Listening to Bill: what the “turbulent AI era” means for the people who make workplaces

If you only skim the headlines, the AI debate comes in two settings: the end of the world, or the end of all our problems. Both are wonderful for engagement and useless for anyone who has to sign a lease, plan a floor or run a change program next year. The doomers can’t tell you what to do on Monday. The cheerleaders can’t tell you what happens to the people. Neither camp has ever had to stand in front of a workforce and explain a decision.

Which is why Bill Gates’s essay last month, The turbulent AI era is here. The choices we make now are critical, deserves more attention than the average AI hot take. Gates has spent fifty years as technology’s great optimist, yet his central claim is sober: even under the best circumstances, the transition to the AI era will be one of the most turbulent periods in human history, and nobody has a plan for it. He is equally clear that the benefits, in health, education and productivity, could be profound. Both things are true at once. That is the middle ground, and it is uncomfortable, which is precisely why so few people stand on it.

So let’s take Gates at his word and ask a narrower question: if he’s broadly right, what does it mean for the people who plan, create and manage corporate workplaces, and for the people who occupy them?

What Gates actually says

Three of his arguments matter most for our industry. First, this transition is different because AI substitutes for cognition itself, not just tasks, and it will compress into a decade rather than the generations previous shifts took. Second, the jobs at greatest risk are entry- and mid-level roles (sales, customer support, software, paralegal work), and the effect is already measurable among young workers. Third, and least noticed: Gates argues some work should be deliberately reserved for humans. He calls it “Human Reserved”, after nature reserves, because certain roles carry a value that has nothing to do with whether a machine could technically perform them.

The hollowing out of the workplace

Australia’s largest office occupiers, the big banks among them, long ago moved to agile, activity-based workplaces where people choose settings to suit their work. The design is modern. But look at what actually fills those buildings: enormous volumes of processing work. Transactions, documents, analysis, reporting. That has always been the volume work, and it is precisely what AI absorbs first.

The banks are simply the biggest and most visible example. The same applies to insurers, law firms, government agencies, professional services and thousands of mid-sized businesses whose headcount is anchored in exactly this kind of work. Nobody is exempt. The large end of town just shows us the shape of things first, at scale.

The plausible mid-case is that many large workplaces get hollowed out from within: not emptied, but progressively thinned. And this lands on a market already softened. Most of these workplaces have been underutilised since COVID normalised working from home. AI doesn’t create the underutilisation problem; it compounds one we already have. Worse, some of the buildings involved were purpose-built for single occupiers and were never designed to be carved up and subleased. The adjustment will not be graceful.

Then there is the clock-speed problem. Corporate real estate decisions carry lead times measured in decades; AI-driven organisational change is now moving in quarters. Sydney offers a live illustration. Atlassian Central, the striking timber tower rising beside Central Station, was commissioned as a home for around 4,000 people. It completes next year. In March, mid-construction, Atlassian announced 1,600 roles cut globally, citing AI’s impact on its labour needs. This is not a criticism of Atlassian; every major occupier faces some version of the same mismatch. The point is that buildings conceived in one era are being delivered into another, and if AI changes organisational sizes as fast as Gates predicts, a fifteen-year lease starts to look less like a commitment and more like a bet.

What remains in these thinner workplaces is, in effect, Gates’s Human Reserved domain: judgment, negotiation, creativity, client trust, mentoring, and the collaborative work of deciding what the machines do next. The office becomes its habitat. A smaller footprint in places, but a more demanding one, designed for connection, learning and trust rather than throughput.

There’s a sting in that word learning. Offices have always been apprenticeship infrastructure, where juniors learned by proximity and correction. Cut the junior roles AI can do, and you cut the pipeline that produces the senior people AI can’t replace. We wrote about this expertise cliff over a year ago in The Experience Paradox, and everything since has strengthened rather than weakened the argument.

The question nobody is answering

Here is where the comfortable version of the story falls apart. The standard reassurance is retraining. Consider a 200-person contact centre automated down to ten people overseeing the systems. What, exactly, do the other 190 retrain to do that an AI couldn’t already have been trained to do instead? They cannot all become baristas and physiotherapists. Anyone who answers this question with the word “reskilling” and nothing else has not thought about it.

Gates, to his credit, confronts it directly: the new jobs being created mostly require skills that take years to build, and communities with concentrated job losses suffer ripple effects far beyond the pay packet. Acknowledging a real risk of rising unemployment in the early years of this transition, and the social strain that follows, isn’t dystopian thinking. It’s the reason Gates calls for safety nets, transition planning and deliberate choices now, rather than after trust has been burnt.

What it means for change management

Gates observes that AI, unlike the PC, requires no adaptation from us: “We don’t have to adapt to it because it can adapt to us.” Technically true, and for anyone who has run an organisational change program, dangerously incomplete. The friction was never really about learning the software. It’s about what the technology means for people: their competence, their status, their sense of being needed. Gates himself makes the point elsewhere: work is not just income but dignity and social connection. An AI rollout is never just an IT rollout. It is a negotiation with people about identity, and the frictionless interface makes that negotiation harder to see, not easier to have.

Three predictions, then. Organisations that introduce AI as a quiet headcount play will get quiet resistance: shadow processes, hoarded knowledge, and a collapse of the trust every subsequent change depends on. Organisations that involve their people in deciding what AI does, and what stays human, will adopt faster and keep the capability they need. And change managers will find their work shifting from managing adoption to managing loss: helping people and teams renegotiate what their work means, not just how it’s done.

The middle ground is a choice

The essay’s most useful idea isn’t a prediction at all. Gates insists that where the line falls, between what machines do and what people do, should be decided deliberately, not defaulted into. That applies to nations, and it applies with equal force to every organisation signing a lease, fitting out a floor or planning a program next year.

The dystopians are wrong that nothing can be done. The utopians are wrong that nothing needs to be. Both positions have the same appeal: they let you off the hook. The workplace decisions being made right now, what to automate, what to reserve, what kind of environment the remaining human work deserves, are exactly the choices Gates says will determine which future we get.

We’d rather be in the room where those choices are made than in either echo chamber. We suspect our readers feel the same.

Bill Gates’s essay was published on Gates Notes on 26 August 2026. What’s your organisation choosing to keep human? We’d genuinely like to know.