Harvard Business School researchers sat down with 18 people at two consulting firms this year. Partners, managers, junior consultants. They asked a simple question: what actually happens when a firm rolls out AI across a team.

The partners were fine. They signed off on the tools, saw the strategic upside, and moved on to the next initiative. The junior consultants were mostly fine too, they picked up the tools fast the way theyre used to picking up any new software. It was the managers in the middle who absorbed the hit.

Julia Shin and Sandra J. Sucher, the researchers behind the study, published their findings in Harvard Business Review in June. Their core finding isnt really about whether AI works. Its about who ends up doing the invisible labor of making it work, and thats almost never the person who approved the rollout.

Who Decides vs. Who Absorbs

Senior leaders lean into AI because the incentives point that way. Expanding its scope looks good on a slide, it signals the company is moving fast, and the strategic upside is real enough to justify the bet.

Middle managers experience a different rollout entirely. They validate what the AI produces before it goes out the door. They catch the errors nobody else is positioned to catch, because theyre the ones who actually know both the client and the work well enough to spot when something is off. They coach their direct reports on tools that change every few months. None of that shows up as a new line item anywhere. Delivery pressure doesnt drop to make room for it. If anything it goes up, because leadership now expects AI to have made the team faster.

What “Validating AI Output” Actually Looks Like

This is the part that gets lost in the strategy deck. A manager cant just skim an AI-generated deliverable and forward it along. If its wrong, its the manager’s name on it, not the tool’s. So they read it line by line, the way you’d check a junior analyst’s first draft, except now theyre doing that for every deliverable the team touches, not just the risky ones.

That reading time isnt in anyone’s capacity plan. Nobody added hours to the week for it. It just gets absorbed, quietly, by the same person who was already the busiest one on the team before AI showed up.

“Most organizations treat AI adoption as a technology challenge, a software rollout managed by IT and celebrated by the C-suite. Some even see it as a fast track to headcount reduction.”

The Coaching Job Nobody Assigned

Then theres the training layer. Someone has to teach the team how to actually use these tools well, when to trust an output and when to double check it, how to prompt for something usable instead of something generic. That job defaults to the manager too, with no curriculum, no protected hours, and no acknowledgment that its a real second job stacked on top of the first one.

This is precisely the kind of operational gap a company doesnt see until its already causing turnover, which is exactly what an AI Assessment for companies is built to surface, before a manager quits or a deliverable goes out with an error nobody caught in time.

Where the Rollout Actually Breaks

Executives tend to picture AI rollouts failing at the pilot stage. A tool doesnt get adopted, a use case doesnt pan out, the budget gets pulled. Thats not where this one breaks.

It breaks quietly, months in, when the managers holding the middle start burning out from a job that was never formally created. Some absorb it and get resentful. Some start pushing errors through because they no longer have time to catch them. Some leave. None of that shows up on an adoption dashboard.

Ask your middle managers what changed for them this year. If the answer is everything, and the org chart still says nothing did, you already know where this breaks first.