Skip to main content Scroll Top

Proofseconds: How Verification Behavior Decays in AI-Assisted Knowledge Work (SSRN Preprint)

Knowledge workers using AI assistants check less over time. The question nobody has answered: does that drop mean the system got better, or the human stopped checking? Both trajectories look identical in the data.

This preprint introduces proofseconds — the elapsed time a worker spends verifying an AI output before acting on it — as the construct that tells them apart. Two decay patterns. Eight testable propositions. One uncomfortable finding: your highest-expertise people collapse first, in a single interaction.

Read the full paper on SSRN.