Drucker wrote that in 1963, when efficiency still meant stopwatches and time-and-motion studies. He was drawing a line most organisations still blur between: doing things right (i.e., efficiency) and doing the right things (effectiveness). His warning was that the first, performed well enough, can disguise the failure to ask the second.

AI is the most powerful efficiency instrument ever built. Which makes it the most powerful way yet devised to do work faster that should have been redesigned — or stopped.

Two editions ago I priced the cost of not investing in your people. Last time, the return on doing it. The question underneath both was whether people are a cost to be managed down or an asset to be funded.

This year that question stopped being philosophical. It now arrives as a business case with a number attached, and it forks two ways.

Most organisations will claim route two in the town hall and fund route one in the model. And they will call it discipline.

There is a real case for route one, and it deserves to be stated at full strength. Margins are under pressure, shareholders have already priced AI in, and some work genuinely does disappear — the tasks now scored as fully automatable are not coming back, and pretending otherwise is its own negligence. An executive who cuts is not being stupid. They are responding to the only incentive that arrives with a date.

Grant all of that.

So let me put my position before the evidence: the substitution route isn't cautious. It's late. The companies capturing the largest AI gains are growing headcount, not shrinking it — and the forecast the cutters are working from has already been overtaken by the one they're reading.

Tasks first — because the fork exists at the level of the work, long before it reaches a spreadsheet.

— Tanguy

Four numbers, from the tasks inside every job down to the person doing them. Skim them here; the one worth sitting with is below — along with the question I'd put to your CFO this quarter.

⚙️ THE WORK ITSELF ⚙️ | New Work, New World 2026: How AI Is Reshaping Work Faster Than Expected | Cognizant (2026) | How much of the work has actually changed? Cognizant scores job families on two axes: exposure to AI, and the velocity at which that exposure is moving. The three-year shift is the story. Tasks classified as fully automatable have risen to 10%, from 1% three years ago, two points short of the 12% originally forecast for 2032. Nearly 40% are now partially or mostly assistable, up from 15% previously—overshooting the 2032 forecast of 31%. Cognizant puts $4.5 trillion of labour inside that velocity. But 40% of tasks assistable is not 40% of people removable. Every one of those tasks is now either a line in a headcount reduction or a brief for a redesign. Same task. Two entirely different business cases. And the six years you thought you had to choose have already gone.

📈 THE MARKET 📈 | 2026 Global AI Jobs Barometer: Two Futures for Jobs in an AI Era | PwC (2026) | So which route actually pays? PwC analysed over a billion job advertisements across 27 countries and territories, combined with company and task data, and reports what it calls a counter-intuitive finding: greater AI exposure is linked to headcount growth, not decrease. Headcount at the most exposed organisations is growing at twice the rate of the least exposed, and wages are rising faster, too. On revenue per employee against a 2018 baseline, the most exposed show 33.5% productivity growth; the top fifth, PwC's "superstars", show 163%. PwC's own reading of why: the firms achieving the largest gains are redesigning how they operate rather than simply substituting. Reinvention, not replacement. And the capability is repricing — workers with AI skills now command a 62% wage premium, up from 57% last year.

A caveat worth taking on the chin rather than burying at the end: this is observational data. Well-run companies may both adopt AI more quickly and grow for reasons unrelated to AI, and PwC says as much. A strong relationship with a plausible mechanism, not proof of cause — and anyone selling it to you as proof is selling you something else too.

🏢 THE BOARDROOM 🏢 | Global Talent Trends 2026: Driving Exponential Performance | Mercer (2026) | And what are executives planning instead? Mercer surveyed approximately 12,000 respondents across 16 geographies — 825 C-suite, 1,650 HR leaders, 9,250 employees, fielded September to October 2025 — and found that 99% of executives expect AI to lead to at least some headcount reduction within two years, with 98% planning organisational design changes over the same period. Near-universal intent to restructure, and near-universal expectation that people are the adjustment variable. Set that against PwC's evidence and the two do not describe the same strategy. One of them is the route the market is rewarding. The other is the route almost everyone has chosen.

