How many jobs are being eliminated by AI? We surely hear a lot of stories about how AI is making it harder for this or that group to find jobs, and plenty on how AI is going to have a major impact on employment down the line. OK, everyone has a view on the future, and most of them don’t pan out, but what about the present? If AI is stealing a lot of jobs, then we should see it in the project outcomes enterprises report. Do we?
I’ve been digging into enterprise use of AI agents for some time. This is the class of AI applications that the IT organization is most likely to be involved in, and so the class I get the most comment on. I’ve dug through what I’ve heard on the topic of job loss, and this is what I’ve found.
First, enterprises divide their agent applications into three groups. The first, the “reactive” group, accounts for roughly 45% of agent applications, the second “embedded” group includes roughly 31%, and the final “workflow” group, 24%. However, enterprises say that when all is said and done with AI, the majority of the new applications will fall in the last group, so they expect to see workflow AI accounting for two-thirds of all AI usage, reactive for a fifth, and embedded for the remaining roughly quarter.
None of these enterprises indicated that workflow AI agents had any impact on employment in their companies, but most believed that when fully deployed it might have a “ten percent” impact on jobs related to the applications, which were said to be only a quarter of all their jobs. Thus, they say workflow AI’s potential is to cut jobs by only around 2-3%.
Reactive AI projects and embedded AI are different. According to enterprises, both are justified largely by presumed labor efficiency gains, which can be monetized primarily by using fewer workers. Reactive AI includes self-hosted customized chatbot applications in sales and customer support, and these are justified in part by a reduction in call center personnel. However, few of these people are employees, according to enterprises. There is a job loss, but not for the company. Embedded AI, in contrast, almost always includes reductions in labor cost, with enterprises saying that successful projects include a reduction in labor cost of approximately 25% (but with wide variation—from 10% to nearly 50%).
The total AI impact on jobs? No enterprise said that they’d seen even a ten percent reduction in workforce. Only a tenth said that they saw any job category employment level decline by more than a fifth. Even in the longer term, nobody thought AI would reduce jobs by ten percent or more. This raises three obvious questions. First, why are we hearing so much about AI stealing jobs? Second, how is AI going to be such a revolution if there is no actual job impact? Third, what specifically works against AI as a reducer of jobs? Here, obviously, we’re going to have to take some guesses.
Enterprises in 2025 have pretty consistently told me they expected to reduce hiring, and a majority said they would target reducing workforce overall. Not because of AI, but because of economic uncertainty, new administration, tariffs, uncertain interest rate policies, you name it. All these are way ahead of AI as a reason to reduce jobs. Why then the AI stories? Two reasons, one really cynical and one a little bit so.
The really cynical reason was put this way by an enterprise: “Let’s see, we can tell Wall Street that the business outlook is uncertain, so we’re cutting jobs. Or we can tell them AI is letting us cut. The first makes our stock go down, and the second makes it go up.”
Of course, nobody wants to outright lie to Wall Street; they might be caught. And there is a possibility that the economic-reduced workforce might be able to pick up for the rest through the use of AI. So, while AI may not be (read “is not”) the real reason for the cuts, there’s a hope that it might let the company proceed along its way normally with a slightly smaller workforce. Hope, not necessarily expectation. Certainly, no enterprise suggested that they were sure they could respond to economic uncertainty with an AI-reduced workforce. My impression of their comments was that they were instead convinced that AI was only a PR play to explain the decision to reduce staff, as the comment said.
OK, then, are the expectations associated with AI and jobs totally unrealistic? Yes, absolutely. But is reality still good enough to make a strong AI business case? Let’s see how much labor cost might be on the table? In the US, labor compensation overall is roughly 59% of GDP, which is now roughly $30 trillion. That means total labor compensation of just under $18 trillion. If we take the impact of AI from enterprise comments, we find that they think AI will eventually impact labor that represents 45% of that, we get roughly $8 trillion in impact, and if their long-term AI impact was to reduce jobs by 8% as the suggested maximum, then we have a benefit of over six hundred billion dollars to justify AI. To put that in perspective, that’s about half the current total of IT spending.
To me, this is the core of the whole AI story. We don’t need AI to displace a lot of jobs. We really don’t even need to see AI meet the seemingly modest expectations of enterprises. If AI just hits these conservative targets, it would revolutionize IT spending, and IT, and vendors, and probably our whole relationship with technology.
Which brings us back to all these stories about AI and job loss. Ask the following: Why do stories get written? For someone to read. Therefore, the best stories are ones that a lot of people will read. That’s particularly true when you factor in the reason that stories are even available to read, which for tech and economic stories is that they generate clicks that serve ads. I’d bet many of you, my own readers, find this whole story a bit weighty. There’s probably some who wouldn’t even finish it. There’s next to no chance these details will ever be widely disseminated, however true they may be.
The sad thing is that the truth is good enough to make AI a success, a great success. And think about my numbers a minute. Labor that enterprises think will be impacted by AI doesn’t include much outside of office workers. Suppose we go after the 40% of workers who haven’t really been touched by IT empowerment yet. If we figure out how to get them involved, we raise the stakes considerably; they make up 55% of the compensation pie. Talk about IT revolution! Worth thinking about.
Want to see what AI does in analyzing this specific blog? I submitted it to Google’s Notebook LM, with no extra guidance or inputs, and this is an audio dialog it generated.
