There is a conversation happening in the enterprise market where the two sides speak different languages, and neither one is wrong. On one side, technology companies bringing AI to market with the urgency of a quarterly sales cycle. On the other, executives who listen, approve a pilot, and walk out still unsure why any of it matters to the business they run. I spent twenty years on the side that presents. I know the mechanics from the inside: nobody is pushing anything. Each party is doing exactly what its own incentive structure demands — which is why the gap between them never closes on its own.
The clearest symptom is the kind of proof that circulates. The verbs never change: accelerate, reduce, cut. All of them act on processes that existed before AI and that no one thought to question. The promise is honest and the ROI is real. But every verb of that family carries a silent assumption: that the process deserved to exist in the first place. Does it? Neither side of the table has the mandate to ask. The side deciding wants to know whether the tool pays for itself. On the other side, the question is narrower still: whether the case lands well enough to authorize the next step. Both are honest questions, and neither is the one that matters.
There is an uncomfortable reason for this. The market speaks about AI in tactical terms because it has not learned to speak any other way. There are no mature sector-level theses — barring rare exceptions — on how AI reorganizes value chains, redistributes margin, shifts barriers to entry, redraws the cost structure of a specific industry. Without that thesis, there is no ballast. And without ballast, an AI investment decision stops being capital committed against a business hypothesis and becomes something else: a technology budget line that had to be spent before the fiscal year closed; a response to pressure from the board and the market for visible signs of innovation; and, underneath it all, the need to feel that one is innovating. AI gets purchased the way one buys insurance against one’s own irrelevance.
Improving processes and delivering visible gains is how any serious organization builds credibility for the step that follows. I am not questioning whether these motivations are legitimate — at least in the rooms I have sat in, they tend to be sincere. It is just that none of them amounts to a thesis. None of them answers where the company will compete now that AI is no longer anyone’s advantage. It is already the floor, only a floor nobody laid. While the committee debates the pilot, the organization has already decided on its own.
Verizon’s 2026 Data Breach Investigations Report, the nineteenth edition of a study built on more than 31,000 security incidents, records that, in a single year, the share of employees turning to unapproved AI tripled — from 15% to 45% — and that this is widening corporate data leakage.
It is called Shadow AI. The label matters little. What it exposes is that governance showed up late: people were already using the tool long before anyone decided it would be used.
The result is the vacuum. The technology arrives before the thesis. The tool arrives before the problem. And AI, which ought to be a means in service of a strategy that exists without it, is treated as an end: as a destination, as a project that justifies itself. This is where my frustration sets in. Not with the technology — I like it, I have worked with it my whole career. What bothers me is how accustomed we have become to treating the largest structural shift since digitalization as though it were a tool upgrade.
This is the vacuum I have chosen to work in. Not because I hold the thesis — I do not, and I am wary of anyone who claims to. But because the question nobody asks in the room is the only one that interests me, and because someone has to ask it before the next round of pilots is approved.
No organization transformed because it bought software. Organizations transform when they change how they decide, who decides, and what they decide about. AI enters that conversation afterwards — and only if someone has done the hard work of answering first: to what end?
Reference
Verizon. (2026). 2026 Data Breach Investigations Report (DBIR). https://www.verizon.com/about/news/breach-industry-wide-dbir-finds