Companies are threatening layoffs over AI adoption. Almost none defined what 'adopted' means.

Writer's 2026 enterprise AI survey found that 60% of companies plan to lay off employees who won't adopt AI. Fifty-four percent of C-suite executives admit the adoption push is tearing their company apart. Read past the headline and the real story isn't resistance. It's that almost none of these companies ever defined what "adopted" means.

I've stood on both ends of that gap. Running program management on GenAI data operations, rolling a new QA rubric out to annotator teams only worked once we defined what "following it" looked like in specific, checkable terms, not once we announced it in a memo. Building a 0-to-1 analytics product for SMBs, the customers who actually stuck with it were the ones whose day-to-day workflow we redesigned around the product, not the ones we onboarded and hoped for the best with.

The Writer numbers back this up at scale. 79% of organizations say they're facing real challenges adopting AI, a double-digit jump from last year, not an improvement. 59% of companies are spending over a million dollars a year on the technology. Fewer than a third report seeing significant return on it. That gap isn't a model problem. Frontier models in 2026 are good enough for nearly every workflow someone is trying to bolt one onto. It's a rollout problem, and rollout has always been a program management job, not a procurement job.

Publicis Sapient's 2026 Global Enterprise AI Report shows the same gap from a different angle. AI now touches 53% of enterprise work by their count, but only 18% of organizations call themselves fully integrated, and just 10% say AI is core to how the business actually operates. Most companies didn't fail to deploy AI. They deployed it everywhere and integrated it nowhere. A tool sitting next to the old workflow, unredesigned, is optional by definition, and optional things don't get adopted, they get ignored.

Adoption needs the same discipline evals brought to AI quality: define done before you roll out, not after the usage numbers disappoint the board. That means naming, per team, what "adopted" actually looks like. Not "logged in once this month." A specific, observable behavior: this workflow now runs through the tool by default, this manual step got deleted, this metric moved by this much this quarter.

The sequence that works is boring on purpose. Map the current workflow before you insert an AI step into it, the same way you'd map a user journey before a redesign. Pick one metric the rollout has to move, cycle time, defect rate, hours saved, and record the baseline before anyone touches the new tool. Name an adoption owner per team who's accountable for that number, not just for scheduling the training session. Redesign the workflow around the tool instead of parking the tool next to the old one.

EY's 2026 survey on autonomous AI found the same asymmetry from the governance side: agent adoption at tech companies is surging faster than the oversight built to manage it. That's the mandate-without-redesign problem again, wearing a different hat. Push the technology out ahead of the operating model built to hold it, and you get shadow usage in one direction and abandoned pilots in the other. Neither one shows up as ROI.

A layoff threat is not an adoption plan. It's an admission that the real plan, the workflow redesign, the baseline metric, the named owner, was never written down, so leadership reached for the only lever left. Employees don't resist AI because they're stubborn. They resist a tool that got dropped into their workflow with no redesign and no definition of what using it well was even supposed to look like.

Acceptance criteria tell you when a feature is done. An adoption plan tells you when a rollout is real. Skip the second one and you haven't launched AI across your company. You've launched a mandate, and mandates without a workflow behind them don't get adopted. They get worked around.