AI’s impact on human jobs is a hot topic of late, with all kinds of forecasting, reporting, and sounding of alarm bells.

But two new analyses seem to indicate that AI isn’t yet fundamentally disrupting the employment landscape, and that traditional metrics don’t accurately capture its actual impact on work.

According to a job cut announcement report from employment analyst firm Challenger, Gray & Christmas, so far in 2026, AI has displaced 12,304 jobs, representing just 8% of job cuts.

And researchers at Anthropic have introduced a whole new way to analyze AI’s impact on work, arguing that there’s still a huge gap between what large language models (LLMs) are capable of, and real-world deployment.

Using this methodology, they have determined that “AI is far from reaching its theoretical capability: Actual coverage remains a fraction of what’s feasible.”

AI-related cuts still barely in the double digits

Challenger, Gray & Christmas began tracking AI’s impact on jobs in 2023. Since then, the technology has been cited in 91,753 job cut announcements, or roughly 3% of all layoff plans. All told, AI was the stated reason for 54,836 layoffs in 2025, representing 5% of total cuts throughout the year.

More recently, AI was cited as the reason for 4,680 job cuts in February, roughly 10% of total cuts for the month.

It should be noted that there have been reverberations across the tech world as some companies slash human workforces in favor of AI. Most recently, payments and financial services company Block saw a 50% headcount reduction as CEO Jack Dorsey emphasized a shift to an “intelligence-native” model.

More alarmingly, the technology sector announced 11,039 job cuts in February. So far in 2026, big tech has lost 33,330 jobs. This represents an increase of more than 50% in the sector over the same period last year. However, they’re not all due to AI.

“Tech is responding to a number of pressures right now,” Challenger notes. While “AI is the big story,” there are also concerns around global regulations, digital advertising slowdowns due to tariffs and economic uncertainty, and overall higher costs.

Computer programmers, customer service reps most at risk

Meanwhile, Anthropic is measuring AI displacement risk with a new “observed exposure” technique, saying that the track record of past approaches is a “reason for humility.” This method combines the “theoretical” capabilities of large language models (LLMs) with real-world usage data, and weighs automated and work-related uses more heavily than what the company calls “augmentative” uses.

Based on this formula, there has been no “systematic increase” in unemployment for highly-exposed workers since late 2022, write Anthropic researchers Maxim Massenkoff and Peter McCrory. However, they note that there is “suggestive evidence” that the hiring of younger workers has slowed in some occupations.

Anthropic’s approach combines data from the O*NET database, which associates tasks with hundreds of jobs in the US, task-level exposure estimates around whether LLMs can perform a task at least twice as fast as a human, and usage data measured in the Anthropic Economic Index.

The researchers analyzed work-related usage in Claude traffic and adjusted for whether tasks performed by AI were carried out fully-autonomously or they augmented human workflows. They considered a job more exposed if associated tasks show “significant usage” in the Anthropic Economic Index, are performed in work-related contexts, and take up a larger share of an overall job role. They also factored in how often tasks for a given job showed up in automated use patterns or API implementations.

Based on this approach, some of the occupations most exposed to displacement by AI include:

• Computer programmers, who can use AI to do 75% of their jobs (this is in line with other data showing that Claude is being used “extensively” for coding).
• Customer service representatives, who can use AI for 70% of their work (the researchers point out that their main tasks are increasingly seen in first-party API traffic).
• Data entry keyers (67%)
• Market research analysts and marketing specialists (65%)
• Sales representatives in wholesale and manufacturing (excluding technical and scientific products) (63%)
• Software quality assurance analysts and testers (52%)
• Information security analysts (49%)
• Computer user support specialists (47%)

This approach brings many interesting points to the forefront, noted Jason Andersen, VP and principal analyst at Moor Insights & Strategy.

“Usage does not equal theoretical capabilities,” he said. “People are still figuring out AI’s capabilities and the risks it presents.” The methodology is interesting for two reasons: It reinforces what he and other analysts are seeing around tasks and roles, and it has a relatively simple rubric for making determinations.

“These are the types of methods that have staying power,” said Andersen.

Massenkoff and McCrory acknowledge that Anthropic’s approach won’t “capture every channel” where AI could reshape the labor market, but the method can help to “more reliably identify economic disruption than post-hoc analyses.”

“It is possible that the impacts of AI will be unmistakable,” they write. “This framework is most useful when the effects are ambiguous — and could help identify the most vulnerable jobs before displacement is visible.”

AI requires a significant industry-by-industry shift

In line with this research, Andersen said he is not seeing AI being deployed in a way that would eliminate entire roles. Some tasks are being re-engineered by AI, but they are largely still driven by humans. “Task-based automation has an incremental effect, making employees more productive and increasing their capacity.”

What’s missing from both of these analyses, however, is the bigger-picture impact on how work will be done in the age of AI. To best make use of the emerging technology, workflows and roles must change, and until that is resolved on an industry-by-industry basis, companies will be “kind of stuck where we are,” said Andersen.

This will disproportionately hurt younger workers seeking employment, potentially for some time, he noted. Meanwhile, existing employees may resist changes to workflows unless the changes are “significant and designed to reward experience and expertise.”

Right now, AI is seen as a way to offload work that would normally be handled by less-experienced resources, which Andersen sees as a problem. “We need to realign tasks and roles to balance this,” he said. The good news is that companies will be incentivized to do so, he noted, adding that workplace demographics in most first-world societies are changing as more white collar workers retire.

“It’s not sustainable for [companies] to have only high-cost, high-experience employees,” said Andersen. “They require a long-term balance.”