The usual story about a team learning a new tool goes one way: skepticism first, then trial, then trust and, in theory, anxiety fades as skill builds. If you wait long enough, confidence and fear can’t both keep climbing at once. Right?

A year ago, Beautiful.ai’s second annual survey of American managers reached exactly that conclusion. After a year of declining replacement sentiment and wage-threat concern among the 3,000 managers surveyed, the report concluded that “professionals should feel less threatened by AI tools eliminating their role,” and that AI is “less of a threat to professionals the more widely adopted it becomes.” The third edition, fielded twelve months later from the same annual series and Pollfish panel (though not identically screened), found the opposite. The share of managers who rate AI’s output as matching or beating an experienced manager jumped to 58%, up from 36%.

In that same wave, the share who believe their employees fear being made less valuable by AI climbed to 72% (up from 64%), and the share who believe employees fear eventual job loss reached 70% (up from 58%). The report’s own section on that finding is titled “the psychological gap widens.”

Confidence and fear didn’t trade places the way the earlier edition said they would. They moved up together (confidently anxious?), and the survey that called this pattern “reassuring” twelve months ago now describes it as a “widening gap.”

Not Patience, But Perception

That “usual story” assumes fear runs on a timeline and that enough exposure eventually resolves the question. What actually moved between these two reports wasn’t time on the job, it was what managers believe the tool can do. The 2025 optimism rested on a year of managers rating AI’s output as roughly novice-level work: useful but not good enough to reopen the question of who does the job. The jump to 58% rating it at experienced-manager level or better in 2026 isn’t more familiarity showing up in the numbers. It’s a real change in perceived capability, unaccounted for by the “give it time” variable.

I don’t have an independent benchmark saying AI tooling actually crossed an experienced-manager threshold this year, and neither survey supplies one. What I have is a related pattern from a different measurement. Section’s 2026 AI Proficiency Report tested 5,026 U.S. knowledge workers’ own AI skill directly, through hands-on tasks rather than self-rating, and found managers’ measured proficiency barely exceeds the individual contributors they manage. That’s a different question than whether the tool’s output is any good, but it’s the same shape of gap: what people report about AI and what a direct proficiency test finds aren’t reliably aligned.

I don’t think Beautiful.ai’s 58% is exempt from that gap just because the report doesn’t test for it. None of this proves the figure is wrong, but it’s a reason to hold it as what managers believe, not as an independently verified capability jump.

None of that invalidates the underlying trend. METR, a research nonprofit with no product to sell either direction, has tracked the length of software task an AI agent can complete on its own since 2019, and its January 2026 update shows that length doubling on a fairly steady clock the whole time.

Something real is accelerating underneath all of this. But let me be clear again: I don’t have a study connecting that specific curve to the claim that AI output now reads as experienced-manager-level work. METR measures how long a task an agent can complete without help. Beautiful.ai’s survey measures what a manager believes about output quality. Those are different questions, and nobody has shown me they move together (yet).

I’ve led people through tool changes before. People got better and the tool stopped being scary, then the team moved to the next problem. What’s different here: believing you’ve gotten better at the tool doesn’t resolve the harder question underneath: if the tool is this good now, what exactly is still mine to do?

The Manager’s Seat

It’s tempting to assume managers are watching this from a safe distance, coaching their teams through something they’ve already resolved for themselves. A separate Salesforce survey of managers, fielded March 2026, says that’s more complicated.

Two-thirds of the 500+ managers surveyed said they’re optimistic about AI’s role in the future of work and most log real time savings. But 78% also said they feel personally responsible for their team’s successful AI adoption, and 51% said they’re anxious about keeping up with the technology themselves. Nearly half feel pressure from leadership to visibly demonstrate adoption, while less than a third work anywhere with formal tracking of what that adoption is even supposed to look like — the same governance gap I’ve written about before.

That confidence doesn’t obviously extend to what it means for the person wielding the AI at the desk. Leadership IQ’s 2025 AI Readiness study, fielded June 2025 among 1,251 executives, directors, and managers, found 79.5% personally use AI tools, and 46% either don’t believe AI will affect their own job or aren’t sure. Asked about industry-wide job replacement over the next three years, 56% gave the same answer or shrugged. Rating the tool near expert level and doubting it touches your own role aren’t two findings pulling against each other so much as the same split showing up a second time, aimed inward instead of at the team. I wish those 46% good luck — they’ll need it.

So the manager isn’t a neutral party translating leadership’s AI strategy down to a nervous team. They’re carrying their own version of the split: confident the tool works, unsure whether that confidence says anything about their own job, and still catching up to the tool themselves. Unfortunately, that makes the accountability they’ve been handed feel assigned rather than earned.

What To Coach

I don’t think the answer is managing anxiety separately from confidence as if they’re unrelated. They’re the same dial, and I don’t trust that more time cranks the anxiety down.

What’s worked better for me is naming the split out loud. Not a scripted reassurance that jobs are safe as I don’t actually know that with certainty for every role, and I trust that reassurance less now that I’ve watched a survey walk back its own version of it in a single year. It’s an honest acknowledgment that getting good at this keeps raising harder questions instead of fewer, and I’d rather have those conversations early and often than let them go unspoken until someone acts on the fear alone.

When the question actually gets asked, in a 1:1 or a team meeting, it usually sounds close to what I named earlier: if the tool is this good now, what exactly is still mine to do? I don’t reach for reassurance there. I reach for the same argument I’ve made about cognitive capacity generally: a skill or task moving to the tool doesn’t leave a void behind it; it frees up room. The harder, more honest coaching question isn’t whether that shift is happening. It’s what the newly freed capacity gets pointed at — and that’s worth naming directly instead of letting the silence answer it as nothing.

We’ll see if next year’s edition reverses course again. I know that betting a team’s morale on “give it time” is the exact bet last year’s report made, right before its own data stopped supporting it.