# Observed Exposure — Anthropic's AI Penetration Metric > Anthropic's "observed exposure" combines what AI *could theoretically do* with what it's *actually doing* in the real world today. The gap between the two numbers is the real story. --- ## The Numbers | Occupation | Theoretical coverage | Observed coverage | |---|---|---| | Computer & math | 94% | 33% | | Office & admin | 90% | 14% | | Business & finance | 87% | 18% | | Computer programmers | ~90% | 75% | | Customer service reps | ~80% | 70% | | Data entry keyers | ~85% | 67% | | Education, healthcare, construction | varies | single digits | Computer programmers are the single most exposed occupation (75% observed), followed by customer service reps (70%) and data entry keyers (67%). --- ## Why the Gap Exists The frontier is jagged. AI is shockingly good at some tasks and poor at adjacent ones, and the boundary is unpredictable and moving fast. Alvaro Higes (Luzia) calls this the *jagged frontier* — you cannot predict where capability ends. High theoretical coverage with low observed penetration points to an adoption and product problem, not a capability problem. The work is there to be done; the workflows haven't been built yet. --- ## The Leading Indicator Hiring for 22–25 year olds in exposed occupations has slowed 14%. Entry points to the labour market narrow before unemployment data shows any signal — the jobs dry up before incumbents lose theirs. --- **Related:** [[AI Eats Services Not Software]] | [[The Gap of Imagination]] | [[AI Disruption Risk Is Not Uniform - Thoma Bravo]] Tags: #AIstrategy #labor #data #Anthropic