You have surely been told that AI is going to wipe out whole swaths of jobs. The reality, measured on actual usage data, is more nuanced. Where AI models could in theory handle 94% of the tasks of computer and math occupations, today they actually cover only 33% of them (Anthropic, 2026).

So it isn’t AI that replaces people. It is the people who master AI who, little by little, replace those who don’t. The threat is real, but it is misnamed: it doesn’t weigh on whole occupations, it weighs on the gap in mastery within each occupation.

The good news? That gap can be closed.

Which jobs are threatened by AI? What the Anthropic Economic Index shows

In March 2026, Anthropic published a study that changes the way the question is asked (Anthropic, “Labor market impacts of AI”, 2026). Rather than estimating what AI could do in theory, it measures what AI actually automates. The method combines three sources: the US O*NET database (a public reference that lists the tasks of about 800 occupations), the theoretical capabilities of the models, and the observed real-world use of the Claude assistant (the Anthropic Economic Index).

The main finding is a yawning gap between what is possible and what actually happens. Computer programming is the most exposed occupation; at the other end, physical or manual occupations (cooks, mechanics, lifeguards, bartenders) remain almost out of AI’s reach. But here is the point that contradicts the doom-laden talk: no significant rise in unemployment has been detected among the most exposed workers since late 2022. Better still: these exposed occupations pay on average 47% more than the least exposed ones (Anthropic, 2026). At this stage, being exposed to AI doesn’t mean being replaced.

Does that mean declaring victory? No, because this gap between potential and practice is anything but permanent.

Why AI doesn’t replace a whole job: the role of task automation

The gap between AI’s theoretical potential and observed use at work

AI doesn’t swallow a job whole: it nibbles at tasks. And a job is a combination of tasks, trade-offs, relationships and judgment. When AI takes on a first draft or a data extraction, it doesn’t eliminate the position: it shifts the value toward what the machine can’t do, which is deciding, putting things in context and taking responsibility. That is precisely why the theoretical potential of 94% translates into only 33% of actual use: in between lies the real work, with its legal constraints, its human checks and its complexity (Anthropic, 2026).

The AI skills gap and its impact on hiring juniors

Two signals reveal it. First, the hiring of 22- to 25-year-olds in exposed occupations has fallen by about 14% since generative AI arrived: companies hesitate to hire juniors for tasks that can now be automated (Anthropic, 2026). Second, the US Bureau of Labor Statistics projects weaker employment growth through 2034 for the occupations most exposed to AI.

In other words: some occupations will hire less, and some ways in are closing. But it isn’t the jobs that are disappearing, it is the advantage that is shifting toward those who know how to work with AI. Within the same team, two people in the same position will soon no longer bring the same value, depending on whether or not they master these tools.

How do you close the AI skills gap in your team?

For now, then, the useful question is no longer “will my job disappear?”, but “do my teams know how to turn AI into a concrete advantage?”. And on the ground, the answer is clear: it isn’t the tools that are missing, it is their anchoring in real work.

Start from frontline pain points, not from tools

Rolling out an AI license across the whole company doesn’t create mastery. What creates it is starting from the time-consuming tasks your teams know better than anyone, and turning them into concrete use cases. The best automation ideas come from the people who do the work, not from a plan decided at the top.

To go further on how to structure this adoption, read our guide AI at work: the complete guide to a successful rollout.

Measure real gains, not the number of licenses

Mastery can’t be decreed, it is measured. Not in licenses bought, but in real uses and time saved, quarter after quarter. It is this discipline (starting from the front line, turning expertise into skills, tracking the impact) that separates the teams that take the advantage from the teams that suffer it.

Conclusion

Yes, AI is going to reshuffle the cards: some occupations will hire less, some junior tasks will be automated, and the gap between teams will widen. But the data doesn’t point to mass replacement. What it points to is value shifting toward the people who know how to use AI.

Note that the gap between the possible and the actual isn’t fixed. Anthropic’s researchers say so themselves: as AI spreads through companies, real use will catch up with the theoretical potential. And the day it does, the impact on the labor market will become far more visible, and so will the lead of those who already master AI.

The real question isn’t whether AI will replace your employees. It is which of them will take the lead before the others.

FAQ: your questions about AI’s impact on jobs

Will AI replace jobs?

Real usage data shows no mass replacement. According to Anthropic (2026), no significant rise in unemployment has been detected among the most exposed workers since late 2022, and these exposed occupations pay on average 47% more than the least exposed ones. AI doesn’t replace a whole job: it takes over tasks, and the advantage shifts toward those who know how to work with it.

Which jobs are most threatened by AI?

Computer programming is the most exposed occupation, while physical or manual occupations (cooks, mechanics, lifeguards, bartenders) remain almost out of AI’s reach (Anthropic, 2026). But theoretical exposure isn’t actual use: in computer and math occupations, AI could handle 94% of the tasks, and today it actually covers only 33% of them.

Is AI reducing the hiring of juniors?

It is a first signal: the hiring of 22- to 25-year-olds in exposed occupations has fallen by about 14% since generative AI arrived, as companies hesitate to hire juniors for tasks that can now be automated (Anthropic, 2026). The US Bureau of Labor Statistics also projects weaker employment growth through 2034 for the most exposed occupations.

How do you close the AI skills gap in a team?

By starting from the time-consuming tasks the teams know better than anyone and turning them into concrete use cases, rather than rolling out a license across the whole company. Then by measuring the real gains, in uses and time saved quarter after quarter, rather than the number of licenses bought.