The 100-Year Pattern — Part I: The Wave

November 2021. Google launches Performance Max, a campaign type where an AI algorithm handles all optimization. Every manual setting removed. Our Google partner pushes it hard. We build it for clients.

Twelve months later, Berlin startups run million-euro ad budgets through the same system with one person. The friction we had been paid to remove had been removed. Not by us. By the infrastructure we were building on top of.

A few months after that, Meta announces Advantage+. Same architecture, different platform.

Our work hadn't been outsourced. It had been absorbed. I didn't have a framework for it yet. But I read the signs. A few months later I started pulling the data together. What I found had happened before. Every time.

In the last 600 years, five technologies arrived that didn't improve one industry. They transformed the economic logic of everything they touched. Economists call them General Purpose Technologies. GPTs, not the AI kind.

The printing press. Steam. Electricity. The internet. AI.

Each one followed the same six stages. Without exception.

Nothing about this is new. The signal was visible each time — to anyone looking. The question has never been whether you can see it coming. It is whether you act before the window closes.

Stage 1: Invention. The technical breakthrough arrives. Not visible to most people yet.

Gutenberg printed the first Bible in 1440. Watt patented his improved engine in 1769. Edison switched on the Pearl Street grid in 1882. Berners-Lee published the World Wide Web protocol in 1991. And in 2017, a team at Google published a research paper called "Attention Is All You Need" — the Transformer architecture that made modern AI possible. None of these felt like a revolution at the time. Most people didn't notice them at all.

One pattern inside the pattern: every wave has two dates. The technical invention, and the public moment when everyone feels it. For steam: Watt's engine (1769), then the first commercial railway (1829), sixty years apart. For electricity: Faraday's induction (1831), then Edison's grid (1882). For AI: the Transformer paper (2017), then ChatGPT, November 2022. One million users in five days. One hundred million in two months. The fastest adoption of any consumer technology in recorded history.

Stage 2: Friction Phase. The technology exists. Early adopters use it. Most people don't feel it yet. The old system still dominates and looks fine.

This is when incumbents feel safest. Sales are holding. The new thing seems like a toy for enthusiasts. The people who sense the threat early are told they are overreacting.

For AI, Stage 2 ran from 2017 to 2022. GPT-3 existed from 2020, access by invitation only. Most of the world had never heard the term "large language model." OpenAI published research papers. The rest of the world scrolled past.

Stage 3: Paradox Phase. Everywhere in conversation. Nowhere in the statistics.

Nobel laureate Robert Solow, 1987: "You can see the computer age everywhere but in the productivity statistics." He was right, and this was not a paradox unique to computers. It is the signature of every GPT wave. The technology spreads before anyone has rebuilt their infrastructure around it. Awareness rises. Productivity data doesn't move. This can last years, sometimes decades.

Enterprise adoption of large language models: under 5% in 2023. Projected above 80% by 2026. We are inside Stage 3 right now.

Stage 4: Reorganization. This is the phase most people don't see coming, and the most important one to understand.

You cannot just add the new technology onto existing infrastructure. You have to rebuild.

The clearest example: electric motors existed in the 1880s and worked perfectly. But factories had been built around a central steam engine, one power source driving everything through belts and shafts. You couldn't swap the engine. You had to tear it down and rebuild it around distributed motors, a completely different physical architecture.

Edison's first grid: 1882. Factory productivity explosion: 1920s. The gap was 40 years. Not because the technology wasn't ready. Because the organizational structure wasn't.

Reorganization is the most turbulent phase. The old way of working is visibly breaking down. The new way hasn't proven itself yet. People, institutions, and governments are all confused about what comes next. The Luddites smashed machines during Reorganization. The recording industry sued its own customers during Internet Reorganization. The pattern holds: the more painful the rebuild, the louder the resistance.

The models are ready. The workflows are not. Every business running AI experiments on top of processes built for human labor is inside Stage 4.

Stage 5: Compression. The old infrastructure collapses. Fast.

The pattern is accelerating. Each wave compresses faster than the last.

Farm laborers: 11 million in the US in 1910. Under 700,000 today. Not because farming disappeared — because the tractor, the combine harvester, and then GPS-guided automation removed the need for human hands, one task at a time.

These are not industries that failed to see it coming. Travel agents knew Expedia existed. Blockbuster executives reviewed streaming proposals. Newspapers wrote about the internet throughout the 1990s. Awareness was not the problem. Position was.

Stage 6: New Formation. New roles emerge after every compression. Always. But they require different skills than the ones that disappeared. After print: publishers. After steam: railway engineers. After electricity: grid operators. After internet: UX designers, data analysts, social media managers. The transitions were not painless. The retraining took years. The unemployment in between was real. It worked — eventually.

For AI, Stage 6 is already forming: AI workflow architects, domain experts who supervise model outputs, training data specialists. But this wave has one unusual feature: unlike Steam or Electricity, which displaced manual labor first and left knowledge workers untouched, AI is hitting cognitive work first. Writing. Analysis. Legal review. Code. Stage 6 is not consolation. The new roles form — they always do — but not in the same companies, not in the same regions, and not for the same people. The handweavers did not become factory managers. The travel agents did not become Expedia engineers. Stage 6 is how the economy rebuilds. It is not how the individuals inside Stage 5 recover.

When I watched Google absorb our campaign management work, I was watching Stage 5 from the inside. Not theoretical. Not historical. Live.

The question is not whether the compression is coming. Across 600 years and five waves, the pattern has never failed to complete. Every profession in that table believed something about their situation was different. None of it was.

Handweavers had 30 years of warning. Video store managers had about ten. We are year 9 of the AI wave, and the productivity data still shows almost nothing. That is exactly what Stage 3 turning into Stage 4 looks like. The gap between now and Stage 5 is not a generation. It is a planning horizon.

I read the signs first in November 2021. Not because I was smarter than anyone else. Because I was looking. The pattern is not hidden. The signal is there — the same shape it has always been. The question is not whether you can see it. It is what you do next.

In Part II, I look at what the people funding and building this transition have said — on record, publicly, with their names attached. Some said it in earnings calls. Some said it at conferences. All of them kept building while they said it.  The transcripts are public. Most people haven't read them.

Historical data sourced from BLS, academic research (UC Davis, NBER, Oxford), and published industry reports. Productivity statistics from peer-reviewed economic literature on GPT adoption cycles.