How we date a role's AI clock

Our working assumption is that AI is compressing roughly a century of scientific, economic, and technological change into five to ten years. That pace does not hit every industry at once: it moves in phases, arriving first where work is digital, text-based, and easy to verify, and later where it is physical, regulated, or relationship-bound.

A "clock" is our estimate of when AI materially reshapes a role's core task mix — not when the job title disappears. Roles rarely vanish on a schedule. What changes first is what the role spends its hours on, how many people are needed to do it, and what skills the market pays for within it.

1. Task-mix exposure. We decompose a role into its recurring task types and ask which are structurally exposed: pattern-bound work, text-in/text-out work, and work whose output can be cheaply verified automates first. Work that carries legal accountability, physical presence, relationships, or judgment under ambiguity holds out longest.

2. Capability thresholds. For each exposed task type we ask whether current AI systems can already do it at a usable quality, and if not, what specific capability would change that. A clock tightens when the blocking capability is already visible in frontier systems.

3. Industry adoption phase. Capability is not adoption. Each industry sits somewhere on an adoption curve shaped by regulation, margin pressure, incumbent software, and culture. The same task automates years apart in adtech and in a public-sector agency.

4. Economic pressure. We ask who profits from automating the task and how strong that incentive is. High-volume, high-cost, low-switching-cost work moves first.

Public data anchors.Where an official statistic exists for the closest occupation, the page shows it in a "numbers behind this page" box: US employment and the official 2025–35 projection, median pay, and the relative AI exposure category published by the U.S. Bureau of Labor Statistics in 2026, plus an independent AI applicability score from Microsoft Research. Those figures are copied from their sources and dated; they anchor the analysis but do not set the clock, because the official projections do not model AI adoption speed. Exposure, in the statisticians' own words, does not imply job loss.

The free pages under /jobs and /industries apply this framework at the category level: the average task mix of a role, in the average company, at the industry's typical adoption phase. They are honest averages — and averages hide a lot. A senior specialist with rare context can sit years behind the category clock; a junior generalist doing the role's most exposed slice can sit years ahead of it.

The personalized reports run the same framework on your actual inputs — your day-to-day tasks, skills, seniority, company size, and market — which is what moves a verdict from "the average for your title" to "your position specifically."

These are structured estimates, not measurements. Adoption can lurch with one regulatory decision or one capability jump, in either direction. We publish month ranges rather than dates, we update the corpus quarterly so every page reflects the current state of the transition, and we treat every clock as falsifiable: when the ground moves, the pages move.

Nothing on this site is legal, financial, or medical advice. It is strategic analysis — input to your own decisions, not a substitute for them.