Your most experienced people are your best AI users

If you are waiting for the youngest, most technical people in your firm to lead on AI, the evidence suggests you have it the wrong way round. The people who get the most out of these tools are the ones with the deepest experience in the work, not the ones most comfortable with software. That points to a decision most leaders have not made on purpose: who in the firm should be using AI first. The answer is your senior fee-earners, and the reason is the thing they already have and cannot delegate, which is judgement about what good looks like.

What the evidence actually shows

The instinct to hand AI to the digital natives is understandable. It looks like a technology problem, and younger staff tend to pick up new software faster. But when researchers watched professionals actually use these tools on real work, the pattern that decided who succeeded was not age or technical background.

Ethan Mollick, who studies how people work with AI at Wharton, set this out in The twilight of the chatbots. Citing a study of people using an AI coding tool across different professions, he notes that software engineers had a similar success rate to everyone else. What separated the strong users from the weak ones was domain expertise. The more someone knew about the field they were working in, the better their results, and the more useful the output they got from each instruction. Knowing the subject mattered more than knowing the tool.

That should reframe the question for a leader. The bottleneck is not whether your people can operate the software. It is whether the person at the keyboard knows the work well enough to ask for the right thing and to spot when the answer is wrong.

Why expertise is the multiplier

AI is fluent, fast and confident, and it is confidently wrong often enough to matter. It will produce a research summary that reads perfectly and misstates a figure. It will draft an argument that hangs together and rests on a case that does not say what it claims. It will write a valuation memo in the right register with a number that came from nowhere. The output looks finished. Whether it is correct is a separate question, and answering it takes someone who already knows the answer, or knows enough to sense that something is off.

That is what a domain expert brings. A senior accountant reads an AI-drafted analysis and pauses at the line that does not sit right, because thirty years of doing the work has taught them what a normal set of numbers looks like. A junior taking the same draft has no such alarm. They are more likely to pass it on intact, partly because they cannot see the error and partly because they have less standing to challenge a tool that sounds so sure. Hand AI to the least experienced people and you get plausible work with no one able to check it. Hand it to the most experienced and you get a genuine multiplier, because the tool does the grunt work and the expert does the part only they can do.

The same expertise tells a leader where the tool is worth using at all. Capability is uneven: AI is strong on some tasks in your field and weak on others, and the boundary is not obvious from the outside. Someone who knows the work can see where the model helps and where it quietly makes things worse. Someone who does not will trust it everywhere or nowhere. This is the durable skill underneath the tools, the judgement that outlasts any particular product, and it is concentrated in your senior people.

The shift that makes this urgent

For a while the main way to use AI was a conversation: ask, check, ask again, steering the tool sentence by sentence. That is changing. As Mollick describes it, work is moving from chatting with a tool towards assigning a task to an agent that runs on its own and reports back. The useful mental model, in his words, is to think of yourself as a manager.

Managing is exactly what experienced people are already good at. Setting a clear brief, knowing what a good result looks like, reviewing the work and sending it back when it is not right: these are the habits of a senior professional, now pointed at an AI rather than a junior colleague. And this is not a distant prospect. In the Thomson Reuters 2026 AI in Professional Services Report, a survey of more than 1,500 professionals across 27 countries, 15 per cent of organisations had already adopted an agentic AI tool and a further 53 per cent were planning or considering it. As more of the work shifts to delegated tasks, the person doing the delegating needs to be the one who can judge the result. That is your expert, not your newest hire.

What this means for how you roll it out

None of this is an argument for keeping AI away from junior staff. They need to learn it, and they will use it heavily. It is an argument about sequence, about who goes first and who sets the standard.

When the most experienced people in a firm use AI seriously, three things follow. They work out where it genuinely helps in your specific work, so the rest of the firm is not guessing. They set the norm that AI output gets checked by someone who understands it before it goes anywhere near a client. And they make it clear the tool is for everyone's work, including the hardest and most senior, which is the signal a team actually reads. Do it the other way, with seniors treating AI as something for the juniors, and you get the worst arrangement available: unchecked work produced by the people least able to catch its mistakes, while the judgement that could have caught them sits unused in the corner office.

So the practical move is not another tool licence or another all-staff webinar. It is to get your experienced people using AI on their own real work, closely enough that they can tell where it helps, where it fails, and what good supervision of it looks like. That is what turns a capable tool into reliable output, and it is the work we do in AI Lessons for Leaders: one-to-one coaching built around the decisions, drafting and judgement calls a leader already handles, so the person best placed to get value from AI actually does.

If you have been waiting for someone junior to show you the way, stop waiting. Book a one-to-one discovery call and start with the person whose judgement the firm already trusts, which is almost certainly you.

Questions leaders ask

Should junior staff be kept away from AI, then?

No. Junior staff will use AI and should learn it early. The point is about who leads and who sets the standard. Experienced people should go first so they can work out where the tool helps in your specific work and establish that AI output gets checked by someone who understands it. Once that norm is set, wider use across the team is safer, not riskier.

Why are experienced people better at AI than technical people?

Because the hard part is not operating the software, it is judging the output. Research on professionals using AI tools found that domain expertise, not technical background, predicted who got good results. Someone who knows the work can frame a useful request and, more importantly, can tell when a confident-sounding answer is wrong. Technical fluency helps you drive the tool; expertise tells you whether to trust what comes out.

Our senior people say they are too busy to learn AI. What is the answer?

Start with their real work, not a training course. The fastest way for a busy expert to see the point is to run one task they already do, a first-pass research note, a draft, a summary, through the tool and watch where it helps and where it needs correcting. That takes an hour, not a term, and it tends to convert scepticism faster than any demo, because they are judging it on ground they know.