More than four in five UK workers now use AI at work. Fewer than one in twenty use it in any meaningful way.

That second number is the one nobody quotes.

We’ve decided the AI story at work is one of adoption. Look how many people are using it. But scratch the surface and most of that use is searching and summarising, the shallow end. Only about one in five feel confident using AI at work. Only about one in ten have had any proper training. Most UK workplaces have no plan, no policy, no guidance.

So what’s happening? People are teaching themselves, quietly. The Microsoft figures put it bluntly: most UK employees have used AI tools their employer never approved, half of them every week. Not because they’re reckless, because no one gave them an alternative.

This isn’t an AI success story. It’s a vacuum. We bought the tools, skipped the training, wrote no rules, and called the result “adoption”.

And here’s where it stops being a tech story and becomes a wellbeing one. We’ve handed people one more thing to be quietly responsible for, with no support and no permission, then we act surprised when it goes wrong, and blame them when it does.

Unsupported responsibility is how you manufacture stress. It’s the same pattern behind so much burnout. The load falls on the individual. The structure does nothing.

The skills gap everyone’s panicking about isn’t really in your people. It’s in everything around them: the strategy, the rules, and the basic decision to say “here’s how we do this, and here’s the help”.

So before the next headline about AI-ready workforces, a quieter question. Have you given anyone the training, the tools and the permission to use this well? Or have you left them to it, and called that progress?

Sources: DSIT, AI Skills for Life and Work; EY 2025 Work Reimagined Survey (UK); ONS Business Insights and Conditions Survey, January 2026; Microsoft UK research, October 2025.