Three technology waves have hit the enterprise. The fourth one rhymes.

My first enterprise technology wave was client-server, which dates me precisely enough that you can stop doing the math. I was the junior person in the room when a division of a large insurer decided the mainframe was history and the future was applications with buttons. I have since sat through the cloud wave and the mobile wave in progressively better chairs, and I am watching the AI wave from the best chair yet, which is the one beside the client rather than under the org chart.

Four waves is enough for the pattern to stop looking like coincidence. So this is a longer piece than usual, with a story per wave and a ledger of what each one teaches about the current one. Pour something appropriate.

wave one: the technology arrives before the organization does

The client-server project at that insurer had wonderful software for its day. It failed anyway, twice, before it succeeded in year four. Not because the technology fell short; because the organization around it had not moved. The mainframe team owned change control, the new applications team owned enthusiasm, and nobody owned the seam between them. (The seam is always where transformations die. Write that down.)

Economists spent that whole decade puzzled about the same thing at national scale. Computers were everywhere and the productivity statistics could not find them, a riddle famous enough to earn a name, the productivity paradox. The resolution, visible only later, was that the gains arrived after organizations restructured around the technology, which took a decade, not a budget cycle.

I think about that insurer whenever an executive shows me an AI roadmap with transformation scheduled for next fiscal year. The models are ready. Organizations move at the speed of ownership, incentives and habit, which is the speed they have always moved at. The teams seeing real AI gains today are mostly the ones that restructured the work first and added the technology second.

wave two: the finance model was the real migration

Cloud looked like a technology story. It was a finance story wearing a technology costume. The genuinely hard part of every early cloud program I saw was never the workload; it was capital expenditure versus operating expenditure, budget owners losing servers they could point at, and procurement processes built for three-year purchases colliding with services billed by the hour.

And there was a second story inside it. Engineers with corporate cards had been quietly using AWS for months before any cloud strategy existed. Governance did not lead that wave; it chased it. The companies that did well were the ones that treated the shadow usage as evidence of demand rather than a disciplinary problem, paved the path, and put guardrails where the traffic already was.

If that sounds familiar, it should. It is the shadow AI problem with different logos. Your staff are already using AI tools, sanctioned or not, and the cloud wave's lesson is unambiguous: demand always outruns policy, and the winning move is to make the safe path the convenient one before the unsafe path calcifies into habit.

The finance lesson transfers too, almost embarrassingly directly. Per-token pricing is hourly billing all over again, complete with the bill shock, the reserved-capacity discounts, and the eventual emergence of a FinOps-shaped discipline. We priced this pattern in the honest ROI math piece; the short version is that variable-cost technology needs variable-cost thinking, and most enterprise budgeting still does not have it.

wave three: the pilot was a press release

Mobile is the wave everyone forgets, because it ended in such total victory that the struggle looks inevitable in hindsight. It was not. Around 2011, every large company built apps. Hundreds of them. There were executive app showcases. (I attended one with a lanyard and everything.) The overwhelming majority of those apps were used by approximately nobody, because they were built to demonstrate that the company was mobile, not to remove a step from anyone's actual day.

The mobile projects that survived shared one boring trait: they started from a workflow that was genuinely broken without mobility, field service, deliveries, approvals stuck in someone's desktop inbox, and worked backward to the device. The showcase apps started from the device and went looking for a justification.

Swap "app" for "copilot" and the sentence survives contact with 2026 perfectly. A distressing share of enterprise AI initiatives exist to demonstrate that the company is doing AI. They pilot well, because pilots built as performances always do, and then usage decays the moment the spotlight moves, because no workflow actually needed them. Meanwhile the initiatives that compound quietly are the ones that started from a specific broken workflow: the claims summary that took forty minutes, the support reply that required three systems, the contract clause hunt.

Start from the workflow. The wave rewards it every single time.

the ledger, and what the fourth wave does differently

Let me put the three lessons in one place, since this is long and you may be skimming by now (I would be).

First, gains arrive after the organization restructures, on a timescale of years, and the technology's readiness date is not the transformation's start date. Second, demand outruns policy, so pave the path people are already walking and put the guardrails there. Third, initiatives that start from a broken workflow compound; initiatives that start from the technology decay.

Now, honesty requires me to say where the rhyme breaks down, because no analogy survives unqualified. This wave is faster on the demand side than anything I have seen. Cloud adoption was gated by engineers; AI adoption is walking in through every employee with a browser tab. That compresses the shadow-usage timeline from years to weeks, which means the governance chase from wave two happens at a tempo wave two never faced. And the capability curve under this wave is still rising steeply, where client-server and mobile plateaued early enough to plan against. Planning against a moving capability is a genuinely new problem, and I will not pretend the 1990s prepared me for it.

But the human machinery the wave crashes into is unchanged. Ownership seams still kill programs. Finance models still lag the billing model. Showcase projects still pilot beautifully and die quietly. The fourth wave is faster and stranger than its three predecessors, and it is still, stubbornly, an enterprise adoption story, which means it will be won by the unglamorous things: clear owners, honest baselines, paved paths, and workflows chosen because they were broken rather than because they were demo-friendly.

The insurer got there, by the way. Year four. The person who finally made it work was not the visionary who launched it; it was the operations manager who inherited it, cut the scope in half, and gave the seam between the teams a name and a salary. Every wave since, I have watched some version of that person quietly save some version of that program.

Find that person early. Waves are survived by owners, not by surfboards.