How We Score Your Role
Every score is built from real task data — not job title guesswork.
Why task-level scoring matters
Most AI risk tools score entire job titles — "Software Developer: high risk" or "Nurse: low risk." That misses the point. Two people with the same title can have wildly different AI exposure depending on what they actually do every day. We score at the task level.
For each occupation, we analyze every core task from the U.S. Department of Labor's O*NET database (~19,000 tasks across 923 occupations). Each task is scored independently on six dimensions, then rolled up into an overall Shield Score.
How each task is evaluated
Each task is evaluated across a proprietary set of dimensions that determine whether AI is likely to perform that task independently, augment a human performing it, or require uniquely human capabilities.
Exposure factors — things like how structured and digital the task is — increase automation likelihood. Protection factors — things like the degree of judgment, physical presence, or relationship trust required — reduce it.
Our model draws on published research, observed AI capability data, and real-world task performance patterns to calibrate these evaluations. Scores are reviewed and updated quarterly as AI capabilities evolve.
Task classification
Based on those six dimensions, each task is placed into one of three categories:
🔴 Automatable
AI can likely perform this task end-to-end within 2–3 years. The action: learn to supervise and manage the AI doing it.
🟡 AI-Assisted
AI will make humans faster and better, but human oversight remains essential. The action: learn the tools that accelerate this task.
🟢 Human-Led
This task requires uniquely human capabilities for the foreseeable future. The action: invest more time here — this is your competitive edge.
The Shield Score
Task classifications are combined into an overall Shield Score (0–100). Human-led tasks contribute most to your score; automatable tasks contribute least. AI-assisted tasks fall in between — they represent work where AI can reduce headcount even when humans remain involved.
The scoring model is calibrated so that a role where most tasks remain human-led scores meaningfully higher than one where AI can handle the majority of work end-to-end. The goal is to reflect real employment risk, not theoretical capability.
Employment outlook adjustment
We apply a small adjustment (±6 points maximum) based on Bureau of Labor Statistics 10-year employment projections for each occupation. A role projected to grow significantly gets a small upward adjustment. A declining role gets a downward adjustment. This reflects the reality that even a high-scoring role in a shrinking field requires more planning than a high-scoring role in a growing one.
Score tiers
Data sources
923 occupations and ~19,000 task statements from the U.S. Department of Labor. License: CC BY 4.0.
Real-world AI usage patterns showing observed exposure rates — not projections, but actual usage data from Anthropic's models. License: CC BY.
10-year employment outlook by occupation from the U.S. Bureau of Labor Statistics. Public domain.
How often scores update
Scores are refreshed quarterly as AI capabilities evolve. The current scores reflect AI capabilities as of Q1 2026. The next update is scheduled for Q3 2026.
AI capability is moving fast. A role that scores 72 today may score differently in 12 months — which is why we provide quarterly refreshes for Career Command subscribers.