The Myth That Adopting AI Means Getting AI

The myth

I have three numbers pinned at the top of my notebook this week. Sixty-eight. Forty. And five hundred million.

The first two come from the AWS UK AI Adoption report that landed on Thursday. The third surfaced on Bill McDermott’s Q1 earnings call at ServiceNow on Friday afternoon. Different sources, different industries, different framings, same conclusion: the assumption that having AI is the same as getting AI just became measurable, and the gap is embarrassing.

The myth, plainly stated: if a company has adopted AI broadly, it is getting AI productivity.

Why it sounds right

This belief is everywhere because adoption is what shows up on dashboards. Seat counts. Pilot counts. Tool subscriptions. Internal posts about prompt libraries. Conference slides that boast “97% of executives have deployed AI agents in the past year.” All of those numbers feel like progress, because they are easy to count and easy to put in a board update.

And honestly, three years ago, adoption was the bottleneck. People were skeptical. IT was nervous. Procurement was slow. Just getting the tool into the hands of employees was a real fight, and the win was real. The reflex to celebrate adoption is not stupid; it is just two years out of date.

The problem is that access stopped being scarce in early 2025 and is now functionally free. The cost of a GPT-3.5-level query has dropped from $20 to $0.07 per million tokens in about eighteen months, according to the Stanford AI Index figures cited in Allwork’s Friday analysis. When access becomes that cheap, access stops being the moat.

What the evidence says

The AWS report published Thursday, conducted by Strand Partners, surveyed UK organisations on the gap between basic AI use (summarising documents, answering questions, drafting emails) and advanced AI use (redesigned workflows, agentic decision-making, new products built on top of models). Sixty-four percent of UK firms now use AI in some form, up from 52% the year before. That is the adoption number, and it looks great on a slide.

Then the report split the population by maturity. The split is the headline.

The verbatim language matters here, because everyone is going to misquote it next week:

Twenty-eight percentage points of separation between adopters and operators. Same tools. Same models. Same vendor invoices. Different operating model.

The Allwork.space analysis on Friday made the same point in a sharper sentence: “Productivity gains appear when AI is fitted to a task, a workflow, and a management system. They do not appear because a company bought a license.” That piece also flagged a customer-support economics study showing a 15% productivity gain on average, with the largest gains going to less-experienced workers who finally have a system that compensates for thin domain context. The gain is real and the gain is conditional.

Then ServiceNow reported on Friday and turned the same story sideways. Bill McDermott told analysts the company itself has captured roughly $500 million of internal productivity from its own AI deployment. Inside ServiceNow, AI specialists are now closing assigned cases 99% faster than human agents and resolving 90% of internal IT requests autonomously. On the customer side, Now Assist is tracking toward $1.5 billion of AI annual contract value this year, up from a $1 billion target. Customers spending $1M or more on Now Assist grew over 130% year-over-year. Deals that include three or more Now Assist products grew almost 70%.

That is not the shape of adoption. That is the shape of customers committing to a redesigned operating model and paying more for it. McDermott called it the real story behind AI: customers spending more on AI as they go deeper into the operating model, not headcount being cut.

The reframe

Here is the better mental model.

Adoption is access. Productivity is what happens when access meets workflow redesign, named ownership, and a management system willing to enforce both. The 68 versus 40 gap is not about the AI. The AI is identical for both groups. The gap is about the operating discipline that surrounds the AI.

Gartner’s report from earlier in the same week put a complementary number on this. Only 39% of technology leaders believe their current AI efforts will actually improve financial performance, and the leaders who do report transformative results are investing up to four times more in the boring foundations: data quality, governance, talent, and change management. Boring is doing the heavy lifting.

Competitive position over the next twelve months will be decided less by which model a company bought and more by whether anyone is responsible, by name, for the workflow that runs on it.

If a company is somewhere in the middle of the 64% who have adopted AI but stuck below the 24% who have actually redesigned around it, this is not a tooling decision anymore. It is an operating decision. The choice is whether to spend the next two quarters buying more tools or spending the next two quarters fixing the workflows the existing tools already have permission to touch.

So what

For Series B operators, this changes what to measure on Monday morning. Stop reporting seat counts and pilot counts up to the board. Start reporting how many core workflows have been redesigned around AI, who owns each one by name, and what the before-and-after metric is. If the answer is “everyone has Copilot” but no workflow has been touched, the honest read is that the company has bought access, not productivity, and that distinction is going to show up in margin within two quarters.

For CEOs, the budget conversation is the same conversation wearing a different suit. The line items that drive the 68 are not the model subscription. They are the workflow analyst, the senior owner with authority to redesign across teams, and the manager willing to retire the old process. Underfund those three roles and the AI invoice becomes an expense without a return.

The most useful thing I can tell you this week is that the productivity question has stopped being a technology question. It is a management question dressed up in a model card. The companies that have understood this are pulling away. The companies still counting seats and waiting for productivity to materialise are going to keep waiting, with very up-to-date tools.