CEOs Bet on AI Agents to Expand Payrolls, Not Shrink Them

Most chief executives of large U.S. companies expect artificial intelligence to help their companies hire more people over the next three years, not fewer.

This view emerges clearly from a fresh survey of 400 leaders at large U.S. firms. The poll, conducted by KPMG between August 11 and September 21, captured executives running businesses with at least $500 million in annual revenue. Their outlook challenges the widespread fear that smarter machines will hollow out the workforce.

Business Insider reported the results. Eighty-nine percent of those polled expect headcount to rise. The reasons tie directly to new business opportunities that AI creates. As AI frees up capacity, growth follows, and companies staff up to chase it.

Yet optimism comes mixed with realism. Only 9 percent of CEOs describe current AI agents as transformational for their organizations. About four in 10 report significant gains in efficiency or growth from these systems. A similar share sees moderate benefits in specific applications. Tim Walsh, KPMG U.S. CEO, put it plainly. “It’s not plug and play.”

Walsh spoke with Axios too. The path to AI-powered job growth is “directionally very clear,” he said. Still, serious risks remain. Cybersecurity threats top the list. Eighty-five percent of CEOs expect cyber issues to hit their prosperity in the coming years. AI-powered attacks only sharpen that anxiety.

Executives aren’t ignoring the human side. Eighty-six percent plan to redesign roles, career paths and team structures for human-AI collaboration in the next two to five years. Seventy-nine percent are hiring talent with AI and technology skills. Eighty-three percent focus on retaining and retraining high-potential workers. These figures have climbed from prior years. Skills gaps, training demands and talent attraction rank among the biggest workforce headaches.

AI has already prompted changes in how companies develop employees. Roughly 80 percent say so. Three-quarters have revisited entry-level skill requirements. The technology is forcing a rethink of what workers need to succeed.

But job growth isn’t uniform. Fifty-three percent of CEOs still plan workforce reductions in certain areas. The composition of teams is shifting. Some roles fade while others expand or emerge entirely, so net hiring may rise even as automation claims specific tasks.

Spending shows no sign of slowing. Eighty-four percent list AI as a top investment priority for the next three years. Leaders recognize that AI will alter business models. Four in five say as much. Returns, however, lag the outlays. Walsh noted this pattern in a Marketplace interview. ROI trails the investment period. Transformation requires more than new tools. It demands new ways of operating.

Recent reports reinforce the tension. A Deloitte study found only one in five organizations feels prepared for autonomous AI agents. Three-quarters expect half their business processes redesigned around such systems within four years. Human oversight remains central. Seventy-five percent see greater value in human-agent collaboration than pure automation. HR Dive covered those findings in August.

Accenture data points to another mismatch. Agentic AI exceeds expectations on productivity and employee satisfaction. More than two-thirds of C-suite leaders report better-than-expected productivity gains. Yet fewer companies capture sustained, board-reportable business value from AI investments. The share fell to 23 percent in Accenture’s July survey, from 32 percent earlier in the year. CIO Dive detailed the gap in August.

Governance questions grow louder. An IBM survey of EMEA executives revealed that 80 percent saw an AI-related incident reach the board or C-suite in the past year. Only 40 percent changed governance policies in response. Seventy-five percent expect AI agents to hold some decision autonomy by 2027. Just 37 percent believe their current frameworks can scale accordingly. IBM released the study on October 14.

Enterprise architects face their own struggles. A SAP LeanIX survey showed 98 percent of companies deploying or planning AI agents. Yet 48 percent have no one clearly responsible for governing them. Only 17 percent can fully track agent performance. Workmate reported those numbers earlier this month.

UK CFOs tell a different story from their American peers. More than half expect AI to cut headcount in the coming year. Adoption remains early there too. Two-thirds say they’ve achieved just a quarter of the changes they ultimately anticipate. Bloomberg highlighted the contrast on October 13.

Accountability questions surface as agents gain independence. A Wharton School study found managers split on who bears blame when an AI system errs. Roughly equal shares point to the deploying employee, the manager, or the company’s risk team. Few blame the vendor. Uncertainty reigns. Bloomberg covered the research.

Oyster’s global AI talent report adds another layer. Ninety-five percent of companies have employees performing AI work outside their official job descriptions. Two-thirds have added AI skills to non-technical roles. Many do so without updating titles, pay, or even notifying staff. Training existing workers beats external hiring for most firms. Morningstar summarized the October 14 release.

Microsoft’s internal metrics hint at scale. The company reported 15-fold growth in active AI agents inside its 365 suite. Large enterprises saw 18 times more. These numbers suggest agents could multiply faster than humans. Yet LinkedIn data shows 1.3 million new AI-related jobs created in two years. Demand for engineers, annotators and related skills keeps climbing. Data Society explored this dynamic on October 15.

Oliver Wyman’s parallel survey of CEOs and CIOs offers a tempered productivity forecast. Sixty-four percent of CEOs expect productivity gains of 11 to 20 percent from agentic AI over the next two to three years, yet almost half of CIOs and CTOs say current IT funding can deliver only a 1 to 10 percent gain. Ambition stays grounded. Governance, talent and operating models will decide who captures the upside. The firm’s September report laid out those expectations.

Walsh still wants people checking the agents’ work. “I need the humans to review this stuff,” he told Business Insider. “We can’t just rely on these agents. They’re not always right.”