What Six Hundred Years of History Actually Taught Me

I started writing this series because I was bored of AI doomsayers or evangelists telling me AI would either save civilisation or end it, maybe by Thursday. What struck me, after spending more time than is healthy reading about the printing press, the railways, the telephone and the early internet, is that we have been here before. The pattern is almost embarrassingly consistent.

Fourteen posts later, here's what I think history is telling us. Five things — the main conclusions I kept arriving at from different directions, across the eras we looked at.

1. Understanding humans usually beats understanding the technology

Go back to the butcher I opened this whole series with. He didn't understand refrigeration when it arrived in the 1890s — he understood he had a problem keeping meat fresh in August.

August was the month he lost stock, turned away customers, and watched his margins shrink in the heat. Refrigeration solved August. He didn't need to know how the compressor worked; he needed to know what he could do with it.

This pattern runs through everything we've covered. Piggly Wiggly's Clarence Saunders didn't understand supply chain theory. What bothered him was the sheer waste of the old system — the army of clerks fetching every item while customers queued. So in 1916 he let people walk the aisles and choose for themselves. The bigger effect was one he probably didn't foresee: handed the freedom to browse, shoppers started making their own choices, and brands suddenly had to earn attention on the shelf. Self-service turned out to be the foundation of modern consumer marketing. Lovable's founders didn't invent most of the technology stack they use. They understood that the gap between "having an idea" and "having working software" was one of the most painful bottlenecks in modern business. And they closed it.

The businesses that win technology waves — consistently, across every era — are not led by people who love technology. They're led by people who spot the problems it can solve.

This matters for AI right now because the investment in AI literacy is enormous and mostly aimed at the wrong target. The useful question is not "how does a large language model work?" The useful question is "what does my customer hate doing that currently requires a human?" That's where the August problems are.

2. The pattern is predictive — here's what comes next

One of the things I didn't expect when I started this was how mechanically the same sequence plays out. Technology arrives, a small number of people spot something real in it, and the investment community piles in. The mania peaks, the crash comes, and the infrastructure survives. Then the businesses that were quietly built on top of that infrastructure, during the mania, turn out to be the ones that matter.

We saw it with railways: the Brontë sisters lost money in 1845, the British seaside resort — Blackpool, Brighton, Scarborough — was built on the track they left behind. We saw it with the dot-com crash: telecom companies raised $1.6 trillion to lay 80 million miles of fibre optic cable, went bankrupt, and left behind infrastructure so cheap that by 2004 the cost of bandwidth had fallen 90%. YouTube, Netflix, and Spotify were built on cable paid for by people who didn't survive to see it used.

The reason this pattern holds for technology manias specifically — and not for pure financial bubbles like Tulip Mania — is that real physical infrastructure gets built. The track goes in the ground, the cable goes under the ocean, and it outlasts the companies that funded it. That's what appears to be happening with AI data centres. The investors may not all survive, but the computing power will.

So, three things I think will come to pass with AI — not predictions, but pattern-matches:

The infrastructure investment will look wildly over-optimistic in hindsight, and the infrastructure will survive anyway. The data centres, the chips, the model training — most of it will not generate the returns investors are currently modelling. Some of the companies funding it will not exist in ten years. The infrastructure will outlast them and become cheap, the way railway track became cheap and fibre cable became cheap, and every business built on top of cheap AI capability will benefit from capital that wasn't theirs to lose.

A significant AI-created mess will generate a significant industry selling the mop. Spam created email security. Ecommerce returns created recommerce. AI-generated content is already creating demand for AI-detection, AI-sourcing, and trust infrastructure. The specific shape of the biggest AI-created problems is not yet clear. The people who find and solve them first are going to be sitting on a large business.

The interface will move again. Voice, gesture, image, ambient computing — at some point the way we interact with AI will shift in a way that the current generation of AI products didn't anticipate, and the companies who built only for the current interface will find themselves where the typewriter manufacturers found themselves when word processors arrived.

3. The map is not the territory

Alfred Korzybski, the Polish-American philosopher, spent most of the 1930s making one point: the map is not the territory. The description of a thing is not the thing itself. The word "fire" doesn't burn you.

