1 in 5 US Workers Delegate Tasks to AI Over Colleagues

What happened

A new survey from Epoch AI finds that one in five employed Americans now hands off at least one work task to an AI tool rather than to a human colleague. According to the research, these workers typically accept the AI's output with little or no editing before it feeds into their work.

The finding points to AI tools moving beyond drafting assistance or idea generation into a role previously filled by coworkers — receiving delegated work and returning a finished output that is used largely as-is.

Why it matters

This marks a shift in how work actually gets distributed inside organisations, not just how individuals use AI as a productivity aid. When a fifth of a workforce is routing tasks to a machine instead of a teammate, it changes the shape of collaboration, informal mentoring, quality checks and the accountability chain that normally accompanies delegated work between people.

For leaders in digital transformation and workforce design, the signal is less about AI capability and more about behaviour: employees are already restructuring their own workflows around AI, often ahead of formal policy, training or oversight. That has implications for quality control, skills development for junior staff who might otherwise absorb tacit knowledge through being delegated tasks, and for how organisations define "good enough" output when review is minimal.

By the numbers

• 20% of employed Americans surveyed say they delegate at least one work task to AI instead of a human colleague

The Renascence take

The headline number is really a story about trust thresholds, not technology adoption. People aren't just using AI — they're substituting it for a social and accountability relationship, and doing so with minimal editing, which suggests confidence is outpacing verification.

The interesting shift here isn't that workers are using AI — it's that they're skipping the review step they'd normally apply to a colleague's work. Delegating to a person carries implicit accountability and a feedback loop; delegating to AI with little editing removes both. Organisations should treat this as a service-design problem, not just a productivity one: define where light-touch AI output is acceptable, where it isn't, and make sure someone is still responsible for the outcome, not just the output.