The V-Index Framework

The V-Index measures how vulnerable a profession is to AI automation by analyzing the work people actually do today.

Traditional automation frameworks ask whether an entire job will disappear. The V-Index instead evaluates how the individual tasks that make up today's professions are expected to change as AI capabilities improve.

The framework breaks every occupation into five primary current-day tasks, and evaluates each one independently.

The index covers a cross-section of U.S. jobs spanning every major sector, ranked among themselves by automation exposure.

Every task is scored on Task Verifiability (T1–T5) – how easy it is for AI to perform the task and objectively verify that it was done correctly.

Each task is also assigned an Automation Vector, describing how it is expected to evolve:

• Evaporative – the task largely disappears because AI performs it autonomously.
• Mutating – the task remains, but humans perform it differently alongside AI.
• Amplified – AI increases the importance of the human role.
• Suppressed – the task becomes less valuable or less common due to AI.

Because technical capability alone does not determine real-world automation, the framework also evaluates two environmental factors:

• Environmental Friction – how difficult the real-world environment makes automation.
• Liability Brake – how much legal, regulatory, or safety responsibility slows adoption.

These task-level assessments combine to produce the occupation's overall V-Index grade (V1–V5), which emerges from the individual assessments rather than being assigned independently. This gives a transparent explanation of why different professions are expected to experience different levels of automation.

Roles are ranked by overall automation vulnerability, most vulnerable first. The rank orders by V-Index grade, then by how verifiable the work is, then by how little environmental friction and legal or safety liability slow adoption.

The V-Index measures automation potential, not current market adoption. It evaluates what AI is fundamentally capable of automating based on the verifiability of the work, recognizing that real-world deployment may lag due to regulation, economics, organizational inertia, or customer preferences.

This index builds on an observation by Andrej Karpathy: in the current era of AI, verifiability is the strongest predictor of what gets automated. Work whose output can be checked cheaply and reliably can be practiced and optimized, so it automates first, while hard-to-verify work lags. The Task Verifiability scale applies that idea at the level of individual tasks. The two role-level factors, Environmental Friction and Liability Brake, along with the V-Index grade and the ranking, are extensions of that starting point rather than Karpathy's own work; this index is not affiliated with or endorsed by him. See his essay on verifiability at karpathy.bearblog.dev/verifiability.