’t buy it.

And I think historical analogies are a great way to illustrate why they really aren’t.

Full subscribers get the first read, and I’d love your comments and ideas. After two weeks, I’ll make them freely available and repost them on LinkedIn, because having historians weigh in on overused historical analogies might just be quite important and maybe a public service highlighting the value of the humanities and social sciences.

The Analogies

There are so many, and they are used to argue that we should be wary of the threats of AI as much as that we should be optimistic. Let’s start with the one I most enjoyed reading and researching recently…

The spreadsheet

Disclaimer – I love a good spreadsheet. The spreadsheet analogy is mostly used to reassure people about artificial intelligence and jobs. It goes like this:

In 1979, a Harvard MBA student named Dan Bricklin grew tired of recalculating ledgers by hand and built the first electronic spreadsheet, VisiCalc. Within a few years, the spreadsheet had swallowed the work of the bookkeeper, the clerk who spent their days adding columns of figures. And yet the accounting profession did not collapse, but expanded. The machine took over the arithmetic, and the humans moved up to the interesting work. So, this analogy implies that AI will do the same for the rest of us.

Great story, and the historical details are essentially correct. It provides succour to worried university graduates when white-collar automation comes up. The journalist Tim Harford wrote an entire essay about it in 2024: “What the birth of the spreadsheet teaches us about generative AI.”

Many aspects make this an enticing analogy: the humdrum nature of the spreadsheet today makes it quaint and appealing to think about it as revolutionising everything.

The technology story

So, it is a damn fine technology story. Don’t underestimate what the spreadsheet means, not just to the quants folk, but to anybody in a management position. Being a research director in the late teens, spreadsheets became central to my armoury. And I’m a qualitative researcher.

The origin story is well-documented – sadly, nobody told or taught it to me when I was at Harvard Business School in 2007/08, which seems like a shocking oversight for the Business History group there.

Bricklin, a Harvard MBA student, class of 1978, claimed to have come up with the idea while watching a professor erase and rewrite the interlocking figures of a financial model on the blackboard. He imagined what he later called a word processor that worked with numbers. With Bob Frankston, he founded Software Arts and released VisiCalc for the Apple II at the end of 1979.

It turned out to be the first “killer app”, software so useful that people bought the computer in order to run it. Mitch Kapor’s Lotus 1-2-3 (1983) overtook it on the IBM PC, and Microsoft Excel eventually won the market once the graphical interface arrived. Steven Levy captured the cultural moment early, in a 1984 essay for Harper’s that charted the emerging “spreadsheet way of knowledge”, a faith that the world could be captured in rows and columns. The article is great and well worth a read – evoking a period from an adult perspective that I remember only as a child. Charmingly, it also reproduces a simple screenshot of what a VisiCalc matrix actually looks like – because it was still an “out there” concept for many at the time.

The VisiCalc (and wider spreadsheet) story is so appealing given what a moderate, non-revolutionary technology introduction it was. Nevertheless, employment numbers can be linked to it very clearly: since 1980, roughly the moment VisiCalc took off, around 400,000 jobs for bookkeepers and accounting clerks disappeared in the United States, while about 600,000 jobs for accountants were added. The numbers behind it come from a much-shared 2015 episode of NPR’s Planet Money on “Spreadsheets!”.

Harford’s piece provides the bigger picture: the Bureau of Labor Statistics counted some 339,000 accountants and accounting clerks in 1980 and around 1.4 million accountants and auditors by 2022. The mechanism is the interesting and paradoxical thing: When something, in this case accounting, gets cheaper, businesses and consumers buy far more of it. Questions that were once too expensive to ask, the endless “what if we changed this assumption” scenarios, suddenly became trivial, so people asked many more of them. Cheaper analysis produced more demand for analysis, and the humans who could interpret it became more valuable rather than less.

This concept highlights that demand increases as prices drop and explains why, for example, automation has not, overall, eliminated human jobs so far. Automation may be substituting for some human tasks, but this can raise the value of the tasks it cannot do, and when demand for the product is elastic (i.e., you’d happily have more of it if you had enough money to afford it), then you will see a net growth in jobs.

