Who Is Actually Absorbing Your AI Strategy?

Ask a director or a senior manager what changed in their job this year, and most won't mention a single new tool. They'll mention their evenings.

Picture the manager who used to leave work at six. Now she stays until eight. AI didn't give her less to do. It gave her a new job nobody put in writing. Check the AI-drafted report before it reaches the client. Catch the error the intern didn't know to look for. Explain to her own team why the tool that was supposed to save time keeps landing back on her desk. Her title didn't change. Her job did.

Harvard Business Review researchers Julia Shin and Sandra J. Sucher spent 2026 inside two major consulting firms. They ran 18 interviews across partners, managers, and junior staff to see how AI adoption spreads inside a company. Junior consultants reported real productivity gains. Senior partners used AI to sell and deliver more. The people in the middle got neither. They got the validating, the error-catching, and the coaching. Delivery targets never moved. No formal support came with the new load. Shin and Sucher's plain description: that middle layer is failing under the weight.

This is not the AI story most boards are tracking. Boards track adoption rates and pilot counts. Nobody has a slide for what happens to the manager standing between a model's output and a client's inbox.

Here's the number that should worry any executive who thinks their AI rollout is going fine. BetterUp Labs and the Stanford Social Media Lab surveyed 1,004 full-time U.S. desk workers in September 2025. They measured something the researchers call workslop: AI-generated work that looks finished but isn't. It gets passed along to someone else to fix. Managers report receiving it at a 54% rate. Individual contributors report 38.5%. Fixing one instance took an average of 1 hour and 51 minutes, about 20 minutes longer than if the task had never touched AI. Scaled across a 10,000-person company, BetterUp puts the lost productivity above $9 million a year. The tool didn't remove the work. It moved the work sideways and up. It landed hardest on the layer with the least room to say no.

Some companies read that pressure correctly, and they're removing the layer instead of supporting it. Amazon cut 14,000 corporate jobs in October 2025. Beth Galetti, the company's SVP of People Experience, said the goal was to reduce bureaucracy and remove organizational layers. The freed budget went to generative AI. Gartner's read on the broader trend: through 2026, one in five organizations will use AI to flatten their structure, eliminating more than half of their current middle-management positions. Korn Ferry's Workforce 2025 research found 44% of U.S. employees already report cutbacks to management levels at their company. 40% say the result is a felt lack of direction at work. 72% of U.S. senior executives say they're stretched beyond their own capabilities. The layer isn't disappearing without a trace. Employees can feel the hole where it used to be.

There's a precedent for what happens when a company treats a layer of the org chart as overhead instead of infrastructure. When ATMs arrived, banks expected teller jobs to vanish. Teller counts per branch did fall. But branches multiplied. The job that survived shifted from counting cash to relationship work: the conversation a machine couldn't have. Economist James Bessen's research on this is one of the clearer proofs that automation reshapes a role before it eliminates one. Today's middle manager is not the teller who gets automated away. The middle manager is already doing the relationship work, unpaid and unnamed, while everyone waits to see if the role gets redesigned on purpose, or just left to absorb the difference.

Every AI rollout reaches a layer that has to decide, case by case, whether the output is good enough to ship. That decision is judgment, and judgment is the terrain I call Above the Algorithm in the Relearning Operating System. It's the work agents can't do. It needs accountability, taste, and the willingness to put your name on a call. Companies that treat this as a staffing cost to trim are cutting the one function that makes agentic AI safe to deploy at scale. Companies that treat it as a skill to build are the ones whose AI programs will still be standing in three years.

Pilots impress boards. Middle managers keep the pilots from becoming liabilities.

I spent 20 years inside a Fortune 100 watching what happens when a company changes the work without changing the job description. The tools get credit for the gains. The people absorbing the friction get longer days. Some get a layoff notice that blames the very tool they were propping up. If your AI strategy funds the model but not the manager supervising it, you already know which line is undersized.

If you're the one in the middle, reading this on a laptop at eight at night, you can name the new job out loud in your next one-on-one. The tool didn't ask permission to change your role. You're allowed to ask for the title, the training, or the headcount to match it.

So take one question into your next leadership meeting, the one I keep bringing into mine. When we measure whether AI adoption is working, are we asking the people in the middle, or just the people presenting the results?

I'm easy to find if you want to compare notes.

Sources: Harvard Business Review, "AI Adoption Is Overloading Your Middle Managers," June 2026 (Julia Shin and Sandra J. Sucher, 18 interviews, two consulting firms) · BetterUp Labs and Stanford Social Media Lab, "Workslop" study, September 2025 (1,004 full-time U.S. desk workers) · CNBC, "Amazon layoffs: corporate workers as it invests more in AI," October 28, 2025 · Fortune, October 29, 2025 · Gartner, cited via Inc., 2026 (prediction: one in five organizations will use AI to eliminate more than half of current middle-management positions through 2026) · Korn Ferry, "Workforce 2025 Research," April 2025.