Career Sentinel
Transparency by design
Your Career Sentinel number is an estimate of AI exposure — how much of the work in your role can be done or accelerated by today’s AI. It is computed deterministically from open, peer-reviewed labor research. It is not an LLM guessing, and it is not a prediction that you will lose your job. Below is the entire method: the inputs, the math, why the score drops as you act, and the limits we hold ourselves to.
Every input is free, public, and peer-reviewed or government-published. No proprietary black box. Here is each source and exactly what it contributes:
The backbone. ~1,000 occupations broken into task statements, Detailed Work Activities, 52 abilities, and work-context factors. This is what lets us score your job at the task level instead of slapping one number on an entire title — and the importance ratings become each task’s weight.
Task-level LLM exposure labels — E0 / E1 / E2 (no exposure / direct / LLM-plus-tooling). These tell us, task by task, how automatable the work is. The E1 and E2 labels also define the lower and upper bounds of the range we show you, so we never report a fake-precise single percentage.
The AI Occupational Exposure index: an occupation-level prior mapping 10 AI applications onto O*NET’s 52 abilities. We blend it with the task math to stabilize the estimate so a single noisy task can’t swing your whole score.
Real-world behavior: how people actually use AI per task, split into automation vs. augmentation. This is the calibrator — it grounds the theoretical exposure in what AI is observed doing, and informs whether a task is the kind AI replaces or the kind it amplifies (the part you can lean into).
The Occupational Employment and Wage Statistics: wages and employment by occupation and geography (SOC codes). Powers the pay benchmarks, the “you may be underpaid by ~$X” signals, and the reward side of any role comparison.
Webb (2020) validates patent-task overlap (a defensibility cross-check), and macro reports (PwC, WEF, McKinsey) supply the priors and the evidence base for the mitigator — including PwC’s finding that AI-skilled workers command a +56% wage premium.
For your occupation, your set of tasks, and your self-reported time on each, the engine runs a fixed sequence. Same inputs always produce the same number — that’s what “deterministic and auditable” means.
For each task: taskWeight × taskAutomatability, summed and normalized.
Automatability comes from Eloundou as β = E1 + 0.5·E2; taskWeight is
O*NET importance × your time-share on that task. Your day, not a generic title.
0.7 × exposure + 0.3 × AIOE_prior. Anchoring 30% to the occupation-level
prior keeps a single odd task from distorting the whole estimate.
× (1 − λ·R) where R is the physical / social / creative /
judgment content of your role from O*NET (the classic “bottlenecks” AI struggles with),
and λ ≈ 0.4. More human-bottleneck work → lower exposure.
The result becomes your RawScore on a 0–100 exposure scale — always shown as a range with a confidence band (e.g. “62–74, Moderate–High”), driven by the Eloundou lower/upper bounds. Never “73.4%.”
Score = RawScore × (1 − M). M is everything you’ve actually
done to use AI — capped at 0.35. The score drops as you build skills.
The cap is honest: you can meaningfully lower your exposure through action, but you can’t
fully opt out of structural change.
This is the heart of Career Sentinel: the number isn’t a verdict you’re stuck with — it’s a gauge that responds to real AI skill-building. The evidence is clear that AI-fluent workers grow faster and earn more (PwC’s +56% premium; augmentable roles expanding faster than automatable ones). So each genuine action lowers your exposure, each with a cited rationale shown in-app, capped at 0.35 total:
We never show one bare number. You always see two: a “do nothing” scenario (a drift toward Eloundou’s upper bound over ~24 months, explicitly labeled a scenario, not a forecast) and a “with your plan” number that falls as you build skills. The gap between them is the whole point — it’s a map of what AI fluency is worth to you.
Disclaimer. AI exposure is not a forecast of certain displacement. Scores reflect technical capability and current research — not your employer’s decisions — and update as the underlying data updates. Career Sentinel provides general information and is not career, financial, or employment advice.