“For every one of my engagements,” Walsh told Fortune, the question is, “What is my mix of labor? What’s my mix of technology? And what’s the overall cost of delivering that engagement?”

He said he would expect the “labor cost in the mix” to go down and for his technology costs within that same engagement to go up. “And at the end of the day, I’m going to be able to run a lot more volume through my business in ways that I couldn’t before.”

That logic—lower labor cost per unit of work, more total volume, net growth—is the quiet calculus running beneath nearly every major AI investment decision in corporate America today. And according to the 2026 KPMG U.S. CEO Outlook Pulse Survey, released Tuesday, the pace at which executives are moving toward that model is accelerating far faster than the public debate and hype around AI jobs has accounted for.​ It’s “dizzying” to do business in a genuine economic boom, he added.

They think the hype is real—just not yet

The survey, which polled 100 CEOs of large U.S. companies, found that 77% agreed with the statement that generative AI was overhyped over the past year, but also that its true disruptive potential over the next five to 10 years is likely to be under-hyped. It is a distinction that doesn’t quite land in what Walsh called “the noise” of the broader conversation, which has oscillated between Silicon Valley triumphalism and doomsday predictions about mass unemployment. The CEOs that KPMG surveyed largely rejected both poles. What they are describing instead is something more structurally significant and harder to see coming: a gradual, then sudden, rewiring of how work gets done and who—or what—does it.​

“There is no doubt that every single layer within the labor pool is going to be disrupted,” Walsh said. “But anyone who tells you what it’s going to do or knows what the shape of it’s going to be isn’t being truthful, because it’s unclear at the moment.”​

The numbers in the survey support that uncertainty as well as the scale of the bet being made in spite of it. Nearly 80% of CEOs said they are allocating at least 5% of their total capital budgets to AI, and 41% are putting in at least 10%. Thirty-five percent are spending between 11% and 20% of their entire capital budget on the technology.

For context, that level of allocation rivals what companies were devoting to cloud infrastructure at the height of the cloud transition—and the cloud took a decade to fully reshape the economy.​

The jobs that are ‘a scary place to be right now’

The workforce picture that emerges is one of deliberate, if uncertain, transformation. Fifty-five percent of CEOs said AI will lead them to increase hiring over the next year. Walsh said his own headcount at KPMG is not down, but the composition of who he’s hiring has changed fundamentally.

“We’re hiring technologists in ways that we never did before,” he said. “We’re hiring people that we call orchestrators, people that are actually managing gigantic parts of our workflow to make sure they’re complete, they’re accurate, that they’re getting to the right output.” KPMG also told Fortune that it needs to hire AI agent adoption strategists (responsible for aligning AI agents with strategy, design, and workforce planning, and ensuring adoption among workers); AI agent orchestration engineers (connecting agents, tools, and workflows, and defining autonomy and guardrails for agents); and AI agent operations managers (managing agents’ day-to-day performance, incidents, and changes).

That is the new shape of white-collar work that is coming into focus: not elimination but stratification. The jobs most at risk, Walsh said, are unambiguous: “You can look at those types of jobs that are repetitive tasks, people that are doing the same thing every single day, day in and day out. That’s a scary place to be right now.”

But he argued that most knowledge workers are not in that bucket. Work isn’t “just one thing” for this breed of white-collar worker. “It’s about building relationships. It’s about building business. It’s about taking judgments on what work I do … Not all of that just fits nicely into an automated solution.”

Still, two-thirds of CEOs surveyed by KPMG admitted they have not yet actually redefined roles or career paths to account for AI, a striking admission given the scale of investment underway. The survey also found that 31% of CEOs cited their top concern about AI’s impact on leadership development as reduced opportunities for early-career employees to build judgment through real-world experience. The worry, in plain terms, is that companies may be training a generation of managers who have never had to figure anything out for themselves.​

The pressure to keep up

The metric Walsh is watching—labor cost margin—is essentially the financial expression of all of this. It captures the substitution of technology for labor; the expansion of capacity without proportional headcount growth; and ultimately the productivity gains that every CEO is under pressure to deliver. And that pressure is real, he agreed, as every CEO is under the microscope, expected to increase that labor cost margin.

“It’s stressful if you’re not investing, if you’re not keeping up,” Walsh said. “Because if you’re not keeping up, you have the risk of losing market share.”

That competitive pressure—to automate faster than your rivals; to find productivity gains before investors demand to see them; to retrain a workforce for jobs that don’t fully exist yet—is the hidden texture of the AI moment that this survey captures. Sixty percent of CEOs identified the pace of AI innovation and risk management as the single biggest factor impacting their organization’s prosperity over the next three years. Not tariffs. Not interest rates. Not geopolitics.​

“It’s dizzying,” Walsh acknowledged, adding that he sees CEOs as very resilient in the mid-2020s.

The machines aren’t taking over. But the people running the largest companies in America are quietly, methodically, recalculating exactly how many humans they need, and the number they’re arriving at looks very different from the one they started with.