Accountants have an interest in saying that artificial intelligence pays. So when PwC — a firm with more than 360,000 employees worldwide, which pitches itself as “client zero” for AI transformation and sells advice on the same shift to others — reports that the technology is splitting the workforce in two, the confession carries some weight.

The firm’s 2026 Global Workforce Hopes and Fears Survey, published on Tuesday, drew responses from almost 50,000 workers — 49,364, by PwC’s own count — across 48 countries and 29 sectors, questioned between May and June this year.

Use of the technology is rising. Sixty-four per cent of workers said they had used AI at work in the past 12 months, up ten percentage points on a year earlier, and the share using generative AI every day rose from 14% to 22%.

The spread of the benefits is another matter. From the answers, PwC sorts employees into four camps. At the top sit the “front-runners”, 14% of the whole, holding skills employers want and getting strong returns from the machines: just over half use generative AI daily and nearly 80% say they have access to learning and development.

Then comes what PwC calls the “engine room” — 56% of workers, the people the firm describes as the core of an organisation’s day-to-day delivery. Few of them use generative AI daily, and fewer than 40% have access to learning and development resources. “They’re not getting the same access to learning. They’re not getting the opportunity to innovate. They are not getting to use AI in a meaningful way,” says Peter Brown, PwC’s global workforce leader.

Beside these two groups stand the “AI insurgents”, about a fifth of staff, whose skills are less in demand but who push the tools hard all the same, and the “indispensables”, whose scarce skills employers prize.

The consequences show up where you would expect them to. Job security, confidence in asking for promotion, trust in managers and skills development all came in lower for the engine-room cohort. Mr Brown calls the result a “two-speed” workforce.

The caveats matter. The survey does not establish why workers’ experiences differ, and the four categories rest partly on what respondents say about their own use of AI and on how in demand they believe their skills to be.

Even so, the findings pose a plain problem for any employer that has bought the software and is waiting for the return: buying tools is not the same as changing work. “The danger is that you actually could render a big chunk of your workforce largely irrelevant in the world of work,” Mr Brown said.

PwC has reorganised its own house on the same bet. In February it rewrote its training agenda around 30 core skills, 15 of them about AI and 15 about human qualities, and it has cut the number of offices its entry-level American consultants may join, in the name of community and learning opportunities. Mr Brown concedes the firm has made some mistakes along the way.

His advice to bosses is to say plainly what the machines are for and what result is hoped for. “Workers aren’t expecting leaders to, I think, sugarcoat everything,” he said. “They just want to understand what’s going on.” Widening access, he insists, is not a choice between backing the front-runners and investing in everyone else: “there’s an enormous amount of value there that can be tapped.” He is presumably right. It remains the case that the group PwC says keeps organisations running day to day — more than half the workforce — is the one the machines and the training are mostly not reaching.