The Lights Came On in 1882. The Factory Changed in 1920.
In 1882, Thomas Edison flicked a switch and lower Manhattan had electric light. Everyone assumed that was the hard part. The technology worked. The capability was real. And for the next four decades, it made almost no difference to how businesses operated or what showed up in productivity data.
Not because people were slow, or resistant. But because you can't just bolt a new technology onto old ways of working and expect transformation. The factories built around steam power - their layouts, their workflows, their management structures - weren't designed for electricity. To get the real gains, you had to redesign the factory from scratch. Retrain workers. Change management structures. Rethink supply chains.
The productivity gains didn't show up in the data until the 1920s. Nearly forty years after the lights came on.
Economist Paul David documented this in his influential 1990 paper The Dynamo and the Computer, published in the American Economic Review. His argument: new general-purpose technologies don't deliver economic transformation on arrival. They deliver it after organisations have fundamentally restructured themselves around the new capability - and that restructuring takes decades. The gap between technological arrival and measurable economic impact he called the productivity paradox. It turned out to be one of the most useful frameworks for understanding how transformative technologies actually behave in the real world.
We are, right now, living inside one.
AI is the same story. But compressed.
The timeline won't be forty years. Software deploys faster than physical infrastructure. Global communication means learnings spread instantly. Capital is flooding in at an unprecedented rate - Goldman Sachs projects AI companies alone could invest more than $500 billion in 2026, and McKinsey estimates generative AI could add between $2.6 trillion and $4.4 trillion in annual economic value once widely adopted.
And yet Goldman Sachs chief economist Jan Hatzius observed earlier this year that despite all that investment, AI contributed "basically zero" to US economic growth in 2025. The Paul David paradox, playing out in real time.
The friction is real, and it's worth being honest about why:
Organisational change is slow regardless of the technology
Trust and verification takes time, especially in regulated sectors
The "last kilometre" - getting AI actually embedded in how work gets done — is genuinely hard
Skills gaps are real, not just a training problem
A rough timeline, grounded in current adoption research and historical technology S-curves:
Early experimentation: now–2027 - tools proliferating, individuals and teams finding value, patchy and uncoordinated
Mainstream business adoption (the messy middle): 2025–2035 - strategy and governance catching up with capability, competitive differentiation beginning to emerge
Deep structural transformation of knowledge work: 2030–2040 - roles, business models, and industries genuinely reshaped
Full economic impact visible in productivity data: probably 2035+ - the point where the Paul David paradox finally resolves
Roughly fifteen to twenty-five years for the bulk of it. Not eighty. Not three to five either.
Two ways to get this wrong.
The first is to flail. To treat every new model release as a five-alarm emergency, chase tools without strategy, and exhaust your team with initiatives that don't connect to anything. This mostly produces noise, some wasted budget, and a workforce that's learned to wait out the next wave of enthusiasm.
The second is to wail. To declare it overhyped, point to the hallucinations and the wrong answers and the underwhelming pilots, and conclude that AI isn't as transformative as advertised - that it's a fad, that it'll pass, that the sensible move is to hold off until someone else figures it out. This feels like prudence. It isn't.
Both failure modes come from the same misreading of the timeline: either assuming it's happening faster than it is, or assuming there's more time than there is.
The actual opportunity.
The firms that will be well-positioned in 2035 probably aren't the ones with the most AI tools right now. They're the ones that used this window - the messy, uncertain, pre-peak-productivity window - to build the things that take time to build.
Clear data foundations. Leadership that can navigate AI trade-offs with confidence rather than anxiety. Governance that enables fast, safe decisions. A workforce that's genuinely capable, not just vaguely aware. Processes clean enough to automate. A strategy that connects AI to actual business outcomes rather than just possibility.
None of that happens overnight. None of it comes from a subscription. It's organisational work, and organisational work takes years.
The firms that will succeed in 2035 are making foundational decisions right now - not because the technology is mature, but because organisational capability, data foundations, and cultural readiness take years to build. The window to get positioned ahead of this is open precisely because the full transformation is still a decade or more away.
The electricity analogy holds here too. The factories that survived and succeeded through the industrial transformation of the early twentieth century weren't the ones with the fanciest generators. They were the ones that had done the harder work of redesigning themselves around the new capability.
So what does that mean practically?
It means you don't need to panic. But you do need to move - deliberately, with a clear sense of where you're going and why.
The question isn't "should we adopt AI?" It's "are we building the foundations that will make AI work for us, rather than against us, over the next decade?"
The lights came on in 1882. The factories that survived and succeeded through the industrial transformation of the early twentieth century didn't get there by being first to flick the switch. They got there by being willing to redesign the factory.
Transformation isn't a switch. It's a redesign.
That's the work. And the best time to start it is now.
Sources
Paul A. David, The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox, American Economic Review, Vol. 80, No. 2, May 1990 — jstor.org/stable/2006600
Goldman Sachs Research, Why AI Companies May Invest More Than $500 Billion in 2026, December 2025 — goldmansachs.com
Goldman Sachs / Jan Hatzius, reported via ML Quantified, January 2026 — mlq.ai
McKinsey Global Institute, The Economic Potential of Generative AI: The Next Productivity Frontier, June 2023 — mckinsey.com
