AI Skills & Talent Trends
Every week brings a new number for how much of your work AI will take. Most are two to three times too high. This measures what AI can actually take from a role — once the work that’s physical, regulated, or built on trust is set aside — across twelve industries.
An AI-exposure report built on the AI Task Exposure (AITX). Shared in a briefing, not sold as a download.
That’s the raw pool the headlines start from — every US wage dollar in tasks AI could theoretically touch. Most of it is protected by physical, regulated or trust-bound work. This report measures what’s genuinely exposable, industry by industry — usually far less.
- How AI exposure is really measured — jobs broken into tasks, scored, minus the work each industry protects
- The twelve industries analysed, and the one pattern that holds across all of them
- The four reasons AI stops at the door — physical work, regulation, legal accountability, and trust
- The specific skills AI can take on in each industry, and what the freed-up time is worth
- What to treat as solid, and what is still an estimate
The familiar “AI will automate X% of jobs” figures start from raw capability — everything today’s models could conceivably do. But no industry is only its automatable tasks. A winery has a physical core; a law firm runs on privilege and signature; a brokerage runs on trust. Set that protected work aside and the genuinely exposable share is far smaller — and far more useful to plan around. This report scores it industry by industry: which of the top roles’ tasks AI can really take, the skills that go with them, and what the freed-up time is worth. Twelve industries, one method, one clear pattern — and the numbers walked through in a briefing.
A note on rigor: exposure is freed task time valued at wage — redeployable to relationship, judgment and new work, never a headcount forecast. Figures carry a stated uncertainty band; nothing is described as proven, and pricing is discussed on the call.
Not the headline number.
AI Task Exposure (AITX) breaks every job into the underlying tasks it’s really made of, scores how much of each task today’s AI can actually do, then sets aside the tasks the industry’s constraints protect. What’s left is the real exposure — and the specific skills that come with it.
Start from raw capability and you overstate every industry two- or threefold. The chart shows the shape — what survives once the protected work is set aside — measured for each industry, not asserted.
Four constraints, twelve industries.
Every industry gives back its open work to one modal constraint. Each has its own brief — request the one that's yours.
Physically gated
A physical core the model never touches — the bundle buys back the scheduling, records and desk work around it.
Person as firm
The solo practice as a ten-hat micro-firm — the week is mostly back-office.
Relational core
Research, drafting and reporting are open; discovery, negotiation and trust are not.
Licence-fenced
Held by signature, disclosure and privilege — not by touching a thing.
830 SOCs × 16,600 tasks → industry staffing → top-ten roles → constrained exposure → skill packages → freed-time model
Don't see your industry? The instrument runs against any of them — request a run.
In collaboration with the Barry Hurd Intelligence Lab.
Back to the lab