Most AI adoption programs I see are training the wrong half of the company.

The plan looks identical from one industry to the next. Pick a prompt engineering curriculum. Get five hundred to five thousand employees through it. Stand up a champions program. Send a memo about responsible use. Then wait six months and discover that 30 percent of the licenses are unused and the productivity metric you promised the board has not moved.

The diagnosis you reach for is that the people are not adopting fast enough. The diagnosis is wrong.

The VP who certified six hundred people

I sat with the VP of Operations at a mid-sized logistics company last month. She had run a thorough internal certification. Six hundred and forty people through it. Three tiers. A leaderboard. A budget line.

Six months in, the dashboard said something was working. Usage was up. Tickets resolved was up by single digits. But she could not explain to her CEO why the program had been worth the spend. The thing that was supposed to compound was not compounding.

I asked her one question. Who, in the org, has changed how they review work because of AI?

She thought about it for a long time. Nobody.

The certified analyst was still being judged on the same KPI from 2023. The certified ops manager was still running the same status meeting. The certified planner was still copying numbers into the same deck for the same Wednesday review. The tool changed. The system around the tool did not.

The certification was real. The system the certified people had to operate in was the same system the uncertified version of them had worked in.

The pattern I see in every rollout

Once you start looking for this, you see it everywhere I work. Three things are usually true at once.

• The employee has access to a new tool and knows how to use it competently.
• The metric the employee is judged on still rewards the shape of the old output.
• The manager above the employee does not use the tool themselves, so they are evaluating the new work with last year's instincts.

In that environment, the rational thing for the employee to do is exactly what they have been doing. They might use AI to get there faster. They will not redesign their work around it. The system around them does not pay for that redesign.

The interesting move, the one almost nobody makes, is to change the system the work runs through before you train one more person.

What the research actually says

Microsoft's 2026 Work Trend Index, published last week, surveyed 20,000 knowledge workers across ten countries and analyzed more than 100,000 Copilot chats. They were trying to understand what predicts whether AI actually delivers value inside an organization.

Their answer is uncomfortable.

Organizational systems account for 67 percent of AI's reported impact. Individual mindset and behavior account for 32 percent.

Two thirds of the outcome lives outside the employee. The systems are doing the heavy lifting, or they are quietly canceling the work out.

A second number from the same report sits next to that one and explains the first. Managers who actively use AI themselves drive a 17-point lift in reported employee value from AI and a 30-point lift in trust of agentic systems. The single biggest variable inside an organization is whether the person above you is also doing the new work, or just asking you to.

Only one in four AI users in the survey say their leadership is clearly and consistently aligned on AI. That is the gap. The certification programs are filling in the wrong third of it.

Talking with Fatjon Kalemaj about this last week, I described it as the training paradox. The companies that need new training the least are the ones already running their systems on the new work, because the manager already models it and the metric already rewards it. The companies that buy the most training keep needing more of it, because nothing they have trained ever cashes through to a changed system.

If you are an executive responsible for AI inside your company, two things are worth doing this quarter, in this order.

First, audit what your managers actually do with AI in front of their teams. Not what they say. What their team sees. If your VPs are still asking for the same kind of memo, in the same format, on the same cadence, they are silently telling everyone below them that the old work is the work. No certification beats that signal.

Second, take one workflow that AI should have visibly changed by now and ask, on paper, what would change if the tool were genuinely doing what it is capable of. Then change the metric, the review cadence, and the artifact that comes out of it. Train nobody. Just rebuild the system around the work. See what moves.

The hard part is that this is a different job than the one most executives signed up for. Buying training is purchasing. Redesigning the workflow is leadership. The first is cheap and visible. The second is slow and almost invisible until it shows up in the numbers a year later.

If two thirds of AI's value lives in the system, and the system is the executive's job, then AI is not a workforce upgrade. It is an executive performance review.

The companies that figure that out this year will compound. The ones still counting how many people passed the prompt course will spend another six months training the wrong half.