60% of companies say they'd lay off employees who won't use AI. Most haven't sorted out which ones actually won't.

Written for anyone deciding what happens next to the employee who hasn't touched the AI tool the rest of the team already uses. The number driving that decision is real and well documented. The diagnosis underneath it, at most companies making the call, isn't.

Sixty percent of C-suite leaders now say they'd consider laying off an employee who refuses to use AI. That's from the second annual enterprise AI-adoption survey run by Writer, the generative-AI platform, with the research firm Workplace Intelligence — 2,400 knowledge workers across the US, UK, Ireland, the Benelux, France and Germany, fielded over the winter and published in April 2026. It reads like a verdict: adopt or be replaced. It isn't one. It's a threat aimed at a category — "won't use AI" — that quietly contains two different employees, and only one of them is the problem the threat is built to solve.

A threat built for one problem, aimed at four

I've written before about the four things that actually stop someone using a new tool at work: they don't know how; they can't see where it fits; the old way is still faster; or they could, and won't — deliberate, informed caution. Each needs a different response, and a layoff threat only ever fixes one of the four. It works on "don't know how," because that's the one genuine knowledge gap in the set. It does nothing for the other three. Someone who can't see how the tool fits their actual task doesn't get clearer under threat — you've just removed a person who was right to be confused. Someone for whom the old method is still quicker on their specific task hasn't had the arithmetic changed by leadership's impatience. And someone who could use it and won't isn't tested by a threat at all. They're punished for saying the objection out loud.

That fourth category is the one policy keeps mistaking for the first. Senior people, and increasingly anyone who has watched a tool embarrass them once in public, are careful because the cost of being wrong lands on them, not on whoever mandated the rollout. Scepticism isn't a knowledge gap you can out-train — sometimes it's the correct read of the tool, reached faster than leadership's own. The same Writer survey found 77% of executives say AI-refusers won't be considered for promotion, but only 27% of employees think refusing AI could actually cost them their job. That's not workers being naive about the stakes. That's a company where the people setting the policy and the people living under it are reading the identical threat completely differently — usually a sign it was never built as a real decision rule, just a compliance temperature check.

The regret rate is already in the data

Skipping the diagnosis has a track record, and it isn't a good one. In an Orgvue survey of 1,163 senior business leaders, fielded in early 2025, 39% said their organisation had already made people redundant because of AI — and of that group, 55% admitted the redundancy decision itself was wrong. That isn't a company second-guessing whether the AI investment paid off in general. That's a company admitting it cut a specific person's job and shouldn't have. A regret rate above one in two, on a decision nobody can quietly reverse, is what "won't use AI" being treated as one problem instead of two actually costs.

The threat is also running well ahead of the action. Challenger, Gray & Christmas — which has tracked employer layoff announcements since 1993 — reports AI as the single most-cited reason for job cuts through the first eight months of 2026: more than 116,000 of them, roughly 22% of everything announced this year. But the New York Fed's own survey work finds realised AI-driven layoffs still rare on the ground: about 4% of service-sector firms report cutting jobs over AI in the past six months, up from 1% a year earlier, and zero percent of manufacturers. Companies are talking about the axe considerably faster than they're swinging it, which fits a workforce being managed by public statement more than by a tested rule.

The company skipped its own test, too

There's a reason for caution on the employer's side, not only the employee's. A Federal Reserve Bank of Atlanta working paper, surveying nearly 750 corporate executives and published this spring, found 89% report no measurable productivity gain from AI so far — the average effect over the past three years came in near 0.3%, against roughly 1.4% executives now expect over the next three. If most companies can't yet point to a number their own AI rollout actually moved, "you're not using it right" is a strange basis for cutting someone whose real objection might be that it doesn't do the job.

ROI here was never a company-wide multiplier borrowed from a vendor's press release. It's a specific workflow, measured, before and after, on the same terms I'd apply to any training claim. A firm that hasn't run that test on itself has no real standing to run a verdict on the one employee who says the tool didn't help.

What to check before the next "adopt or else"

The fix isn't softer language wrapped around the same threat. It's separating the two employees before either one gets cut. One genuinely doesn't know how — training on the actual task, on the account they'll really use, closes that gap in weeks, and it's the only one of the four constraints a mandate was ever capable of solving. The other has usually already run their own test and reached a specific, statable objection: which step, which output, which client-facing risk it creates. That person isn't the one to threaten. Their objection is the review your rollout skipped, arriving a year early and for free.

Before the next round of "adopt or else" goes out, run the arithmetic the Orgvue respondents didn't run first: what changed, measurably, on this one person's actual work, against what baseline. If nothing did, the diagnosis was never really about AI adoption. That's the slower conversation. It's also the one that doesn't come with a better-than-even chance of being reversed a year later.

This is a position, not a finding of my own — built on named, dated third-party research linked throughout. Where competing outlets have blended several different surveys into one round "55%" figure, I've cited the single study behind the number I used (Orgvue) rather than repeat the blend.