There is a particular kind of vertigo that comes from reading about AI and your own work. One headline says your job is finished. The next says nothing will change. Both are confident; neither shows you the number it is built on. This magazine exists to do the unglamorous opposite: to put the actual, measured figure in front of you, name the dataset it came from, and tell you — plainly — what it does and does not mean.
Start with the number that breaks the most intuitions.
This is the whole thesis of the site in one line: exposure is not a verdict. “AI task overlap” measures how much of the work today’s models can attempt — draft, summarize, classify, suggest. It says nothing, by itself, about whether a job disappears, whether wages move, or whether anyone adopts the tool. Those are separate questions with separate data, and we keep them separate on purpose.
The shape of it
Once you stop reading exposure as doom, a real shape appears underneath the noise.
That is the quiet finding the panic misses. The work most reachable by today’s AI is the desk-bound, language-heavy, screen-mediated work — and a lot of it is well-paid, white-collar, and historically considered safe. The work least reachable is physical, situated, and embodied. A decade ago the forecasts pointed the other way.
What’s new in the map
This is a living instrument, and it changed a lot this period. The build log is public and honest — here is what actually shipped.
What's new in the map
211 changes shipped across 5 active days — most recent 2026-06-05.
- Feature P4 S4 honest AEI-usage viz on geography pages 2026-06-05
- Feature P4 S1-S3 exposure quadrant on viz-less entity pages 2026-06-05
- Feature P3 grounded example prompts from measured delegated tasks 2026-06-05
- Feature P2 S4 — context pack for occupation-groups + industries 2026-06-05
- Feature P2 — per-occupation AI context pack (lib + route + affordance + llms) 2026-06-05
- Feature P1 S3 — macro exposure quadrants on /ai-exposure + /outlook 2026-06-05
Read deeper
The instrument Where every kind of work sits on AI exposureTen honest lenses on the same question — by occupation, by observed AI use, by growth, by pay, by remote-work feasibility, by industry. Each one states what it measures and what it refuses to claim.
Read the analysis → Your coordinate Find your own point on the gradientType one occupation and get a one-page, source-backed brief: where the work sits, where AI already shows up in observed use, and where humans are still needed. Task overlap, not a prediction.
Read the analysis →What to watch
What to watch
- Whether the high-overlap, high-demand occupations (like customer service) start to show movement in BLS openings — the first place adoption would actually surface. Outlook hub →
- The gap between what AI can attempt (potential exposure) and where it's actually being used (observed). Today they disagree most in regulated, legal, and judgment-heavy work. Divergence lens →
- The tacit-knowledge premium: roles whose value lives in unwritten judgment, where overlap stays low not because the tasks are simple but because they resist being written down at all. The report →
The map will keep changing as the data does. What won’t change is the rule this whole thing is built on: every number names its source, every figure carries its caveat, and the map is never sold to you as the verdict. That’s the job of the reader — and now you have the instrument to do it.