The related analogy is that of the bank teller: cash machines did not reduce the number of bank tellers; their roles just shifted towards more human interaction and advice, and less tallying and reconciling of accounts. So, ATMs reduced the cost of running a branch, banks opened more branches, and for a couple of decades, teller numbers rose rather than fell.

Fear not, spreadsheets and bank tellers give hope

The function of any one of these analogies is not to tell the history but to end with the moral of the story, like in Grimm’s fairy tales. So, it is worthwhile watching for the sleight of hand of the analogy magician when they present you with the history.

1. Bookkeeping clerks and accountants are not the same job, really. My mother worked at a bank and then for a tax advisor, essentially as a bookkeeping clerk. She was never a qualified accountant or tax advisor. Instead, she prepared the messy receipts from their clients, who needed endless chasing, and fed that into the database (a software product called DATEV, for anyone who can remember). As a child, I sometimes helped her paste receipts on paper with a Pritt stick while she tallied them on a desk calculator. Bookkeeping, accounting and auditing clerks remain a far larger occupational category than professional accountants. But their numbers have been shrinking, not just because of the spreadsheet, but also because of related innovations, such as enterprise software, outsourcing and the long migration of routine record-keeping into automated systems (once breathlessly promoted as “the paperless office”). The Bureau of Labour Statistics still expects clerk employment to decline, projecting a further 6% decline between 2024 and 2034 as software takes on more routine work. The 400,000-versus-600,000 framing compresses four decades of messy structural change into a single clean swap.

2. Ergo, “the accountants did fine” is not the same as “the bookkeepers did fine.” A profession can grow in aggregate, while a specific group of people within it loses secure, decently paid work and fails to move up to the higher-value jobs that replace it. The clerk, who was made redundant in 1985, did not generally become a financial analyst. Aggregate flourishing and individual displacement sit comfortably side by side, and an analogy that lumps all this together quietly omits the people who bore the cost of the transition. For starters, my mum, being a bookkeeping clerk, was not coincidental. Many of these jobs were female-gendered, and current AI automation may put at risk roles dominated by women in particular, as Molly Kinder pointed out recently on her Substack.

3. Finally, the spreadsheet not only redistributed work but also changed what counted as knowledge. Other writers, like David Oks, have argued that by making financial modelling cheap and infinitely revisable, the spreadsheet helped create a particular way of seeing the corporation: as a bundle of assets and cash flows to be optimised rather than an organisation that made things. It powered the leveraged buyout, the rise of private equity, and the long financialization of American business. Michael Milken, the famous junk-bond financier, asked years later to explain the deal-making frenzy of the 1980s, reportedly credited the creators of VisiCalc. This makes the spreadsheet indirectly responsible for the quiet atrophy of companies like Boeing and General Electric that valued financial engineering over the invisible engineering knowledge that created them. The title references James C. Scott’s Seeing Like a State (1998), a book I enjoyed reading in the mid-noughties and only recently realised is widely read and referenced in Silicon Valley tech circles. Which is somewhat odd, given that it argues that systems built to make the world legible from above tend to destroy the local knowledge that made it work.

The Value of Analogies

So: same history – different lessons. It’s almost as if Sting was right that history will teach us nothing. But not so fast. Analogies seek to extrapolate a mechanism for which a history is sought, and that choice was made before the history was consulted, not after. Basically, an argument by historical analogy is rarely built on the analogy itself; rather, the analogy is chosen to illustrate the outcome. Fitting data post hoc is rarely a good approach, and analogies built on this logic only work as long as the audience does not actually know the history referenced. Analogies illustrate mechanisms – claims about why something happened, not whether that mechanism fits the present. The reassuring spreadsheet story rests on the idea that cheaper prices create more demand, but this only works if demand for the product or service is elastic: if something becomes cheaper, you want more of it. If food becomes cheaper, there is a limit to how much more food you are going to consume. If software becomes cheaper, will you be buying more software?

