AI in your business? Read this before you go on holiday

By Peter Springett · 28 July 2026

It's 1996. Your business has rolled out email to 35 percent of employees. It hasn't gone down well with the rest of the workforce. The result? Another 20 percent bring their own email to work, fed up waiting behind the fax machine while other departments got the proper kit.

And now these rebels are using shadow email for client correspondence: attachments, business information, all through an account you don't control and can't see.

Daft, right? Yet this is exactly what's happening with AI thirty years later according to a recent ONS survey. Ask employers and they'll tell you 35% of their people use AI in the organisation. Ask the employees themselves and one in five are using AI the business doesn't know about.

The good news: there's latent demand for AI in the UK workforce. The bad news: companies are slow to close the gap. Figures like these should be giving IT and compliance teams sleepless nights. If you haven't put AI training, governance and guardrails in place, it's time to do so. Now.

From prompting to process

The same ONS data shows 64% of businesses using AI to "improve business operations and efficiency". A fancy way of saying "write emails faster." Right now, the biggest returns in AI come from bolting a chatbot onto 30-year-old tools: word processors and spreadsheets.

But we haven’t plateaued. Beyond shadow AI, there's a growing number of advanced users who use AI for process, not just prompting. The ONS findings chime with global surveys. Microsoft has a name for these advanced users: Frontier Professionals: People who "use agents for multi-step workflows and building multi-agent systems." But they're still a minority: just 16% of the users Microsoft surveyed.

Will Cowork have its moment?

My own experience matches this. When I'm training individuals or organisations, I still see a gap between those using AI as a chatbot and those taking advantage of more advanced agentic platforms. Most people still haven't heard of Claude Cowork, or ChatGPT's equivalent. But they get it almost as soon as they log on. "Completely changed how I work" is a common response.

The biggest discovery is that you don't need to be a coder, or even understand computer logic, to use these tools to reorganise your daily work. The beauty of Cowork is that it does most of the structural analysis, design and build for you. Having a natural-language conversation with a model, then watching it build an app, a workflow or a website, is a genuine revelation.

I'm also talking to very large businesses deploying Cowork across operational teams, including finance and legal. That's quite a leap for a technology that only launched at the start of the year.

Microsoft and Google? All bets are off

But that's the shape of the AI revolution: long flat stretches where nothing looks different, followed by a jump that resets what "normal" means. Companies planning for gradual change risk getting caught out. Not because they weren't paying attention, but because they were watching for the wrong kind of movement.

This is the boiling frog problem. Nothing about any single week feels urgent. The chatbot is still just a chatbot. Then Cowork happens, or whatever the equivalent is once Google or Microsoft stop shipping assistants and start shipping coworkers, and anyone treating this as a slow-burn IT project finds the water's already at a steady boil.

The uncomfortable truth I put to every group I train: the 35% who've officially adopted AI and the fraction who've actually restructured how work gets done are not on the same timeline. And the gap is closing faster than either group expects. The businesses getting real value aren't the ones with the most chatbot licences. They're the ones who looked at what agentic tools can actually do, decided which teams would benefit, and booked in training that give them a head start before the rest of the world catches up.

You don't need to panic. But you do need to stop treating this as background noise. The frog doesn't get a second warning.