The weekly paper
for curious people
See the connections.
Understand the consequences.
Swipe to see more sections →
Work · Issue 010
Workplace AI can raise task productivity without automatically raising pay or reducing hours.
Productivity rose in one setting; broad worker gains were not established
In a customer-support study, generative AI raised issues resolved per hour by 15% on average. In Denmark, workers reported that most saved time went into other work tasks, while administrative records showed no measurable average change in earnings or recorded hours through 2024. Faster work is real in some tasks; who benefits is a separate question.
If generative AI makes a worker faster, the worker gets time back—or the employer gets a matching productivity windfall.
The question: Where did the saved time actually go?
Choose how to read this
The facts do not change. Each view uses the same published sources and leaves the same questions open.
See through another Lens
Each view notices something useful. None is allowed to stand in for the complete evidence.
Worker
If the tool saves time, I may reasonably ask whether workload, hours or pay changes.
What this view explainsWhy productivity alone is not a complete worker outcome.
What it may missA worker can value help even without a recorded hour reduction.
Two studies, two questions
The customer-support study followed 5,172 agents. Access to a generative-AI assistant increased issues resolved per hour by 15% on average, with the largest gains among less experienced and lower-skilled workers.
That is a result in one operational setting. The researchers could not observe wages or overall labour demand, so the productivity number cannot tell us who received the value.
The Danish research combined late-2024 surveys of about 25,000 workers with administrative records. It found no measurable average effect on earnings or recorded hours and could rule out average effects larger than about 2% after two years.
In the survey appendix, about 80% of saved time was redirected to other work tasks. Fewer than 10% reported more breaks or leisure, about a quarter spent more time on the same tasks, and only 3–7% reported higher earnings. These are reported allocations, not a time-and-motion audit.
Both studies show uneven effects. A tool can help a novice more than an expert, change some occupations before others, or turn saved time into higher output without changing scheduled hours.
The evidence does not justify saying employers captured every gain. It does justify asking for records that connect task time, output, hours and pay rather than treating 'productivity' as the answer.
Possible effects · We cannot say how likely
The bargaining question begins after the stopwatch stops. Saved time may become more output, different tasks or smaller teams. Job design and staffing records—not a demonstration of the tool—will show which change occurred. Pay, shorter hours, autonomy or reduced workload should become visible in contracts, payroll, schedules or credible worker surveys. None follows automatically from faster task completion. Large gains for some workers and losses for others can cancel out. Future studies need occupational and distributional results, not only an economy-wide mean.
In a customer-support study, generative AI raised issues resolved per hour by 15% on average. In Denmark, workers reported that most saved time went into other work tasks, while administrative records showed no measurable average change in earnings or recorded hours through 2024. Faster work is real in some tasks; who benefits is a separate question.
The bargaining question begins after the stopwatch stops.
Where the connection stopsThe studies do not measure every change in work quality, stress or bargaining power. Average effects may hide different results by occupation, firm and worker.
How people may respond
Treating output, hours, pay and work experience as separate ledgers could change how firms and workers negotiate AI productivity gains.
What the evidence does not showThe studies do not measure every workplace or prove who captured every gain.
The story presents a task productivity result without converting it into a wage or employment result.
Firms and workers may ask whether saved time becomes output, different tasks or staffing change.
Attention moves to job design and distribution rather than the demonstration alone.
Employers and workers: Record task, staffing, workload and quality changes after adoption.
Work design could change even while average scheduled hours remain stable.
What we know has changedWe have not established that this possible change has happened.
What this does not showA redesign would not establish that all workers benefit or lose.
This is the first time Lens has mapped this path. We have no later evidence showing whether it is happening more, less or about the same.
The story names contracts, payroll, schedules and credible surveys as the places a benefit would become visible.
Workers and representatives may test a productivity claim against pay, autonomy, workload or shorter hours.
Attention shifts from capability to distribution.
Workers, employers and bargaining institutions: Negotiate and measure the form in which gains are shared.
A documented worker benefit could replace an assumed one.
What we know has changedWe have not established that this possible change has happened.
What this does not showOne workplace agreement would not establish the economy-wide average.
This is the first time Lens has mapped this path. We have no later evidence showing whether it is happening more, less or about the same.
If saved time may become more output, different tasks or smaller teams
Then Saved time may become more output, different tasks or smaller teams. Job design and staffing records—not a demonstration of the tool—will show which change occurred.
Would weaken this: The studies do not measure every change in work quality, stress or bargaining power.
Scope: Workplace AI can raise task productivity without automatically raising pay or reducing hours. Horizon: The next policy, reporting or research update.
If pay, shorter hours, autonomy or reduced workload should become visible in contracts, payroll, schedules or credible worker surveys
Then Pay, shorter hours, autonomy or reduced workload should become visible in contracts, payroll, schedules or credible worker surveys. None follows automatically from faster task completion.
Would weaken this: Average effects may hide different results by occupation, firm and worker.
Scope: Workplace AI can raise task productivity without automatically raising pay or reducing hours. Horizon: The next policy, reporting or research update.
A study of 5,172 support agents found a 15% average increase in issues resolved per hour, with substantial variation.
Open evidence ↗Danish surveys and administrative records found precise average null effects on earnings and recorded hours through 2024.
Open evidence ↗About 80% of saved time was redirected to other work tasks; fewer than 10% reported extra breaks or leisure, and 3–7% reported higher earnings.
Open evidence ↗What could change this assessment?
In a customer-support study, generative AI raised issues resolved per hour by 15% on average. In Denmark, workers reported that most saved time went into other work tasks, while administrative records showed no measurable average change in earnings or recorded hours through 2024. Faster work is real in some tasks; who benefits is a separate question.
A study of 5,172 support agents found a 15% average increase in issues resolved per hour, with substantial variation. Danish surveys and administrative records found precise average null effects on earnings and recorded hours through 2024. About 80% of saved time was redirected to other work tasks; fewer than 10% reported extra breaks or leisure, and 3–7% reported higher earnings.
The records support the distinction expressed in the answer, but not a claim about every person, place or future outcome.
The studies do not measure every change in work quality, stress or bargaining power. Average effects may hide different results by occupation, firm and worker.
Original government, institutional and peer-reviewed records listed on this page. Secondary reporting was not used to establish the central answer.
Last checked 2026-10-10.
A study of 5,172 support agents found a 15% average increase in issues resolved per hour, with substantial variation.
Quarterly Journal of Economics · 4 February 2025 · Checked 10 October 2026Danish surveys and administrative records found precise average null effects on earnings and recorded hours through 2024.
National Bureau of Economic Research · September 2025 · Checked 10 October 2026About 80% of saved time was redirected to other work tasks; fewer than 10% reported extra breaks or leisure, and 3–7% reported higher earnings.
Authors' supplementary survey tables · September 2025 · Checked 10 October 2026Free Lens updates
Our free weekly email brings you the latest Lens reporting and what changed.