A new workplace survey out of Germany doesn't say anything dramatic. It measures something quieter: what employees are actually doing with AI, versus what their employers think is happening.

The University of Konstanz's third wave of its ongoing workplace AI study surveyed 1,105 employees in May 2026 and found that AI use at work rose only slightly year on year, from 35% to 38% (reported by Phys.org, 18 September 2026). The more telling figure sits underneath that one: only 55% of AI users say the tool they use most was actually introduced by their employer. Nearly half found it themselves.

The gap widens further along two lines that matter for any business owner planning what to do next. Knowledge and office workers use AI at roughly double the rate of production and manual staff (49% versus 25%). Highly educated employees use AI at nearly three times the rate of those with lower formal qualifications (56% versus 21%).

It's a familiar shape to anyone who spends time inside SMEs on this subject. Two people on the same team, similar role, similar tenure. One has been using an AI tool every day for the best part of a year. The other has never opened one. It rarely comes down to interest. More often, one of them happened to be shown something, or stumbled across it, and the other simply wasn't.

None of this describes a discipline problem. It describes a visibility problem. Most SME leaders who raise AI with me assume they're starting from a blank page, that adoption hasn't really begun, or that it's contained to one enthusiastic person in the office. The Konstanz data suggests otherwise: adoption is already under way, unevenly, informally, and largely out of sight of the people meant to be managing it.

There are two ways to respond to that. One is to tighten the rules quickly: block the tools, write a policy, remind everyone what's allowed. It's an understandable reflex, and it isn't wrong exactly, but it treats the symptom rather than the cause. Rules written without first knowing what's actually happening tend to either miss the real activity entirely or push it further out of sight.

The other is to find out first. Ask people, plainly, what they're already using and why. Not to catch anyone out, but because sensible governance, useful training and fair access can't be built on top of a picture that's wrong. This is why the AI Accelerator Diagnostic starts with exactly that question, mapped across six dimensions rather than one. Tools and Data asks what's already in use. Governance asks what people need to know before using it in real work. People and Culture asks who can use AI well, and whether the business actually supports that. All three sit on the same underlying finding: you can't govern, train or plan around adoption you haven't measured.

The education-level gap deserves a specific mention, because it's easy to read as inevitable, and it isn't. If the people least likely to have used AI at work so far are also the ones least likely to be offered structured training, the gap won't close on its own. It will simply harden into a permanent difference in who benefits from a technology that's supposed to be broadly accessible. That's a capability planning problem, not a talent one, and it's fixable with a plan rather than a policy.

The practical first step is smaller than most leaders expect. It isn't a company-wide audit or new software. It's a short, honest conversation, or a two-minute anonymous survey, asking what people already use, for what, and how they learned to. Businesses that ask this calmly, before deciding what to do about it, tend to end up somewhere sensible. Businesses that skip straight to a policy tend to end up managing a problem they never properly understood.

The German survey isn't a call to panic. Adoption barely moved year on year. It's a reminder that the honest starting point for most SMEs isn't "we haven't started." It's "we don't yet know what's already started." That's a much easier problem to solve, once someone actually asks.

Go wisely.