The AI Trap: Why Using Tech Isn't Transformation.

I have had a version of this conversation more than once in the past year:

"We rolled out Copilot to a hundred people six months ago. Adoption is at seventy-eight percent. Our workflows haven't changed. What did we miss?"

The owners asking are not naïve. They have technology budgets that work. Their teams are using the tools. The dashboards show green. And yet the business they were promised — the one where AI redesigns how value gets created — has not arrived.

What did they miss?

Probably just one simple fact — that buying technology and being transformed by it are not the same thing. They never were. We learned this in the digital wave. We are about to learn it again, faster, harder, and at exponentially higher cost.

The digital wave — an opportunity mostly missed.

What I think of as the Digital Factory era — roughly 2010 to 2020 — was the largest enterprise technology adoption cycle since the introduction of personal computing. Companies bought CMS platforms, marketing automation, CRM systems, the lot. Most of those implementations were technically successful. Most of the transformations were not.

Nicholas Carr saw this two decades earlier when he asked, in a now-famous Harvard Business Review piece, whether IT mattered at all. His answer was unwelcome but clarifying: the strategic value was never in having the technology — it was in what you redesigned around it.

I watched this play out at Allianz. The data was already there. The platforms worked. What was missing was the reframing of digital from a marketing cost line into a P&L position. Sixteen new revenue streams in twelve months came out of that reframing — not out of any new tool.

The technology was the excuse. The strategy was the work.

Bought is not adopted. Adopted is not transformed.

The economist Robert Solow once observed that you could see the computer age everywhere except in the productivity statistics. Erik Brynjolfsson traced this paradox for thirty years and found a consistent answer: the productivity gains from a new general-purpose technology only show up after the organizations using it have redesigned themselves around what it makes possible. There is always a lag. The lag is usually a decade.

Carlota Perez calls this the difference between installation and deployment. We are clearly in the installation phase of AI — chaotic, speculative, full of investment without coherence. The deployment phase, where the real economic value gets unlocked, comes later, and only for the organizations that did the deeper work in the meantime.

Bought, adopted, and transformed are three different states. Most companies confuse the second for the third — and stop there.

Tech-enabled ≠ tech-savvy.

Ethan Mollick, who studies this more honestly than most, keeps repeating the same point: we are still using the worst AI we will ever have, and we still haven't figured out the workflows. The tools are running ahead of the thinking.

Two companies are issued the same Copilot licence. Same training. Same rollout. The first uses it to write emails faster. Reports come together in less time. Meeting notes get cleaner. By every adoption metric, the program has succeeded. Yet the business itself remains as before — just slightly more efficient at doing what it already did.

The second company asks a different question: which decisions are we now making five times a day that we used to make once a week, and are the decision rights still allocated correctly? That question pulls on operating model. It pulls on governance. It pulls on the way work is structured.

Both companies are tech-enabled. But only the second is tech-savvy.

That is where transformation lives — or doesn't.

What it looks like when it actually happens.

DarkTools came to me nine months ago with a website problem. Their content management system was fragile, updates not working, the database collapsing under a complex product catalogue. The brief was a redesign.

Several weeks in, we were no longer talking about a website. We were talking about a Product Information Management architecture — a single source of truth feeding not only the website, but all the channels and platforms in the tech stack, including the operations. We were talking about how AI would govern data quality at the input layer, not generate copy at the output layer. We were talking about who would be allowed to instruct an agent to update a product spec, and who would have to approve it — the kind of governance question I will return to in a separate piece.

That conversation is not a website project. It is an operating model redesign with AI as a structural assumption, not a feature. The website still gets built. But it gets built as one individual skin of an entirely new business model, not as the system itself.

That is what tech-savvy looks like in practice. It is slower. It is harder. It compounds.

Stopping at the first question will probably set you up for epic failure.

Every owner facing an AI decision is implicitly answering two questions. Most only notice the first.

1. What are we buying?
2. What are we redesigning around what we have bought?

The first one is easy. There are vendors, demos, decks, comparison matrices. The procurement process has a budget line and a timeline. Everyone knows what to do.

The second question is harder. Because there is no vendor. No demo. No matrix. The answer involves the operating model, the decision rights, the way the business has organized itself for the past decade or two.

Most companies stop after the first question. But asking the second one, and finding answers, truly sets the others up for success.