They cut the staff. Did AI cost more?

AI could reshape the global labor market in several directions at once: replacing some workers, helping others do more, reducing entry-level hiring and creating jobs around the systems themselves. Higher productivity could mean a smaller team handling the same workload, or the same team taking on more business. Which outcome prevails will depend on demand, how well the automation works and local labor costs. An agent that undercuts an employee in a high-cost office may lose to a worker somewhere wages are lower. Before technical feasibility becomes a reason to cut jobs, the numbers still have to add up.

The pitch for replacing people with AI agents has run into an accounting problem. A salary disappears from the budget, but compute, integration, maintenance and quality control take its place. In theory, an autonomous agent can work around the clock. Its running costs never clock off either. Businesses are starting to ask whether they were too quick to let people go.

KPMG’s Global AI Pulse, based on a survey of 2,145 senior executives, found that 29% struggle to understand their growing AI operating costs. And 49% of companies had questioned, delayed or scaled back agent deployments when expected costs began to outweigh expected value. That is a sobering result for a technology sold as a way to slash labor costs: nearly half the companies surveyed had already reconsidered deployments because the economics looked doubtful.

Then comes the bill for bringing people back. Careerminds surveyed 600 HR professionals and found that two-thirds of employers that had cut jobs because of AI were already hiring people for the eliminated roles. The source has a commercial stake here: Careerminds provides outplacement services for people losing their jobs. Still, the pattern deserves attention. A company cuts its payroll, discovers there is work nobody is doing and returns to the hiring market. The apparent saving can come bundled with implementation costs, lost expertise and a fresh recruitment bill.

Klarna offers a particularly revealing public example. In 2024, the company said its AI assistant had handled 2.3 million conversations in its first month, covering two-thirds of customer-service chats and doing work equivalent to that of 700 full-time employees. It projected a $40 million improvement in profit for the year. By May 2025, CEO Sebastian Siemiatkowski was saying that an excessive focus on cost-cutting had hurt quality, and discussing new hiring in customer support. A chat handled is a convenient measure of success. Whether the customer’s problem was solved is a harder, more useful question.

“As cost unfortunately seems to have been a too predominant evaluation factor … what you end up having is lower quality.” — Sebastian Siemiatkowski, Klarna CEO

There is more tangible evidence for giving workers AI tools. A study published in the Quarterly Journal of Economics examined the introduction of an AI assistant to 5,172 customer-support workers. Issues resolved per hour rose by an average of 15%, with less experienced workers benefiting most. The AI offered guidance; the employee stayed involved in the work. That 15% is a productivity gain, and calculating net savings still requires accounting for the tool’s cost. But it is a measured result from people using AI on the job, rather than a promise about a future digital workforce.

Much of the cost difference comes down to who catches mistakes. With an AI assistant, a worker can spot a wrong turn, clarify the task or stop an unproductive attempt. An autonomous system needs those functions built in: checks, constraints, exception handling and ways to hand the task back to a person. Retries consume compute; investigating failures consumes specialists’ time. Comparing a subscription fee with a salary reveals very little about the cost of replacing someone. The useful figure is the total cost of delivering an acceptable result, including verification and every failed attempt along the way.

The next wave of automation deserves a straightforward question: why pay for autonomy where helping an employee would be cheaper and less socially disruptive? KPMG shows companies already rethinking agent deployments over expected costs. Business experience is also exposing the price of assuming too soon that people can be replaced. Before cutting jobs, companies should compare three options—human labor, a person using AI and an autonomous agent—on the same tasks, held to the same quality standard. Without that calculation, the layoff can easily become the most expensive part of the AI rollout.