AI Didn't Break Your Managers. It Exposed Them.

Many leaders are reading the current moment as AI showed up, and now managers are struggling. . .

That's not what happened. Managers were already the problem. AI just made it impossible to look away.

The data has been screaming this for three years. Gallup shows manager engagement has been falling year over year:

• 30% in 2023

• 27% in 2024

• 22% in 2025

  
  

Half of managers say they give weekly feedback to their direct reports; only 20% of ICs agree they get it. That's not a perception gap. That's a load-bearing wall with a crack in it. Since managers drive 70% of the variance in team engagement, when they break, the team breaks to the tune of $438 billion in lost productivity globally last year alone.

Then AI showed up.

McKinsey's research is clear: 43% of standard managerial tasks are now affected by generative AI: 19% augmented, 24% automated outright. The half that disappeared is the operational half: coordination, status synthesis, scheduling, drafting, project plans. It's the half a weak manager could hide behind.

What's left is what they were already avoiding: coaching, hard conversations, judgment, real-time feedback, growth. It's the half that requires courage and craft. Now it's most of the job.

Here's the kicker: Gallup just found that the single biggest predictor of whether an employee adopts AI is whether their direct manager actively champions it. So the same managers who can't give weekly feedback are now also the bottleneck on your AI ROI.

The two wrong reactions

Most companies are responding in one of two ways. Both are wrong.

The first reaction is to widen span of control. Gartner predicts 20% of organizations will eliminate more than half of their middle management by 2026. Fortune is calling it the "megamanager era." Bosses are going from six direct reports to twelve. The math sounds clean: AI did the busywork, so each manager can carry more people. The problem is what dies first. Deloitte's research is unambiguous: when span of control widens, coaching, mentorship, and real-time feedback are the first casualties. So you've cut headcount by removing the exact thing you needed more of.

The second reaction is to train managers on AI tools. It's necessary, insufficient, and very fashionable. 68% of organizations report efficiency gains from AI; very few translate them into strategic work. Most managers are using freed time to do more email. Teaching a manager to prompt better doesn't fix that they were never coached on coaching. It just makes their existing limitations faster.

Two camps, same problem

In my fractional Head of People work with founder-led teams Series A to Series C, I'm watching two distinct camps emerge and each is producing its own quiet failure mode.

Camp A: the aggressive adopters where leaders mandate AI use.

What I'm seeing:

• Managers drafting performance feedback in ChatGPT. The IC can always tell. The feedback reads generic, the trust erodes, and the manager has no idea anything is wrong because the document looked nice.

• Managers using AI to "prep" for hard conversations, and walking in cold tongued. They outsourced the thinking, and the conversation goes worse than if they'd written nothing at all.

• Teams 3x'ing output with AI tools while the manager's job description is still 2023. The manager becomes a status-meeting bottleneck instead of a coach.

Camp B: the cautious adopters who are worried about IP, brand, quality, governance.

What I'm seeing:

• Top ICs using AI on the side and outpacing their managers 3:1. The manager can't evaluate the output because they don't know what AI did vs. what the human did.

• Performance review fundamentally broken: did this person do this work, or did Claude? Calibration falls apart, promotions become political.

• Cross-functional drift: sales and marketing flying with AI, ops and finance not. Manager-to-manager friction grows because they're operating on different speed clocks.

Two camps. Two failure modes. The same root: the manager role itself was never redesigned.

What the new manager job actually is

After AI absorbs the operational half, four things are left. These are the new job description and most people promoted into management in the last decade weren't selected for any of them:

1. Coaching: Not supportive 1:1s, specific, behavior-changing feedback in the moment. This is skill ICs have always wanted and rarely gotten.

2. Calibration of judgment: Distinguishing AI-assisted work from human work. Knowing when to trust an output and when to push back. Setting the bar for quality when the floor for "good enough" just rose.

3. Growth conversations: Career, performance, and hard truths. These are the conversations weak managers used to defer until an annual review.

4. Trust and relationship building: This is the thing AI structurally cannot do. It's what retains people in a market where ICs can find a new job in a week.

That's the whole job. Everything else is now negotiable.

What leaders should actually do

If you're Series A to Series C, you're still in the window where you can still design management on purpose instead of inheriting the broken version. The companies that win the next decade won't be the flat orgs that delayered, or the bloated ones that doubled down on tool training. They'll be the ones with fewer, better managers selected and built for the four jobs above.

That means three uncomfortable moves:

• Re-select. Some of your current managers don't survive this transition. That's data, not failure.

• Redesign the role. Don't promote your best IC and call them a manager. That pattern was already broken in 2019.

• Measure what matters. Stop measuring throughput. Start measuring growth, decision velocity, and retention of your top performers.

I stand by that culture happens on purpose or by accident. Pick one.