People and Tokens: Rewriting the Workforce Plan

In the days before I stood up to give this keynote in London, a government intervened in AI for the first time and ordered a model offline. Whatever one makes of that particular decision, it is a useful marker. The ground beneath workforce planning is moving faster than most planning cycles can absorb, and an annual forecasting exercise is not built to register a change of that kind.

I called the talk "People and Tokens: Rewriting the Workforce Plan," and the argument is contained in the title. For decades, organisations planned around people, jobs and headcount. That is no longer the whole unit of account. Leaders are now allocating two forms of capital at the same time — human capital and token capital — and competitive advantage increasingly depends on how well the two are combined rather than on how much of either you hold.

Human capability remains the source of judgment, accountability, creativity and enterprise context. AI supplies scale, speed and a kind of intelligence. Neither substitutes for the other. The genuinely difficult question, and the one I think workforce planning has to absorb next, is how work should be distributed between them.

Workforce transformation is a leadership agenda

If that framing is right, HR's remit widens considerably. Five capability areas follow from it, and each asks a forward-looking question rather than a backward-looking one. Strategic workforce planning asks what workforce we will need. Work design and transformation asks how work should be redesigned using AI. Talent intelligence asks which skills and roles are emerging or declining. AI governance and workforce risk asks what the compliance and ethical impacts are. Internal mobility and reskilling asks how we move people into future roles.

The common thread is that none of these can be answered by managing talent alone. The HR function of the next few years will be measured on its ability to transform work. That is a different job, and it cannot be delegated downward or sideways to a technology programme.

The same question at every altitude

I spent time recently with an enterprise of more than 170,000 employees. The board's questions were not about cost optimisation, which surprised the people who had prepared for that conversation. They asked whether the company was building the right workforce, and how AI would actually affect the business. They also noted that any productivity AI created would more likely be reinvested into growth and acquisitions than used to cut costs.

That is one board, and I would not build a trend on it. But it maps onto something I keep seeing: every leadership layer owns a different piece of this. Boards own workforce readiness. Executives own operating model transformation. Managers own adoption and execution. When any one layer treats it as somebody else's programme, the whole thing stalls in procurement. The organisations that do well here are not the ones with the most tools. They are the ones asking better questions.

The same role, three different answers

Then you try to answer those questions and hit the wall. In one enterprise we ran a similarity model across the job profile, the position description and the requisition for a single role — three documents that are meant to describe the same thing. There was almost no overlap between them. The same role read as three unrelated records.

Historically we compensated for that by buying systems. The pattern is familiar to anyone who has run an HR technology stack: the HCM is thin on applicant tracking, so an ATS goes in; sourcing intelligence is missing, so a tool goes in for that; and thirty or more disconnected systems accumulate. Integration is not understanding. A data lake does not solve a context problem. Workforce architecture matters more than technology architecture, and AI does not paper over the gap — it exposes fragmented data far faster than legacy reporting ever did. Without workforce data integrity, every AI workforce decision becomes questionable.

Where AI is applied matters more than how much

The second wall is placement. AI applied upstream changes the work. AI applied downstream mostly audits work that has already happened. Intent classification on an incoming call is worth more than summarising the call afterwards. AI in supplier onboarding is worth more than AI on exception invoice processing. In talent, job-description generation beats résumé reranking, for the plain reason that a poorly written job description makes reranking pointless.

This also disrupts the build/buy/borrow model most workforce plans still use. In practice it has become build, buy, borrow and balance: which foundation model to back, which agents to buy, and whether an agent already exists somewhere in the stack. There is added pressure from providers who increasingly behave like IT services firms, offering near-infinite cloud credits and encouraging teams to just build agents, which sprays AI everywhere without anyone deciding where it belongs. Those decisions now sit inside strategic workforce planning, because the workforce now includes employees, agents, automation, platforms and partners. AI everywhere is not a strategy. Placement is.

Human capital appreciates

Quantifying the return is genuinely hard, and I would rather say so than pretend otherwise. Studies comparing compute spend in 2024 with 2025 show steep year-over-year jumps and translate that compute into intelligence delivered, arriving at large impact figures. Inside an enterprise it remains difficult to separate automation, which reduces headcount, from augmentation, which raises productivity. There is no clean answer yet. What I can say is that the largest gains tend to appear first as reduced friction, faster decisions and better execution — none of which show up promptly in a budget line.

The objective, in any case, is not automation. It is restoring human capacity for higher-value work. Enterprises survive on quality and service-level guarantees, so without enough human-in-the-loop for judgment and verification, AI simply runs in circles at speed. As machine intelligence becomes abundant, human judgment becomes the scarce input that decides whether all that computing produces value or spins. The human role is not getting diminished. It is getting reprioritised to higher-value work.

Making it concrete: tasks, context and one worked example

Skills only create value when you understand how work is actually performed. Many HR teams have already moved from skills to tasks to subtasks and stopped there, but work does not deploy task by task — it flows through processes and workflows. Adding that layer of context is what makes a task-and-skills model usable for redesign. This is the problem Etter, our work redesign agent, was built for: assessing AI's impact at the task level across thousands of roles, and identifying what should be automated, what should be augmented and what stays human-led. It ingests job architecture alongside process maps, KPIs and policies, publishes a dynamic skills architecture, and connects directly into Workday, SuccessFactors and Oracle to flag data that is wrong rather than assuming the data is clean.

Recruiting is the clearest worked example. Sourcing, screening, scheduling and candidate communications are increasingly automated or augmented. Relationship building, stakeholder alignment, candidate assessment and the hiring decision remain human. The recruiter does not disappear; capacity moves toward advisory work, talent strategy and premium hiring, and the human share of the job grows rather than shrinks. In one engagement, this approach moved roughly 20% of roles across about 20 job families to higher-value work, with a reskilling plan built business unit by business unit. The discipline is managing the rate of compression so quality and service guarantees are never traded for speed.

Three questions to take back

If you do nothing else with this, put three questions in front of your leadership team. Which work should be automated, which augmented, and which should remain human? Which capabilities become more valuable as AI becomes more capable? And how quickly can you redesign work compared with your competitors? The third one is the uncomfortable question, because it is the only one with a clock on it. Every conversation I had after the keynote landed in the same place: the challenge is no longer understanding AI. It is understanding how work changes because of AI.