AI is the third general purpose technology in a century. We are in the messy middle — and that is exactly where the opportunity lies.

Every general purpose technology looks overhyped in year three and transformative by year ten. We are in year three.

The steam engine, electricity, and computing each followed the same arc: early exuberance, a messy middle of institutional confusion, then a decades-long restructuring of how economies produce, distribute, and consume. AI is the third general purpose technology in a century, and if you judge it by today’s muddled adoption metrics, you will miss the pattern that history is screaming at you.

The messy middle

When American manufacturers first wired their factories for electricity in the 1890s, most did the obvious thing: they replaced the central steam engine with a central electric motor and kept the same layout. Productivity barely moved. It took thirty years — a full generation — before engineers like Henry Ford redesigned the factory around the new technology, distributing small motors throughout the production line. Only then did productivity explode. Computing followed a similar script. Mainframes arrived in the 1960s; it was not until the 1990s, after networked PCs, ERP systems, and wholesale process re-engineering, that the productivity surge materialised.

AI is compressing this timeline, but it cannot skip the institutional rewiring. The binding constraint today is not capability — it is permission structures, trust frameworks, and organisational redesign. As Anthropic’s head of economics, Peter McCrory, put it on my podcast: implementation follows “a staircase, not a curve.” Capability does not instantly deliver adoption.

The evidence beneath the noise

Yet the staircase is climbing faster than sceptics admit. Three data points tell the story.

First, the share of S&P 500 companies making quantified AI efficiency claims in earnings calls — not vague aspirations but specific numbers — has jumped from 1.9 per cent to 13.2 per cent in under two years. Bank of America reports its AI coding tools cut development time by 30 per cent, saving the equivalent of 2,000 full-time engineers. BNY Mellon has over a hundred AI-powered “digital employees” in production, trimming 5 per cent off the cost of every custody trade. Boring adoption is real adoption.

Second, monthly AI revenue across the ecosystem grew from $772 million in January 2024 to $13.8 billion by December 2025 — an eighteen-fold increase in two years. The industry strain ratio, measuring how much investment chases each dollar of actual revenue, dropped from 6.1x to 4.7x in five months and is heading below 3x by mid-2026. For comparison, the telecoms bubble peaked just above 4x — and that ratio was rising, not falling.

Third, the technology has crossed what I call a coherence threshold. Models can now reliably execute tasks lasting one to two hours — not party tricks, but sustained, structured work. Software that would have cost a million pounds to build with a human team has been produced for under five hundred pounds using AI agents. Claude Code alone became a three-billion-dollar business and doubled its revenue in a single month.

Not mass unemployment — mass reorganisation

The labour implications follow the GPT pattern precisely. Spotify’s top developers have not written a line of code since December 2025. They direct an AI system called Honk that ships features while they review output from their phones. The role has shifted from production to judgement. This is not leisure — UC Berkeley research shows that AI tools increase both productivity and voluntary workload. The scope of what is possible expands faster than the effort compresses.

My house view on knowledge work transformation captures it plainly: coding restructures first. It is the canary for all knowledge work. The mechanism is a knowledge-asymmetry collapse — AI has made engineering capacity nearly free, breaking the intermediation model that sustained an entire class of vendors and consultants.

History tells us GPTs do not destroy employment in aggregate. They reorganise it, painfully and unevenly. Steam created factory workers from farmhands. Electricity created assembly-line operators from craft artisans. AI is creating orchestrators from executors. The transition is real, it is dislocating, and it is not optional.

The window

Leaders who wait for certainty will wait too long. The firms restructuring now — redesigning workflows around AI the way Ford redesigned the factory around distributed motors — are building advantages that compound. Those hanging electric lights in the old workshop, extending the working day but changing nothing fundamental, will find themselves on the wrong side of the staircase.

The ten-year transformation has begun. We are in the messy middle. That is exactly where the leverage is.

Azeem explores the AI transformation with leadership teams worldwide. Begin a conversation.