Abundant Expertise: Middle Managers Face the AI Test

Toronto, 23 September 2026: the argument

At Toronto's Nrth Festival on 23 September 2026, Frankie Llewellyn-Thomas, a partner in KPMG Canada's Transformation practice, brought a structural shift into focus, reported on 28 September by HCAMag^[1]: expertise has begun to separate from experts.

For centuries, specialist knowledge stayed inside people's heads. Schools, universities, law firms, audit practices and governments were built around that condition of scarcity.

Generative AI, in the partner's view, makes a level of expert analysis and reasoning accessible within seconds. The output remains imperfect and rather inhuman, to use her words, yet it is available at scale. A rare resource becomes abundant, and competitive advantage moves elsewhere.

What matters here for those running organisations has little to do with the tools. It has to do with the point at which judgement forms in the people who will have to use them.

The suitcase of books, 2008

The partner described walking into a tax law exam in 2008 with a suitcase full of texts: statutes, regulations, case law.

The purpose of the test was knowing where to look for the answer. Anyone who did not know which part of the code to consult was, for the purposes of that exam, facing information that did not exist.

In 2026 a fresh graduate asks the same questions in plain language and receives a reasoned answer within seconds. The distance between those two scenes measures how much the value of information retrieval has changed.

The point deserves precision. Retrieving information loses value; deciding what to do with it gains value. These are two different skills, and organisations have always trained them together, inside the same craft.

The craft remains; the first of the two skills is leaving its path.

What stays rare

The argument continues with a precise list: judgement, leadership, relationships, ethics, accountability, trust. These are the resources that, according to the KPMG Canada partner, become the scarce factor of the coming decade.

Corroboration comes from another source cited in the same report. PwC's Global AI Jobs Barometer records rising demand for judgement and creativity in the roles most exposed to AI.

A parallel reading appears in the essay "AI Made Knowledge Cheap. The Scarcity Moved Into Nature" (Adaptation Intelligence^[2]), which describes the same shift: when knowledge becomes cheap, value migrates towards whatever remains hard to replicate.

The risk, for anyone reading quickly, lies in treating judgement and trust as character traits. They are the outcome of a path: they form inside an architecture of roles, of attempts and of errors corrected by someone more experienced.

The limit of the argument

The thesis deserves an honest test. Access to an expert answer arrives within seconds; assessing that answer requires someone capable of recognising it.

Here the operational paradox appears. The system produces expert-level material, and it takes a competent person to validate it. That competence comes from the very apprenticeship that automation is shortening.

Organisations that neglect this link find themselves, a few years on, with plenty of output and little capacity for control. The cost surfaces late, in the form of errors corrected too long after the fact.

One answer already exists: make validation an explicit task, assigned and measured, rather than an implicit residue of the craft.

Where judgement is really learned

Professional judgement comes from a long apprenticeship. A junior writes an imperfect draft, a manager corrects it, and the logic of the correction becomes a criterion.

That cycle lives in repetitive work: research, first drafts, reconciliations, summaries. It is the band of tasks that generative systems absorb first.

Here the consequence becomes measurable on careers rather than on tools. When the entry door narrows, the learning curve loses its first steps.

The entry-level data point in this direction. The Federal Reserve Bank of New York measures unemployment among recent graduates at 5.7%, and entry-level payrolls show a relative 16% decline. The World Economic Forum describes careers compressing towards the top.

The question for a middle manager changes accordingly: which decisions to entrust to people with little experience, now that the mechanical part of the work is leaving their hands.

The bottleneck sits higher up

Adoption and readiness remain two different things. Cornerstone measured the gap: 94% of leaders see AI affecting roles, 17% of people feel ready.

The distance between those two figures describes the scope of the problem. Those leading can see the change; those doing the work are waiting for the conditions to get through it.

The evidence gathered so far suggests an order of priority: leadership readiness counts for more than the technology stack. Excellent tools in the hands of unprepared managers deliver less than average tools in the hands of AI-literate ones.

The middle role carries the heaviest weight. It translates a strategy written on the top floor into daily tasks, and decides every day who learns what. The 56% wage premium measured by PwC on AI skills shows where the market is already paying for that translation.

What the organisations closing the gap do

Organisations coping with the transition work on converting talent internally. Morgan Stanley reports 98% adoption among its advisers; JPMorgan has gathered 200,000 voluntary sign-ups to its training programmes.

LHH/Adecco research covering 11,000 executives points to a clear economic case: redeploying people costs less than firing and rehiring. Few organisations have the infrastructure to do it.

Operationally, high-functioning teams redesign three things:

• the entry door: junior tasks chosen for their developmental value rather than their cost
• the review cycle: a correction that is explained is worth more than a correction that is applied
• the manager's time: protected hours for teaching, counted as output

One asset remains available inside many organisations. The people who learned the craft before AI can write down the criteria they use. That material becomes both training for younger colleagues and instruction for the systems.

The common principle is simple. Exposure to judgement becomes a designed condition rather than an individual choice.

The conversation to take to the top

For a CEO, the question to bring to the board concerns organisational readiness more than the number of licences activated. How many people, over the next twenty-four months, will reach a role of independent judgement?

For a CHRO, the priority shifts to organisational design: redesigning the first eighteen months of a career is worth more than a course catalogue.

For a CFO, the investment with a documented return remains internal redeployment, at a per-person cost lower than external replacement. For the board's Talent committee, the metric to follow becomes the average time it takes to develop a middle manager.

The final design question is direct: which part of the work that builds judgement has already passed to the machines, and who, inside this organisation, answers for its replacement.