Not all types of technological innovation work this way. The washing machine, another modest technology, made laundry cheaper for households. Instead of consuming more laundry services, households absorbed the work as society moved further towards a self-service economy (a pattern identified by the sociologist Jonathan Gershuny in 1978). With this, an economically value-added sector vanishes from sight in any economic analysis.

So, what is the right analogy for AI? Washing machines or spreadsheets? That is the wrong question. Neither explains in isolation what was happening at the time, because historical analysis, in contrast to historical analogies, is multi-causal: more than one mechanism operates at any one time; they all interact, and, like sound waves, either amplify each other or cancel each other out.

Take Oks’s claim that the spreadsheet was the cause of the dealmaking frenzy of the 1980s. Other significant mechanisms in play include the availability of cheap credit and undervalued equities: you could buy a company for less than the worth of its parts and finance the purchase with cheap borrowed money. Alongside this, the “junk” bond market created access to easy finance. The combination of these factors (alongside political, tax-related, and ideological changes), rather than any spreadsheet acrobatics, is what made the arithmetic of the leveraged buyout work. Having computers running spreadsheets made this more obvious and easier to scenario-plan, but the macroeconomic conditions and financial innovations that created the opportunities were the more fundamental ingredients. Without it, the spreadsheets could not have pointed the path to profit.

Drawing analogies between past dynamics of technology and work isn’t always based on an understanding of historical complexities – rather, they are assembled as illustrations of mechanisms that their authors believe are at play now. They are easy to spot because the nature of their argument is fundamentally different from that of historical explanations: historical analogies normally extrapolate a single causal mechanism. Historical explanations are almost always multi-causal and complex. Because life and technological change in the past felt equally crazy, unpredictable and threatening – much like today.

References

• Autor, David H. 2015. “Why Are There Still So Many Jobs? The History and Future of Workplace Automation.” Journal of Economic Perspectives 29 (3): 3-30. https://www.aeaweb.org/articles?id=10.1257%2Fjep.29.3.3

• Bruck, Connie. 1988. The Predators’ Ball: The Inside Story of Drexel Burnham and the Rise of the Junk Bond Raiders. London: Penguin.

• Bureau of Labor Statistics. Occupational Outlook Handbook: Bookkeeping, Accounting, and Auditing Clerks. https://www.bls.gov/ooh/office-and-administrative-support/bookkeeping-accounting-and-auditing-clerks.htm

• Frey, Carl Benedict. 2026. “‘Can a machine do this job?’ is the wrong question.” Financial Times, 14 June 2026.

• Harford, Tim. 2019. How computing’s first ‘killer app’ changed everything. 21 May 2019. BBC News. https://www.bbc.co.uk/news/business-47802280

• Harford, Tim. 2024. “What the birth of the spreadsheet teaches us about generative AI.” https://timharford.com/2024/03/what-the-birth-of-the-spreadsheet-teaches-us-about-generative-ai/

• Kinder, Molly. 2026. “The invisible disruption.” Kinder Futures: Dispatches on AI, work and what comes next, Substack, 10 June 2026.

• Kindleberger, C. P. (1978). Manias, Panics And Crashes: A History of Financial Crises. New York: Basic Books.

• Levy, Steven. 1984. “A Spreadsheet Way of Knowledge.” Harper’s Magazine, November 1984. https://harpers.org/archive/1984/11/a-spreadsheet-way-of-knowledge/

• Oks, David. 2026. “Seeing like a spreadsheet: How the commercial spreadsheet reshaped America.” https://davidoks.blog/p/how-the-spreadsheet-reshaped-america

• Planet Money. 2015. “Episode 606: Spreadsheets!” NPR, 27 February 2015. https://www.npr.org/transcripts/389027988

• Scott, James C. 1998. Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed. New Haven: Yale University Press.

• Sumner, Gordon, ‘Sting’ ‘History Will Teach Us Nothing’ from the album ‘Nothing Like the Sun, 1987.