Capability Is Not Strategy: What AI Leaders Cannot Tell You About Your Workforce
Yann LeCun, one of the most senior figures in modern AI, recently pushed back publicly on Dario Amodei’s claim that AI will wipe out half of entry-level white-collar jobs. LeCun’s argument was not that the technology is weak. It was that computer scientists, himself included, are not the right authorities on what large-scale AI adoption will do to labour markets. That job, he said, belongs to economists who actually study how technological change moves through employment, wages and demand.
He is directionally right, and senior leaders should take the point seriously. The people who build AI systems understand capability, speed and unit economics. That is not the same as understanding workforce design, management load, service quality, trust, consumer demand, or the commercial system the technology is entering. Treating one as a substitute for the other is how businesses end up making bold operating decisions on the strength of someone else’s marketing deck.
The Collapse Between Capability and Strategy
This is the core risk in the current cycle. Organisations are being encouraged to confuse capability with strategy. AI can now do a long list of things it could not do eighteen months ago. That is real. It does not follow that every task AI can perform should be automated, that every function built around those tasks should be hollowed out, or that the efficiency gain at the task level will show up cleanly as a profit line at the business level. Capability tells you what is possible. Strategy is meant to tell you what is wise.
Nowhere is this collapse between capability and strategy more common right now than in HR. It is worth looking at the example most often cited by those arguing that AI is ready to replace people functions: IBM’s AskHR.
The IBM Case Study That Proves Less Than You Think
IBM’s own published case study is striking, and worth reading honestly rather than selectively. The platform handled more than 11.5 million employee interactions in 2024, resolved roughly 94% of common inquiries without human involvement, has cut HR support ticket volume by around 75% since 2016, and contributed to a 40% reduction in HR operating costs over four years. Those are serious numbers, and senior leaders should not dismiss them. Routine, transactional, policy-lookup HR work can genuinely be automated at scale.
But the part that tends to drop out of the retelling is the shape of the model. IBM describes AskHR as a two-tier system. AI handles routine inquiries. Human advisers handle the complex ones. The machine is not doing HR. It is handling the most standardised slice of people administration and routing everything else into a smaller, more specialised, more expensive human layer. That is a substantially different story from “AI replaced HR”, and senior leaders should read it accordingly.
The distinction matters because it applies everywhere, not just in HR. Automating tasks is not the same as replacing functions. Removing routine labour does not automatically remove the work. It often shifts the work into oversight, judgement, escalation, service recovery, exception handling, and relationship management, where the remaining people are more senior, more exposed and more expensive. You have fewer hands, but harder problems and a thinner margin for error. At Esbee, when we work with businesses on organisational design and restructuring, this is the question we press hardest: are you removing cost, or are you moving it somewhere you cannot see it yet?
Automating tasks is not the same as replacing functions. Removing routine labour does not remove the work. It shifts it into oversight, judgement, and exception handling, where the remaining people are more senior, more exposed, and more expensive.
The Pipeline You Are Quietly Destroying
There is a pipeline question that rarely features in the sales conversation. Routine transactional work has historically been the training ground for the next generation of managers, advisers and specialists. Close that door cleanly, and you may well reduce short-term operating cost while quietly destroying your medium-term management pipeline. The people who later handle the hard cases learn their craft by first handling the easy ones. If AI absorbs the easy ones, the question of how you produce the next cohort of competent HR professionals, legal advisers, analysts or line managers is a live strategic issue, not a training department detail.
Businesses used to understand this instinctively. The Big Four accounting firms are the obvious example. They built their practices around a simple premise: recruit far more graduates than present demand requires, train them properly, and accept that most will leave. The ones who stay become partners. The ones who leave become clients, contacts, and institutional knowledge in the market. Nobody hired a 22-year-old because they needed that specific person to do that specific job for the next forty years. They hired them because a profession, and a firm, requires a throughput of trained people. The label attached to the future role did not need to be legible at the point of hire. The discipline of bringing people through did.
That instinct is what is being lost. Because we cannot, today, write down the exact job title a 22-year-old will hold in 2038, we have started to treat the trainee layer as unjustifiable cost. This is historically illiterate. The 2010s budding journalists and copywriters became this decade’s heads of content and SEO managers. The machine-learning engineers of the early 2020s are still here, most of them now labelled LLM integration engineers. Very few people are still doing at 40 what they thought they were doing at 21. Job titles drift. Technologies turn over. The trained person, the one who has learnt how to think inside an organisation, is what actually carries forward. An organisation’s capability is built in that layer, not on the cover slide of the tool that currently happens to sit on top of it.
