Gartner pins under 1% of 2025's 1.4 million layoffs on AI productivity

Somebody has to open the queue and decide whether what the agent produced can ship. Tori Paulman, a VP analyst on Gartner's talent research team, said value in agentic workflows depends on redefining decision rights, supervisor roles and approval processes ^[2][18]. She also described the version that goes wrong. "Creating a human approval factory is an expensive bottleneck where humans spend time reviewing, escalating and correcting agents instead of doing higher-value work," Paulman said ^[7].

Gartner's layoff figure is easier to weigh in whole numbers. Less than one percent of 1.4 million comes out below 14,000 positions for the entire year ^[1]. TechTarget reports the figure as Paulman's account of Gartner's analysis and does not describe how cause was assigned to each layoff ^[4].

The University of Maryland's Robert H. Smith School of Business data runs the other way. AI-related roles went from 0.28 percent of all job postings in late 2022 to 1.13 percent at the end of 2025, about four times the share in three years ^[5][2]. Postings aimed at new graduates rose 0.9 percentage points over the same stretch ^[6][3]. If AI were absorbing entry-level work first, the new-graduate share is where it would show, and it went up.

Paulman said businesses overestimate the labor savings because they underestimate how much organizational redesign has to happen before the returns are durable ^[14]. That omission sits in the buyer's own estimate. Randall Hunt, CTO of the cloud native services provider Caylent, said the savings do not reliably arrive: "Surprisingly, costs are not always lower," and said that reaching human-level quality with AI on many tasks is possible but occasionally more expensive than just using a human ^[11][9][10]. "A year ago, everyone was focused on AI adoption," Hunt said. "Today, people are focused on leverage" ^[12].

Teams tell themselves users hand a task to the agent and stop doing it. Once it is live, the work looks like this, in Paulman's words: "To unlock real value, enterprises must decompose workflows, set decision rights, define supervisor roles, and make agents visible and auditable in systems of record. Not only is this not a labor-elimination shortcut, it's a work redesign cost burden that many organizations have not planned for" ^[13].

For anyone who owes finance a number, price the review lane before the savings line. Minutes a reviewer spends per item, times items per day, times the loaded hourly cost of the reviewer. Compare that to what the task cost when one person did all of it. If the first figure is not well under the second, the deployment has landed on the outcome Hunt described, where human-level quality costs more through the model than through the human ^[10].

The work sorts on two things: whether the output can be checked in less time than it takes to produce, and whether a named person has to sign it. The returns sit where checking is cheap and no signature is needed. Work that needs a signature and is slow to check stays in the queue, and reviewer hours cap how much of it clears.

Maruf Ahmed, CEO of the talent and technology consultancy Dexian, said the change is inside the roles: "In most cases, AI is changing the work inside roles far more often than it's eliminating the roles themselves" ^[15][16].