The untrained layer.
Why AI value dies at middle management.

The sharpest AI user in your company is probably 26. They’ve rebuilt their entire workflow around it: research, drafts, analysis, the boring 40% of their job that used to eat their afternoons. Their output roughly doubled sometime last year, and nothing official noticed.

Now look one box up the org chart. Their manager has never had a single hour of training on how to evaluate AI-assisted work. Not one. They’re expected to judge output they don’t know how to produce, catch errors in tools they’ve barely used, and coach a way of working nobody ever taught them.

That reporting line exists in almost every organisation right now. And it’s the exact spot where AI value goes to die.

Only 1 in 8 managers feel confident leading AI

The Chartered Management Institute surveyed more than 1,000 UK managers this year. The numbers are blunt.

Read those together and the story writes itself. The people responsible for turning AI tools into AI results don’t feel equipped to do it, 40% say so outright, and the results follow: only 5% transformational gains, a full quarter seeing nothing.

One more number, because the problem runs all the way up: just 8% of boards have strong AI knowledge. The layer approving the AI budget often can’t evaluate it either.

CMI’s chief executive put it plainly: “Getting AI in the door is the easy part. Making it actually deliver is much harder.”

The invisible second job

Here’s the part most leadership teams haven’t clocked. When your best people adopt AI bottom-up (and they do; nobody waits for the memo), the new work doesn’t disappear. It moves up a level.

Harvard Business Review published a study in June with a title that does most of the work: “AI Adoption Is Overloading Your Middle Managers.” Across 18 interviews at 2 major consulting firms, the researchers found the same shape everywhere. Frontier employees adopt AI on their own. Their managers then absorb a new, unnamed, unpaid job: validating AI outputs, catching the errors, coaching people through the change. Delivery pressure stays exactly where it was. Formal support: none.

So the manager quietly becomes the quality-control layer for a technology they were never trained on. A few rise to it. Most do the rational thing under pressure: slow the whole thing down, or stop looking too closely.

Your most AI-fluent people report to your least AI-trained people.

That’s not a talent problem. It’s a training problem, one level up from where the training budget went.

How to manage AI-augmented teams

Most companies treated AI as an IT rollout. Buy the licences, run a launch webinar, done. But the questions that decide whether AI actually produces value are management questions.

What does good AI-assisted work look like? Which outputs need human review, and how deep? How do you set targets for someone whose output just tripled? Do you reward the analyst who automated half her role, or quietly hand her more work and call it capacity?

None of that comes in the software box. And a manager who can’t answer those questions doesn’t stay neutral. They default to what they know how to evaluate: the old way of working. Which means your fastest people learn to hide how they actually work, effort goes unmeasured, and the productivity gain never reaches anything you’d show a board.

Malaysia makes this sharper, by the way: Malaysian employees are already ahead of the global average on frontier AI use, and we’ve written about the leadership visibility gap that creates; this piece is about the layer in between.

The human layer is trainable

This isn’t an argument that middle managers are obsolete in the age of AI. It’s the opposite. The judgment calls just got harder, and judgment is precisely what managers are for. They’ve simply been left to figure it out alone, on top of their actual jobs.

And here’s the useful thing buried in the CMI numbers: nothing in them is about the technology. Confidence, training, evaluation. All of it is human, and all of it is trainable. A 40% training gap is the most fixable problem in the whole AI conversation.

So if you’re setting AI training budgets this year, consider the reallocation almost nobody makes: some of it belongs to the managers, not just the staff. Manager AI capability is measurable. You can baseline it, build it, and track whether it moved, the same way you’d treat any other business skill. That’s how we’ve built SkillTrainer AI: assessment agents that baseline where people actually stand (managers included), corporate AI training built around applying AI in real work, and analytics that show whether any of it changed. We’re HRD Corp registered and claimable under SBL-Khas, so for most Malaysian companies the budget already exists.

Your best AI user already sorted out their side of the equation, unofficially, months ago. The open question is who’s training the person they report to.