You can't buy AI fluency

Walk into most companies and you'll find a familiar AI shopping list: more tools, a platform, a course library, maybe a Head of AI hire. All of it is an attempt to buy AI fluency, and AI fluency is the one thing on the list that isn't for sale. The evidence is blunt. A widely cited MIT study found that despite some thirty to forty billion dollars of enterprise spend, 95% of generative-AI efforts returned no measurable value, and named the cause a learning gap, money bought, capability didn't follow. Microsoft markets AI as intelligence you can buy on tap, abundant and scalable on demand, and that's true of the intelligence. It's not true of your people's fluency with it. You can buy the model; you cannot buy whether your team is any good at using it.

That distinction is the whole point, and it explains why so much AI spend disappoints. The companies pulling ahead didn't buy something the others couldn't. They built something.

It helps to separate two words people use interchangeably. Literacy is knowing what AI is and using it when given direction. AI fluency is confidently applying AI to novel, real problems, asking better questions, judging the output critically, and reshaping the work around it. You can hand someone literacy in an afternoon. Fluency is rarer because it has to be built: a Google study with Ipsos found only about 5% of workers qualify as AI fluent, redesigning real parts of their job with AI and using it across many distinct tasks, and those people save roughly 8 hours a week against 3 for casual users. Fluency shows up as changed work, not completed courses, which is also why McKinsey found only about 1% of companies have reached real AI maturity and named leadership, not employees, as the bottleneck. The capability has to live in the people doing the work, which is the heart of any real AI transformation.

The most common purchase attempt is to hire the capability: bring in a Head of AI or a few specialists and call it covered. That buys a function, not org-wide fluency. Microsoft's own data shows the gap it leaves, with 67% of leaders familiar with AI agents against just 40% of employees, a chasm a single hire doesn't close. A Head of AI is a catalyst for building fluency across the company, not a substitute for it; concentrate the expertise in one office and the people who actually run the workflows stay exactly where they were. Fluency isn't a thing one person can hold on behalf of everyone else, because the skill is judgment about where and how AI fits a specific job, and that judgment has to live in each role that does the work.

Two objections deserve a straight answer. The first: the tools are getting so good that fluency won't matter, you'll just buy the platform. MIT's 95%-no-return finding is the rebuttal, because generic tools stall in enterprise use precisely when nobody adapts the workflow to them, and adapting the work to the tool is fluency by another name. Better tools raise the ceiling; they don't install the skill. The second objection is sharper and we should concede it: training alone doesn't work either. People finish a course, return to the same meetings, approvals, and handoffs, and the training produces local efficiency, not advantage. That's real, and it's our point, not a counter to it. Buying a course library or a one-time workshop is still buying. Fluency forms only in people doing real work with AI, coached and reinforced, until the work itself gets redesigned, which is exactly the workflow redesign McKinsey identifies as the strongest correlate of returns. Redesign is the output of fluency, and fluency is the output of coached practice on real work, which is what adoption really means.

If you can't buy it, how do you build it? Three things, none of which is a purchase. Real work, not sandboxes, because the skill is judgment about your own deals, reports, and decisions, which a generic course can't transfer. Coaching, not just content, because a coach catches the specific way this person or team is using the tool wrong and redirects it, where a course only delivers information. And reinforcement until the workflow changes, because behavior shifts through repetition and feedback, not a single event. The trap to name out loud is the substitution of a purchase for the practice: every tool, platform, and hire is necessary infrastructure, and none of them is the fluency. McKinsey's finding that employees are ready and leadership is the barrier is good news, because it means the capability is buildable right now. Build it team by team and role by role, the way our AI for teams and enterprise programs do, and you end up with the rarest asset in the market: a workforce that is genuinely fluent, which no competitor can simply buy to catch up.