I've been guilty, across this series, of implying that history is a map precise enough to navigate the present by. Unfortunately it isn't, quite.

Here's what seems to be different this time, and I think it is worth calling out.

Every previous technology wave was domain-specific. The railway improved transport. The telephone improved communication. Electricity improved power delivery. Each one was transformative within its domain and then rippled outward. AI is the first general-purpose cognitive technology. It doesn't improve one thing — it potentially improves any task that currently requires human thought. That is not the same as previous waves. The scope is categorically different, even if the emotional and social response is identical.

The speed of diffusion is also unlike anything in the historical record. Electricity took forty years to reach half of American homes. The internet took about seven years to reach a hundred million users. ChatGPT reached a hundred million users in two months. The window between "technology arrives" and "you have to have a view on it" has compressed to almost nothing.

And the agency question is new. Previous technologies did things. AI is starting to 'decide' things. That's a different category of tool, with implications for accountability and trust that the historical analogies don't map onto cleanly. The printing press didn't recommend which books to read. The telephone didn't decide who to connect you to. The recommendation algorithm does, and the AI agent will.

None of this means the historical patterns are useless — they're not, which is the whole argument of the series. But the patterns hold at the level of human nature and they hold less reliably at the level of institutional or regulatory response. The Church took two centuries to come to terms with the printing press. We probably don't have two centuries to work it out this time.

4. The fear is always wrong in the same way

The fears that generate the most noise — the ones that fill the parliamentary hearings and the newspaper front pages — almost always turn out to be the wrong fears. Not imaginary, just wrong.

The real danger of overhead AC cables in 1880s New York was electrical fires and electrocution, which were genuine risks and were solved by engineering and regulation. The fear that drove the headlines was that electricity would "drip" invisibly from empty sockets and poison families in their beds. The fear of the internet in 1996 was children encountering strangers online. The real long-term damage was surveillance capitalism, political polarisation, and the destruction of local journalism — none of which featured in the Senate hearings.

The current AI fears generating the most noise are job displacement and robot uprisings. These may not be wrong exactly — job displacement in particular is a real and serious question — but I'd bet that the thing AI actually does to society that proves most consequential in thirty years is something nobody is currently devoting much time to. That's how it always seems to work.

The lesson isn't "don't worry." It's: the thing worth worrying about is probably not the thing in the headlines now.

5. The one thing we learn from history is that we don't learn from history

Georg Wilhelm Friedrich Hegel said it first, in approximately 1830, while the Railway Mania was still eighteen months away. He was, of course, proved right almost immediately.

Every single pattern in this series was visible to someone at the time. The railway mania of 1845 was explicitly compared to Tulip Mania of 1637 by contemporary commentators — people literally wrote pamphlets saying "this is a bubble, we have seen this before." It crashed anyway. The dot-com boom of 1999 was explicitly compared to Railway Mania by people who had clearly read their history. It crashed anyway. There are analysts today drawing direct comparisons between AI infrastructure investment and dot-com capex. The market continues uninterrupted.

This is not because people are stupid. It's because knowing the pattern and being able to act on the pattern in real time are different skills. The person who recognises a bubble is not automatically the person who profits from that knowledge — often they're the person who exited twelve months too early and missed the last, largest, most irrational leg of the rally.

What does change, slightly, is what happens at the individual business level. Some of the executives who read about Kodak and Fujifilm and asked themselves which one their organisation resembles will go on to do something different — not many, but maybe some. The data on organisational change in the face of disruption is not encouraging, but it isn't zero either.

The series ends here, which means I've now spent considerable time arguing that history contains lessons, followed by one final lesson which is that the lessons don't reliably get applied. I'm aware of the irony.

But I think there is some value in holding these ideas in tension. The butcher who understood August wasn't smarter than the other butchers. He just paid attention to the right thing, at the right moment, and moved.

That's still the trick.

Article 15 of 15 — the final post in the "Every Era Has Its Reboot" series. | redebuter.com Read more