The Demand-Side Problem Nobody Wants to Discuss
There is a wider point still, and it is the one capability-led arguments tend to skip. A market is not only a production system. It is also a demand system. Firms do not sell into a vacuum; they sell into an economy whose purchasing power is underwritten, in large part, by wages. If enough firms move far enough, fast enough, toward removing paid human participation from the value chain, the aggregate demand that justifies their own output starts to erode. That is not an abstract theoretical concern. It is a commercial one.
Henry Ford is often cited here, sometimes carelessly. The verifiable point is that in 1926, Ford Motor Company moved its factories to a five-day, 40-hour working week, helping normalise what eventually became the two-day weekend. Ford’s own commentary at the time made it plain that this was not simply a welfare gesture. It reflected a recognition that a mass-production economy requires a mass consumer base with both the income and the time to buy what it is producing. That principle has not gone away. An economy that gets very good at production, while getting careless about who can afford the output, does not simply correct itself over a polite quarterly cycle. Historically, it corrects through recessions, asset repricings, political instability and prolonged wage stagnation. The market usually survives technological disruption. It tends to survive it through ugly adjustment.
What Ugly Adjustment Actually Looks Like Inside Your Business
It is worth being clear about what ugly adjustment means in practice, because senior leaders will carry the consequences inside their own organisations. It means displaced workers, thinned-out middle layers, degraded service, strained remaining managers, and a noticeable gap between the slide deck that justified the change and the operating reality that follows it. HR and management teams are the functions that carry most of that weight. Pretending the cost is zero is not commercial realism. It is better PR.
What a lot of organisations are doing now looks structurally similar to hollowing out a load-bearing layer because nobody can immediately label what it is holding up, then booking the saving as efficiency. It is efficient only until you need the layer back. The HR admin role quietly replaced by an LLM this year is not really a clean choice between a person and a chatbot. It is a choice between paying for a function now and paying, later, for the absence of the people that function was quietly producing. How many capable future operators, managers and specialists, of jobs we have not yet invented, are being cut off at the entry point because we cannot neatly justify them on this year’s workforce plan?
Organisations that think of workforce design as a proper commercial discipline, rather than a cost-cutting exercise in another wrapper, will treat that as a front-rank business question. Esbee’s management consultancy work with boards and operating partners addresses exactly this: the gap between the efficiency story on the slide deck and the operating reality that follows it. HR leaders who are serious about the long-term health of their organisations should be willing to say so plainly in the room where the decision gets made.
The Three Questions You Should Be Separating
None of this is an argument against AI. It is an argument for not letting the people who built the technology also become, by default, the authors of your workforce strategy. Their authority is real on capability. It is not automatic on consequence.
The practical position for founders, directors and senior HR leaders is to separate three questions that are currently being conflated. First, what can the technology actually do in our context, tested on our own processes rather than on a vendor demonstration. Second, what should be automated, given the trade-offs on pipeline, judgement, quality and trust, not only the trade-off on headcount cost. Third, what does our operating model need to look like on the other side, including who carries the harder work, how the next generation of capable people is developed, and how the organisation recovers when, not if, the automated layer gets something wrong.
An HR MOT provides a structured baseline for the second question: before deciding what to automate, understand what your people function actually does, where the value sits, and where the risk lies. The businesses that skip this step and automate from a vendor demonstration rather than from an honest assessment of their own operating model tend to discover the consequences on their own P&L.
The Uncomfortable Conclusion
Businesses that take those three questions seriously will use AI well. Businesses that take the task-level efficiency story at face value, and mistake it for an operating-model answer, will discover the difference in their own results. That is the gap between capability and strategy. Closing it is what senior management is for.
The technology is real. The efficiency gains on routine tasks are real. The temptation to treat those gains as a workforce strategy, to let the people who built the tools also author the plan for how your organisation uses them, is where the risk sits. The question is not whether AI can do the work. The question is whether the decisions you are making about your people, your pipeline, and your operating model are being made with the same rigour you would apply to any other capital allocation decision, or whether they are being made on the strength of a capability demonstration and a cost-per-head calculation.
One of those is strategy. The other is a slide deck with a savings number on it. The difference matters, and it will show up.
If you are making decisions about AI, automation and workforce design and want an independent view on what the operating model should look like on the other side, talk to us. Esbee’s management consultancy team works with boards and senior leaders on organisational design, and our HR services team provides the baseline assessment of your people function that any automation decision should start from.