Ask any executive if their company has adopted AI and they'll say yes. Ask them what exists in their organization today that wouldn't have existed without AI, and you'll mostly get silence.
Everyone measures adoption. McKinsey reports 88%. Companies count licenses, logins, the number of employees who have "access to an AI tool." The word shapes the entire conversation: who adopts, who resists, who's behind, who's ahead. But the word itself is the problem. Adoption assumes there's a defined object you either take or don't. It assumes work stays the same and a tool gets added on top. It assumes the issue is resistance, that there are laggards to convince.
That frame was fine for the computer, for the CRM, for the cloud. It doesn't work for AI.
The engine in place of steam
What most companies do with AI today looks a lot like what factories did with electricity in the early twentieth century. Economic historian Paul David documented it in a foundational paper (1990): when electricity arrived, most factories did the obvious thing. They swapped the steam engine for an electric motor and kept the exact same layout. Productivity gains were close to zero for thirty years. The real gains came when factories were redesigned around what electricity made possible.
The equivalent today is AI that helps write emails faster, sort documents, summarize meetings. Real gains, but marginal. The organization stays identical. The work stays the same, just a bit faster. And those marginal gains quickly become the norm. When everyone writes emails with AI assistance, nobody has an edge anymore. AI becomes baseline, not advantage.
The numbers back this up. In its own 2025 report, McKinsey finds that 88% of companies say they've adopted AI, but only 21% have redesigned even a single workflow, and just 6% see real impact on their bottom line. BCG gets to the same place from a different angle: 60% of companies generate no material value from AI, and the firm calls out standard adoption metrics (logins, time spent in tools) as input metrics, not outcome metrics.
The two most influential consulting firms in the world conclude, in their own data, that adoption isn't what drives value. Yet it's the word they put in the headline.
What rethinking the system actually means
Take accounting software. The adoption approach is adding an AI-powered invoice extractor that reads documents a bit faster. That's the electric motor in place of steam. The gain is real, measurable, and it changes nothing strategically. A competitor who rethinks the whole system around what AI makes possible (automatic reconciliation, predictive cash-flow analysis, intelligent supplier matching, real-time anomaly detection) won't offer accounting software "with AI." They'll ship a fundamentally different product. And they'll win. Not because they have a better model, but because they have a better system.
This is what the adoption frame misses. Companies that checked the box think they're safe. But their structure hasn't moved, and someone else is already building the thing that will replace them.
The threat doesn't only come from competitors doing the same thing better. It comes from what didn't exist before. Internal systems nobody would have built. Roles that had no name two years ago. Capabilities that weren't in the org chart. AI doesn't just make existing work faster. It makes work possible where there was none. Autor, Chin, Salomons and Seegmiller measured this over the long term: roughly 60% of jobs held in 2018 were in titles that didn't exist in 1940. Most of today's work isn't accelerated work. It's work that emerged.
Acemoglu estimates that if AI only speeds up existing tasks, the macro effect stays small: less than 0.7% total factor productivity over ten years. The real case for AI needs new tasks, new roles, new categories of work. That's exactly what the adoption frame makes invisible. Work that didn't exist has no adoption rate. It doesn't show up in any report.
The right question
If this is right, then most "AI strategies" floating around are built on the wrong foundation. Measuring adoption means measuring how far a tool has spread. But AI isn't a tool. It's a capability that changes what you can build. Companies that treat it as a tool to adopt are optimizing the wrong thing. The ones asking "what becomes possible now" are playing a completely different game. Give it five years and these two types of companies won't look anything alike.
The question worth asking isn't whether your organization has adopted AI. It's simpler than that: what exists now that wouldn't have existed without it? If the answer is nothing, if it's just that emails get written a bit faster, then nothing has really been adopted. You've plugged in a new motor where the steam engine used to be. And you're waiting.