The Workforce You Can't Plan For

How the AI industry is forcing companies to rethink the relationship between talent, growth, and real estate.

I have sat in hundreds of boardrooms advising on real estate strategy. I have never seen the gap between capital commitments and workforce clarity this wide — or this expensive. After sixteen years at Deloitte and running West Coast occupier strategy at CBRE, I've learned to read that gap before it shows up on a balance sheet.

In 2025, artificial intelligence companies raised $211 billion in venture capital, up 85 percent from the prior year.¹ Half of all global venture funding went to AI. Sixty percent of it landed in the San Francisco Bay Area, concentrated in just 92 companies.¹ OpenAI now occupies more than 1.2 million square feet of office space in San Francisco — roughly where Facebook stood at its 2018 peak.² Sierra AI signed a 300,000-square-foot lease so quickly that its previous headquarters was listed for sublease before employees had even packed their desks.²

That is the expansion story.

The labor story is harder to square. In the first three months of 2026 alone, the U.S. tech sector shed more than 52,000 jobs.³ San Francisco lost 4,400 jobs in 2025, and the city's unemployment rate reached 4.1 percent in January 2026.^4,5 Nationally, tech employment has been under pressure for several years now, even as AI capital and leasing continue to surge.⁶

This fracture exposes a vulnerability the market still tends to glide past. The AI economy has gotten very good at scaling capital. It has been much less consistent in defining the workforce that capital is actually meant to build.

The Commitment Gap

For a long time, the planning logic inside companies was familiar enough that most executives barely had to say it out loud. Revenue targets drove headcount. Headcount drove real estate. Real estate drove capital commitments. You built a hiring plan, translated it into seats, and then went out and leased the space.

That logic doesn't hold as neatly as it used to. I've started calling it the Commitment Gap: the growing distance between what a company signs and what it actually knows about itself.

AI changes the relationship between labor and output. A technical breakthrough can make one category of work thinner almost overnight and, at the same time, create urgent demand for a different capability that barely existed six months earlier. BCG estimates that as much as 55 percent of U.S. jobs could be reshaped by AI over the next two to three years.⁸ For the companies building these systems, that is not some broad future-of-work abstraction. It is already showing up in org design, recruiting decisions, and the way leadership teams think about growth.

What I keep seeing in executive conversations is that the problem often starts with the word headcount. It sounds neutral. It isn't. It carries an older assumption: that people can still be planned as relatively stable units of production. That assumption is getting weaker by the quarter.

One excellent engineer, working with the right AI tooling, can now produce what once required a meaningfully larger team. Not in every function. Not in every company. But often enough that it changes the planning discussion. And once that changes, the real estate discussion changes too.

This is where the tension becomes more than theoretical. The CHRO may need elasticity because the business could hire aggressively in one area, pause in another, and rethink a third within the same fiscal year. Meanwhile, the CFO and Head of Real Estate may still be underwriting long-duration commitments that assume a level of organizational stability the business no longer has.

That is not just a coordination problem. It is a strategic mismatch, and an expensive one if leadership gets too confident too early.

Two Very Different Boardrooms

What makes this moment especially interesting to me is that not every growth boardroom sounds like this.

I spent enough years in more traditional industries to know the contrast. In a high-performing consumer packaged goods company — especially one that has posted double-digit growth and a strong CAGR over a decade — the conversation is rarely about whether the business model itself still makes sense. That part is already proven. The debate is usually about how far the company can push innovation without weakening the system that made the growth possible in the first place: the brand, the margin structure, the retailer relationships, the supply chain, the operating discipline. In those rooms, growth creates confidence, but it also creates restraint. Success gives you more to protect.

AI companies are dealing with a different set of questions. Their challenge is not just how to scale. It is what exactly they are scaling toward. Where will durable value actually sit? Which capabilities are truly strategic, and which are already moving toward commodity status? How much of today's organizational design will still make sense a year from now?

