AI Agents Outnumber Humans: How Tech Leaders Handle High-Stakes Decisions

Executives gathered in Aspen this month. They confronted a stark new reality. AI agents now operate at scales that leave no room for constant human oversight. The conversations at Fortune Brainstorm Tech 2026 revealed both the enormous gains in speed and the persistent risks when machines handle decisions with real consequences.

Zach Maybury watches trillions of transactions flow through DraftKings systems. The chief technology officer described a material increase in the speed with which his team can create things. Yet the introduction of agentic AI changes everything. Agents speak directly to other agents. The resulting volume and complexity overwhelm traditional controls. “I can’t insert humans into those loops,” Maybury said. “We will never have enough humans to insert in all those loops.”

Short pause. The admission hung in the conference air.

LaShonda Anderson-Williams drove the point home from the Salesforce perspective. The chief customer and commercial officer reminded the audience that stakes vary wildly across industries. “In high-stakes environments like health care, it’s not taking the wrong tee-shirt size if you’re a retailer, it’s a life on the other side of it,” she said. Her warning underscored a central tension: businesses chase efficiency while lives, money, and reputations sit on the line.

But what replaces the human in the loop when scale makes oversight impossible? Panelists offered no single blueprint. They shared hard-won practices instead. Anderson-Williams stressed the need to examine each AI use case with clear eyes. Companies must define the exact outcome they seek before deployment. Without that discipline, systems drift into dangerous territory.

Governance emerged as the recurring theme. Not the static policies of yesterday. Something that grows with the technology. Anderson-Williams noted many organizations had simply purchased tools and bolted them onto existing operations. They skipped the hard work of establishing rules. “A lot of people just ran and bought a lot of different tools and technologies and just bolted them on, and there wasn’t any governance on how the tech was applied,” she explained.

Maybury agreed. Solid foundations matter. They provide safeguards and reduce exposure. Yet governance cannot remain frozen in place. “It’s got to be governance that can scale,” he said. The statement carried weight coming from a company that processes sports bets around the clock. One wrong move in such an environment can trigger regulatory headaches or customer losses measured in millions.

Testing at Scale Becomes the New Standard

Anthony Moisant faces a different but related pressure at Indeed. The chief information and security officer oversees technology supporting 645 million job seekers and 3.5 million employers. Human review of every AI-assisted decision simply does not work at that volume. Moisant advocates relentless testing. Teams must compare AI outputs against intended results on an ongoing basis. The approach turns validation into a continuous process rather than a one-time gate.

Diya Jolly, chief product and technology officer at Xero, added nuance around decision types. Accounting software handles both routine calculations and situations that demand judgment. The distinction proves decisive. “If your outcome is deterministic, then you can probably let the agent go pretty far,” Jolly said. Outcomes in those cases lend themselves to straightforward measurement. When judgment enters the picture, however, removing humans grows far more difficult. Errors become harder to define and detect.

Executives from Prudential and Zillow joined similar discussions throughout the conference, according to Fortune’s reporting. Their presence highlighted how insurance and real estate also wrestle with AI in areas where mistakes carry heavy costs. A mispriced policy or flawed property valuation can cascade quickly.

And fresh signals reinforce the urgency. A recent Salesforce analysis from March 2026 found 63% of government leaders expect agentic AI to deliver greater impact than generative systems alone. The report pointed to mission-critical applications in social benefits, public safety, and defense. Those domains tolerate even less error than commercial betting or job matching. The pressure to get governance right intensifies.

Market data tells its own story. The AI governance sector, valued at roughly $430 million this year, is projected to reach over $4 billion by 2033, growing at 38.5% annually, per Persistence Market Research. Demand for frameworks, monitoring tools, and accountability mechanisms is rising fast. Companies that treat governance as an afterthought risk falling behind both competitors and regulators.

Leaders at the Aspen event repeatedly returned to first principles. Understand the desired result. Build policies that adapt. Test continuously. Accept that some decisions will always require human judgment while others can run with greater autonomy. The balance is delicate. Get it wrong and the downside arrives faster than ever. Get it right and the speed advantage becomes lasting.

Maybury’s closing observation captured the moment. Trillions of transactions already move through his systems. Agentic AI will multiply that activity many times over. Humans cannot keep pace. The only viable path lies in systems designed from the start with scalable controls, measurable outcomes, and clear accountability. Anything less invites trouble at a scale previously unimaginable.

So the executives left Aspen with notebooks full of ideas and a shared recognition. The age of AI agents has arrived. Managing them demands new thinking about responsibility, verification, and trust. Those who master the balance will shape the next decade of enterprise technology. Those who don’t may find their organizations overwhelmed by the very systems they deployed.