When AI shares expertise: what should experienced professionals do?

An organisation can use the knowledge of its most experienced people in two ways. It can keep directing the same questions towards them. Or it can make part of their reasoning accessible so that others can make progress themselves. AI expands the possibilities of the second approach. For me, the interesting leadership question begins when that sharing actually works: what responsibilities, capacity and recognition should the people whose experience supplied the starting point receive?

Anyone responsible for clinical education, a scientific department or customer support at a medical technology company should ask this early. A system that distributes existing answers still needs people who can develop new ones. Bringing these activities together is a decision about how the organisation develops.

What the customer-support study shows

Brynjolfsson, Li and Raymond examined the staggered introduction of AI assistance among 5,172 customer-support workers at one software company. The study, published in 2025, uses data from 2019 to 2021. The average number of issues resolved per hour increased by 15%. Less experienced and initially lower-performing workers benefited more; the strongest performers experienced small speed gains and small quality declines. The observational design concerns specific customer enquiries at one company. The author manuscript reviewed here dates from November 2024. [1]

A complementary writing experiment by Noy and Zhang also found larger benefits for work initially rated less highly. The March 2023 preprint read here included 444 college-educated professionals completing short, paid writing tasks. It did not investigate long-term competence development. [2]

The two papers support a limited proposition: the benefits of AI assistance can be unevenly distributed within the same activity. The following argument concerns the organisational consequences if that pattern is confirmed in a local application. It is not an established effect for hospitals or pharmaceutical companies.

Experience contains several capabilities

Consider a fictional example from a medical technology manufacturer's customer-support team. A new employee needs to explain a maintenance procedure. An experienced colleague knows the approved instructions, common misunderstandings and the appropriate contact for an unresolved situation. AI assistance might make some of this accessible, provided the information is suitable and approved for that use.

A closer distinction helps when assigning responsibilities. After receiving assistance, does the new employee know the next step? Does she understand the conditions under which it applies? Can she recognise when the case falls outside those conditions? Can she ask a targeted question about missing information? Each capability can be observed through a different task. A fluent answer can obscure the distinctions.

I would therefore begin by describing a small selection of recurring decisions. Each would have a second version with a changed assumption. For example, the instruction might apply only to a particular device version. The second case would then require the employee to establish which version is involved. This reveals whether someone can reproduce a familiar procedure and also assess its scope.

This example concerns the design of a learning environment. Actual device use and medical decisions remain subject to the relevant approved procedures and professional responsibilities. A persuasive answer in an exercise does not extend anyone's authority to act.

The role of experienced professionals should develop

When assistance works for frequent questions, leaders have choices about how to use the attention it releases. They can increase the number of cases handled. They can ask experienced staff to investigate new classes of problem. They can allocate more time to developing junior colleagues. The appropriate combination depends on demand, workload and quality objectives.

My proposal is to make the contribution of experienced professionals explicit through three responsibilities: explaining difficult boundaries, investigating cases that remain poorly understood, and changing underlying working rules when new knowledge requires it. Each responsibility should produce something visible. That might be an improved example, a resolved interpretation question or a revised instruction.

Time and recognition must accompany the responsibility. If performance is assessed only through completed routine cases, maintaining shared knowledge can become an additional burden. The organisation should agree how much time is available and who decides between competing demands. Simply asking people to document their knowledge alongside their normal workload rarely establishes these conditions.

In a pharmaceutical company, for example, an experienced scientific employee could identify recurring misunderstandings in internal learning materials. In a hospital, an education team could investigate the organisational handovers that regularly leave new colleagues uncertain. Both examples are proposals for possible work. They are not accounts of projects I have carried out.

Make room for a reasoned departure

A shared assistance system should allow someone to depart from its suggestion for a sound professional reason. Otherwise, making a task more consistent could also narrow the thinking around it. This concern does not require a claim that every AI system inevitably weakens experts. A practical organisational question is enough: where does a good explanation go when the suggested answer was unsuitable for a particular case?

I would establish a simple route back. The responsible person records what differed from the usual case, explains the consequence and identifies who can decide whether a change is needed. The technical implementation can remain small. What matters is that these reports are considered and the person raising one learns what happened to it.

Professional errors, ambiguous rules and personal preferences should be distinguished. Preferring a different writing style does not by itself reveal a weakness in a procedure. Demonstrating a missing condition may identify a valuable improvement. Brief joint consideration of selected reports can sharpen this distinction.

The decision to make now

Before a wider introduction, I would ask leaders to agree three things. Which recurring decisions should become more accessible? Which capabilities should new employees develop for themselves while using assistance? And what new responsibilities will experienced people take on if fewer routine questions reach them?

After an initial period of use, the same group should examine whether that distribution has actually emerged. Are difficult cases recognised earlier? Can new colleagues explain their decisions? Are useful improvements being made to the shared knowledge base? These observations complement ordinary performance measures with a question about whether the organisation is developing its understanding.

As a physician and founder, I am particularly interested in the connection between work today and the ability to act tomorrow. An organisation can gain a great deal by making experience more accessible. It also needs to ensure that new experience can develop. Leaders can make that responsibility concrete when they decide how work will be allocated.