Why senior leaders get more from AI than junior staff, and what to do about it

The same tool, two very different experiences

Ask a senior leader how AI is going and you will often hear real enthusiasm. Ask someone two years into their career and the answer can sound quite different. New research from PA Consulting puts numbers on that gap, and it should make every leader pause before assuming their own experience is everyone's.

PA surveyed 2,000 UK employees and published the findings on 9 November 2026. PA is a consultancy that sells AI and change services, and the press release does not give fieldwork dates or full method, so treat this as consultancy research rather than an independent study.

What the research found

Seven in ten senior leaders (70%) say AI helps them achieve things they could not have done before. Among entry-level and junior employees the figure is 38%. Senior leaders use AI more than three times as often each day as junior staff, averaging 6.9 uses a day. Three quarters of senior leaders (72%) say AI produces higher-quality work, against 43% of entry-level and junior employees.

Speed is not the same as benefit. 60% of employees think AI helps them work faster, yet 37% say they then spend as much time reviewing and refining AI-generated work as they would have spent producing it themselves. Almost a third (32%) have yet to see meaningful benefits at all.

Accountability is blurring too. Two thirds (66%) believe accountability stays with the person responsible for an output, whatever AI contributed. But 21% are unsure where it lies, 13% feel less responsible for quality when AI has produced most of the work, and 26% feel uncomfortable signing off work that AI has shaped.

The benefits are also uneven across sectors. Reported benefit is highest in technology, media and telecoms (87%) and financial services (77%), and much lower in healthcare and life sciences (55%), the public sector (53%) and charities and not-for-profits (39%).

A second report the following day points the same way. Payscale's AI Workforce Impact Report, vendor research from a pay data company based on an August 2026 survey of 500 employers and 1,000 employees, found that 43% of employees save three or more hours a week with AI, but only 19% say that time goes into more strategic or higher-value work.

Why this matters: checking is not the same as learning

There is a simple explanation for much of the gap. Experienced people already know what good looks like. When AI hands them a draft, they can spot the weak paragraph in seconds, keep the useful part and move on. The tool multiplies judgement they already have.

Junior staff are in a different position. They are still building that judgement, and for most of them the way they built it was by doing the first draft, getting it wrong and having it sent back. If the machine now writes the first draft and the junior person's job becomes checking it, two things happen. The checking takes longer than expected, because they cannot yet tell a good answer from a plausibly wrong one. And the learning that used to come from doing the work quietly disappears.

That is not an argument against AI. It is an argument for designing the work so that people still learn.

A service management lens

John Sansbury's The Journey to Service Excellence makes a recommendation about new staff that predates any of this: "All new joiners to an enterprise should initially spend some time in the support centre." His reasoning is that the support centre is where people see the real environment, its idiosyncrasies and its problems.

Applying that principle to AI, the lesson is that people learn an organisation by being close to the real work. When AI takes over the routine work that used to provide that closeness, the learning path has to be rebuilt deliberately. It will not happen by accident.

What leaders can do

1. Look at the junior experience directly. Ask early-career colleagues to walk you through a recent AI-assisted task, with the emphasis on how they checked it and what they learned, not just the result.
2. Measure the checking time. If reviewing AI output takes as long as doing the work, the time saving is not real for that task. Count review and correction time alongside any claimed gains.
3. Make sign-off explicit. Decide who signs off AI-shaped work, what they are checking for and how much time they have to do it. A quarter of staff being uneasy is a signal that this has not been made clear.
4. Protect learning on purpose. Keep some tasks done by hand for people who are learning, pair juniors with experienced colleagues when reviewing AI output, and explain why the output is right or wrong rather than just correcting it.
5. Decide what freed time is for. If saved hours simply turn into more volume, people will see AI as more work for the same reward. Say openly where the time should go, including development.

Where training fits

Much of this is leadership capability rather than technical skill. Line managers need to be able to redesign work, coach people through reviewing AI output and hold honest conversations about accountability. Junior staff need structured practice in judging AI output, not just instructions on how to prompt.

Honest limits

Both reports are produced by firms with commercial interests in the topic. The PA release does not give fieldwork dates or weighting. Self-reported benefit is not the same as measured performance, and senior leaders may be more inclined to report success with a programme they sponsored. Questions about legal accountability for AI-assisted decisions in regulated sectors need specialist legal advice, not a survey.

The question to carry

In two years, who on your team will be able to tell a good AI answer from a plausibly wrong one, and how will they have learned?

Tags: ai adoption, leadership, junior staff, accountability, learning and development, future of work

Sources

• https://www.paconsulting.com/newsroom/ai-risks-widening-the-workplace-divide-new-report-finds-9-november-2026

• https://www.payscale.com/press-releases/payscales-ai-workforce-impact-report