You can feel the difference in the room. One boardroom is managing the burden of success; the other is managing the uncertainty of emergence. Both may be growing, but they are not growing from the same premise. A mature CPG company is usually trying to extend a model that already works without damaging it. An AI company is often still trying to determine which parts of its model will prove durable enough to build around. Those are very different executive conversations, and they lead to very different decision styles. The first tends to be disciplined, pattern-based, and protective of what has already been earned. The second is more provisional. Faster. Sometimes more improvisational than anyone in the room would like to admit.

That difference shows up in talent first.

The Talent Layer

Technical skills are losing shelf life faster than many organizations can reprice them. Skills that commanded a premium in 2022 are already being standardized, automated, or folded into a broader tool set. The premium is moving toward judgment: framing the problem correctly, challenging machine output, making decisions with incomplete information, and knowing when not to trust a system simply because it appears efficient.

Deloitte found that only 7 percent of leaders feel equipped to help their workforce adapt continuously to this kind of change.¹⁰ The industry knows how to hire coders. It is still learning how to hire leaders who can govern intelligent systems.

Burnout sits right underneath that. Tech sector job cuts were up 51 percent year to date in 2026.¹¹ Nearly half of employed tech professionals report severe burnout, and most of that group say they are actively looking to leave.¹² That does not stay at the level of morale. It moves into execution. It shows up in slower decisions, weaker judgment, and the quiet loss of people a company can least afford to lose. In my experience, that loss rarely shows up in an exit interview. It shows up six months later, when the decision that needed the right person didn't get made.

Culture matters here too, though I would put it more plainly than most management writing does. In a moment like this, culture is not some soft backdrop. It is how decisions get made when roles are shifting, the data is incomplete, and the operating model is being rewritten in real time. If leadership treats that as secondary, it is missing one of the things holding the whole system together.

Where It Lands: The Office

Eventually, all of this lands somewhere very concrete: the office.

The leasing boom in San Francisco is real. Full-year office leasing activity reached 11 million square feet in 2025.¹³ But the more revealing question is not whether AI companies are taking space. It is what kind of space they are willing to take, how long they are willing to commit to it, and how much flexibility they are trying to preserve. In earlier tech cycles, companies signaled arrival with a skyline logo and a long-term tower lease. Many AI firms are taking a different view. Boston Properties put it plainly with investors: AI demand is not really a tower business.¹⁴

That matters.

Even in a period of extreme growth, the companies thinking most clearly about risk are not necessarily optimizing for permanence. They are optimizing for room to move. Harvey AI, for example, recently doubled its San Francisco footprint to 150,000 square feet, largely because the deal gave it immediate occupancy and room to grow without locking the company into an old-style anchor.¹⁵ Lease terms have shortened too. For AI startups, the average term is now closer to 3.5 years than the decade-long commitments that used to define a major tech statement lease.¹⁶

In calmer markets, efficiency usually wins. In unstable ones, flexibility is survival.

The same logic carries into workplace design. If AI is absorbing a meaningful share of routine execution, companies do not need to plan around simple seat counts in the way they once did. They need spaces that support the kinds of work machines still struggle with: judgment, synthesis, trust, debate, and hard calls made with incomplete information. That is a different brief. Less about density for its own sake. More about creating conditions where high-value interaction can actually happen.

The Deeper Mistake

This is why I don't think real estate can still be treated as a downstream facilities decision. It has become one of the clearest physical expressions of what leadership believes about the future of the organization — even when leadership would rather pretend those beliefs are firmer than they really are.

And that, to me, is the deeper mistake many companies are making right now. It is not that they are moving too slowly on AI. It is that they still think they can modernize the technology layer while leaving the planning model underneath it mostly intact.

They can't.

Once AI changes the relationship between labor and output, it also changes the relationship between hiring plans and buildings. That is the issue underneath the current expansion cycle. The real question is not how much space a company needs. It is whether the CFO, the CHRO, and the Head of Real Estate are in the same room when that decision gets made — and whether any of them has fundamentally changed how they think about it in the last eighteen months.

In most companies I talk to, the answer to both is no.