The Skills of Tomorrow: How AI-exposed is every skill in 2026?

Customer service representatives sit second from the top, which tracks with what Klarna learned the hard way. They replaced 700 CS agents with OpenAI in early 2024, walked it back in mid-2025 when CSAT collapsed, and most of those agents are now back on the payroll.

So exposure is not the same thing as substitution. The score tells you where AI is being used heavily for the task. It doesn't tell you whether the humans who used to do it still have jobs. For any role high on this chart, assume the shape of the work is shifting under them; don't assume the headcount is going anywhere.

The growth column is where this gets interesting. ML Engineer, AI Engineer, Applied Scientist, and AI Product Manager all roughly tripled their posting rate in the last 30 days compared to the 30 before. That's a real signal, even allowing for the fact that our dataset has been widening in the same window. AI Agent Engineer is the one to watch: it barely existed as a job title in early 2025 and is now posting at a pace close to Research Scientist.

These numbers are snapshot audits, not a live feed. We're wiring a live endpoint so the chart updates on every build. And if the numbers look old by the time you're reading them, they probably are.

Senior and staff together run at 59.1% of live postings. Entry-level sits at 13.3%. That's an inverted pyramid, not a talent funnel. If you're running graduate recruitment, the question isn't whether this matters, it's whether your 2027 intake has anywhere to land.

Two caveats to the story. One, our dataset skews toward technical and enterprise employers where juniors were always rarer. Two, "entry" is whatever the enrichment pipeline can infer from the job ad, some firms just don't tag seniority cleanly. We'll keep watching the gap as volume grows.

Read the chart as two axes of market signal, not one. The X axis is market demand (how often employers ask for it). The Y axis is AI intensity (how much Claude work is happening inside the jobs that ask for it). A skill in the top-right is in heavy demand and sits inside AI-heavy workflows. That's the sweet spot for anyone choosing what to learn next.

Some skills look misleading at first glance. Customer service lands at 0.00 exposure because it's a required-skills afterthought inside sales and account-management roles, not the job itself. Data analysis sits around 0.00, moderate, because it's diluted across sectors that all use Claude for different things. Skill-level scores always sit lower than occupation-level scores. That's the dilution effect, not a bug.

Bottom-left skills (low demand, low exposure) aren't obsolete, they're just niche. Bottom-right skills (high demand, low exposure) are the work that still looks like work: communication, project management, teamwork. That matches Anthropic's own finding that collaborative and stakeholder-facing tasks remain stubbornly human-shaped.

Top of the stack: legal at a $209k median. Bottom: manufacturing at $90k. The gap is wide, but more interesting is the width of the IQR bars, the sectors with the biggest AI premium (engineering, IT, finance) also have the fattest spreads, which means the top performers in those sectors are pulling away from the median, not just the floor.

Salary in our dataset is whatever the employer discloses in the ad, normalised to USD. A lot of listings still don't show a range, so these figures are a ceiling on transparency, not a ceiling on the market. Take them as directional, not definitive.