Is AI Replacing Low-Wage Jobs? What the Usage Data Shows

Is AI replacing low wage jobs? The evidence points the other way. Anthropic's own usage data, built from real conversations rather than a guess about a job title, shows AI exposure concentrated in higher paid, more educated office work, while a large group of mostly physical, in person jobs barely register in the data at all.

That is worth sitting with, because the popular fear runs the opposite direction: that AI arrives first for cashiers, cooks, and drivers, and works its way up. The most direct evidence available, a March 2026 report from Anthropic researchers Maxim Massenkoff and Peter McCrory, says the concentration runs toward the desk, not away from it.

What "zero coverage" actually measures

The report is built differently from most exposure research. Instead of asking a model how automatable a job title sounds, the authors matched real Claude conversations to hundreds of occupations from the O*NET database, then checked which tasks showed up in that usage often enough to count. An occupation only counts as covered once its tasks cross that minimum threshold. This is an occupation's real usage in a live tool, not a rating of how automatable its tasks sound. Our own explainer on how job risk scores get built covers the older, guess based approach this method was built to improve on.

The jobs the data barely touches

Roughly 30 percent of workers fall into a zero coverage group, where the tasks that make up their day appeared too rarely in Claude usage to clear the threshold. The named examples include cooks, motorcycle mechanics, lifeguards, bartenders, dishwashers, and dressing room attendants. None of that work routes through a screen or a document. It happens with your hands, in a room, in real time, in front of a person who needs the result right now. A chat based tool has no way into a shift like that yet.

Where the exposure actually concentrates

The contrast sits in who is on the other side of that line. Workers in the most exposed occupations earn 47 percent more on average than workers in the zero coverage group, and graduate degree holders make up 17.4 percent of the most exposed group against just 4.5 percent of the unexposed one, an almost fourfold gap. The most exposed workers also skew older and more likely to be women. The fear that AI would eliminate low wage service work first is not what this particular dataset shows. The exposure sits with the people already paid the most to sit at a keyboard.

The one place the picture gets harder

The honest caveat is about who is trying to get in, not who already has the job. The same report finds a 14 percent drop in the job finding rate for workers aged 22 to 25 in the most exposed occupations, a result the authors flag as only barely significant. It does not show up for workers over 25. Read plainly, that looks like a hiring headwind for new entrants into exposed fields rather than a wave of layoffs among people already working in them, and it is worth naming rather than leaving out.

What this changes if your job is a desk job

If your work is the kind that shows up in this data, an office job built from writing, analysis, or information handling, the honest read is that AI is already part of the texture of that work, not a rumor about someone else's job. That does not make the job replaceable on its own. It makes the differentiator the layer above the routine text tasks a tool already handles: judgement calls, holding a client relationship, catching what a first draft gets wrong. Our checklist for reading your own week for exposure and our rundown of the human skills that resist automation both start from that same layer. If you want a read built from your own answers instead of an occupation average, the check scores six of those domains directly.