👷 THE PERSON DOING THE WORK 👷 | Global Talent Barometer 2026 | ManpowerGroup (2026) | Now bring it to the desk. ManpowerGroup surveyed 13,918 workers across 19 countries, weighted to each country's worker population by gender, age, and region, with each country weighted equally, fielded from 1 September to 1 October 2025. 43% now fear automation may replace them within two years. The adoption curve tells the rest: regular AI use at work rose 13% to reach 45% of workers, while confidence in using that technology fell 18%. Using it more. Trusting themselves with it less. Nobody needs to leak the memo about route one. It arrives in every all-hands, every pilot, every reorganisation. And it lands on people who, at that moment, are being asked to adopt the very thing they have been told will replace them.

Sit with the two middle numbers, held against each other.

99% of executives expect AI to reduce headcount. The companies actually capturing the largest AI productivity gains are adding it.

Both cannot be right about what works. And notice which one is already shaping behaviour inside your organisation: not the evidence, but the expectation. 43% of workers are bracing for replacement in an economy where the most AI-exposed firms are hiring faster and paying more than anyone else.

The fear is not irrational. It is well-informed — informed by what their own leaders have told them they intend. And it arrives well ahead of any evidence that the intent is the right one.

Somewhere in your organisation, someone volunteered for the AI pilot, sat through the training, and is now quietly better at the thing they were told would replace them. They have not mentioned how that feels. They will not.

So why does almost everyone choose it anyway? Not because it is better. Because it is modellable. A headcount saving fits in a spreadsheet — one line, one number, one date. A redesign does not. It is second-order effects arriving over three years, through people not yet hired. The fork is rarely a strategic disagreement. It is an artefact of what a business case can hold: we choose the route the finance template can express.

I spent twenty years inside firms that build those models, and that line is rarely argued over. It is inherited from the last restructuring, because that is the shape the system already holds.

Which is where I'd put the uncomfortable question, and it isn't for the CFO. That decision was taken in a room you were in. If you are the CHRO, you watched the benefit line get written as headcount; you knew what it committed the organisation to, and the moment to contest it was before the model was locked. Finance doesn't choose the fork. Everyone who lets finance choose it chooses it.

And there is a cost to that choice nobody books. The most expensive thing about route one may not be the redundancy programme. It may be that you announced it.

Which raises the question this newsletter will spend the coming editions on. If an announcement alone can reset what people need from you, then employee expectations are not the stable backdrop your EVP assumes. They are a live variable — and they have already moved.

Your turn. One ratio, and you can find it before lunch: your skilling budget as a percentage of your AI technology budget. Most organisations discover nobody has ever calculated it, because it was never a number anyone managed. (Ask your FP&A lead. Watch the pause). What's yours?

If the organisation won't fund the transformation, the workforce will fund it themselves — and this year, for the first time, we can see them doing it. Employability has become an expectation. Next: what employees are prepared to give up to get it, and what that costs an employer who hasn't noticed.

— Tanguy

The Substitution Route — AI adopted to remove work and reduce headcount; the benefit booked as cost saving. Fast to model, and on the current evidence the slower route to productivity.

The Reinvention Route — AI adopted to redesign work and redeploy people; the benefit booked as capability. Slower to model, and the route on which this year's productivity leaders are found.

Velocity score — Cognizant's measure of how fast a job family's exposure to AI is changing, distinct from how exposed it already is.

Employability — the expectation that an employer keeps a person marketable outside the organisation, not merely useful inside it. Next edition's subject.

Cognizant. (2026). New work, new world 2026: How AI is reshaping work faster than expected. cognizant.com

Drucker, P. F. (1963, May). Managing for business effectiveness. Harvard Business Review, 41(3), 53–60. hbr.org/1963/05/managing-for-business-effectiveness

ManpowerGroup. (2026). Global talent barometer 2026. manpowergroup.com

Mercer. (2026). Global talent trends 2026: Driving exponential performance. mercer.com

PwC. (2026). 2026 global AI jobs barometer: Two futures for jobs in an AI era. pwc.com