One in five US workers now delegates tasks to AI

The survey captures a significant inflection point in workplace AI adoption: not mass displacement, but task redistribution. The finding that 20 percent of US workers delegate at least one task to AI reflects widespread but selective use rather than wholesale automation. The variation in adoption rates—57 percent in software development versus 25 percent in record-keeping—suggests that AI's utility and workplace readiness remain task-dependent. The fact that workers accept AI output with minimal editing (66 percent unchanged or lightly tweaked) points to growing trust in AI competence for knowledge work, though the researchers caution this does not necessarily reflect objective output quality.

The timing savings data introduces a nuance: heavier AI use correlates with reported time savings (53 percent vs. 37 percent), but causality remains unclear—AI may genuinely speed work, or workers may use AI preferentially when seeking efficiency gains. The one-in-six tasks that take longer suggests friction: either the cost of prompting and review, or workers choosing to raise quality standards when AI saves time. Epoch AI's framing—AI as "versatile but usually not self-sufficient"—aligns with the evidence: full task automation remains rare, and human judgment remains the norm.