AI can save time on individual tasks. Whether people actually regain that time is a separate question.
Time saved is real.
A randomized field experiment involving 7,137 knowledge workers across 66 firms gave selected employees access to an AI assistant integrated into familiar work tools. In the latter half of the six-month experiment, users spent roughly two fewer hours per week on email and reduced work outside normal hours. Researchers did not detect a broader change in the quantity or composition of tasks from individual access alone.
The distinction matters: a faster email process does not by itself redesign a company's operating model. The study concerns a particular technology, deployment and workforce, so its measured effects should not be generalized to every AI implementation.
Work can also expand.
An eight-month ethnographic study at a roughly 200-person technology company, reported by UC Berkeley researchers in 2026, observed employees taking on more tasks, switching among more work threads and extending work into previously quieter hours. Its findings are qualitative, context-specific observations, not an estimate of how often this occurs across employers.
Both results can be true. AI can make certain tasks faster while the organization absorbs the recovered capacity as additional demands.
Design for time well spent.
Tristan Harris and the Center for Humane Technology have long argued that digital design should respect human attention and deeper interests. In an AI-enabled workplace, that raises practical questions. Can a reporting step disappear? Can a technician get an answer without another dashboard? Can a manager approve an exception without supervising a dozen autonomous agents?
The relevant measure is not simply screen minutes. Some time on technology creates genuine value. The goal is to remove avoidable digital labor while preserving understanding, craftsmanship, good decisions and the freedom to disengage.
What to measure.
Before the intervention, record recurring task time, rework, interruptions, after-hours activity and employee experience. Afterward, measure whether the burden was removed, shifted to another person, or replaced with more work. Include quality and cost, as well as training and control of the resulting system.
For Cyber Agency, this is a design principle and a proposed evaluation method. It is not a claim that a particular deployment has delivered those benefits.
Sources
Dillon et al., Shifting Work Patterns with Generative AI, NBER WP
33795, revised November 2025 ↗
UC Berkeley Research, Does AI Actually Free Up Workers’ Time?, February
2026 ↗
Center for Humane Technology, Impact and Story ↗
Tristan Harris, How Better Tech Could Protect Us from Distraction (TED)
↗