Who's using AI at work, and who's still waiting

At the software company where I work in my day job, we hardly used AI last year. This year we're fully agentic: the team works with AI agents every day, using an approach called compound engineering, where each piece of work makes the next one easier because what we learn goes back into how the agents work. We're not yet as quick as we are at rm-rf software, but we're getting there.

The official figures say the same thing has happened across the economy. In late 2023, about 12% of UK businesses with ten or more staff used any AI at all. By June 2026 it was about 35%, nearly three times as many. The rise has been about ten points a year for the last two years: 15% in June 2024, 25% in June 2025, 35% in June 2026.

But the average hides a big gap. Some industries have taken to AI as a matter of course. Others have barely started.

Who's leading

Across the EU the pattern is the same, from a survey that counts slightly differently (any use of AI during 2025, firms with ten or more staff):

The leaders are the industries whose work happens on a screen. In the UK, 58% of information and communication businesses use AI, followed by education at 51% and professional services at 46%. In the EU it's 63%, and two-thirds of programming and IT consultancy firms. In the US, the Census Bureau's survey found 40% of information businesses and 34% of finance and insurance firms had used AI in the previous fortnight, against 20% of all businesses.

Who's trailing

At the other end are the industries whose work happens on site or on the road. Only 13% of UK construction businesses use AI, and 11% of EU ones. Transport and storage is at 19% in the UK and 11% across the EU, and accommodation and food at 21% and 12%. US retail is at about 14%.

The UK figures also show how deep the use goes. In information and communication, 28% of businesses use two or more kinds of AI. In transport and storage it's 0.3%: the firms that use AI at all mostly use one thing.

These are the industries we build DoRoutes for: removals, delivery and waste firms. So the gap matters to us.

Size matters as much as industry

Bigger firms are far ahead. In the EU, 55% of businesses with 250 or more staff used AI in 2025, against 17% of those with 10 to 49. In the UK it's 49% of the largest firms and 28% of businesses with fewer than ten staff. Most firms in the trailing industries are small, so they're behind on both counts.

Why some industries are slow

The surveys ask firms why. Eurostat asked businesses that had considered AI but weren't using it, and they gave these reasons:

• 71% said they lacked the expertise
• 53% weren't sure of the legal consequences
• 49% were worried about data protection and privacy
• 21% said it wasn't useful for their business

Behind those numbers, I think there are a few plain reasons.

The work isn't on a screen

Today's AI is best at text and code. Most of an office job is exactly that. A removals crew's job is lifting a wardrobe down three flights of stairs. AI can help a firm like that around the work (quotes, planning, paperwork), but it can't do the work itself.

Nobody has the time to find out

A firm of twelve has no one whose job is to try new tools. The owner spends the evenings on quotes and chasing invoices. "Lack of expertise" often means "lack of an afternoon".

The worries are reasonable

These firms hold customers' home addresses and payment details. Being unsure what an AI tool does with that data is a good reason to wait.

The data isn't written down

A lot of the knowledge in these firms lives on a whiteboard, in a WhatsApp group or in someone's head. AI can't plan tomorrow's vans from information nobody has recorded.

Where it takes off

The ONS found that most businesses adopt AI through free tools or by buying software that has it built in, not by building their own. That matches what I've seen. AI spreads fastest when it arrives inside a tool people already use for their job, and someone they trust has checked that it works.

So for the trailing industries the answer probably isn't "learn AI". It's software that does the job properly, records the data as it goes, and brings AI in where it saves time, with a person making the decisions.

How we use it

At rm-rf software, AI agents do most of the building. They work from our own board, DoWorkflows, with the same permissions a person would have, and every change they make is recorded. It's compound engineering here too: every mistake an agent makes becomes a written rule that every agent reads before it starts work. In DoWorkflows, anything AI writes that reaches a customer, like a newsletter or a support reply, is a draft until a person approves it.

This post was made the same way. Claude, our AI assistant, collected the figures from the statistics offices' own pages and datasets, and I checked them and rewrote it.

If you run a firm in one of the trailing industries, start small. Pick one job that's mostly paperwork, like quotes or invoices, and try a tool that does it for you. Keep the decisions yourself.

Sources

• Office for National Statistics, Artificial intelligence in UK businesses: 2023 to 2026 (opens in a new tab), 20 July 2026. Business Insights and Conditions Survey, June 2026; UK businesses with 10 or more employees unless stated.
• Eurostat, Use of artificial intelligence in enterprises (opens in a new tab), 2025 survey; datasets isoc_eb_ai and isoc_eb_ain2, updated 15 June 2026. EU enterprises with 10 or more persons employed.
• US Census Bureau, Large Firms With at Least 20 Employees Biggest AI Users (opens in a new tab), 26 May 2026. Business Trends and Outlook Survey, December 2025 to May 2026; AI used in the previous two weeks.

The three surveys ask different questions of different firms, so compare the patterns between them, not the exact percentages.