Why AI’s Productivity Gains Often Lead to More Work, Not Less

AI feels fast, but work does not get lighter

First‑time users of generative AI report that writing emails, proposals, code, or summaries now takes minutes instead of half an hour or a whole day. The speed creates the impression that AI should make work easier.

Efficiency rises, but workload does not shrink

Research from NBER on 5,179 customer‑service agents shows that using an AI assistant raises average productivity by 14 %, and for novices or low‑skill staff the gain reaches 34 %. Another NBER six‑month experiment with knowledge workers finds that weekly email time drops by about two hours after AI adoption.

Who captures the saved time?

If the saved time benefits employees, they may work fewer overtime hours and repeat fewer low‑value tasks. If the organization captures it, the result is more tasks, faster response expectations, and higher‑frequency deliveries, leaving many feeling more exhausted.

Saved time is quickly refilled

A workplace rule says that once a task becomes faster, it is soon expected to be that fast. Examples: a project summary that once took half a day can now be delivered in two hours, prompting managers to ask for three summaries instead of one. Weekly retrospectives that used to occur once a week become twice a week because AI can generate minutes and action items instantly.

Upwork’s AI survey illustrates the paradox: 96 % of executives expect productivity gains, yet 77 % of employees report that AI has increased their workload.

Work becomes more granular

AI splits a single report into many micro‑steps, such as:

Ask AI to draft titles and choose the most suitable.

Ask AI to outline the structure and verify it matches the problem.

Ask AI to suggest angles and discard irrelevant ones.

Ask AI to expand content while checking facts and boundaries.

Ask AI to adjust tone, compress wording, convert to a presentation format, then manually review, rewrite, merge, and confirm.

Each step is small but requires judgment, creating a new layer of cognitive labor that many describe as a persistent fatigue.

Local efficiency vs. overall workflow

Companies often focus on whether a single task becomes faster, ignoring whether the whole process shortens. An employee who reduces weekly report writing from 40 minutes to 10 minutes may face new expectations to produce daily briefs, risk summaries, and client updates, nullifying any net time gain. Similarly, AI‑generated meeting minutes can lead to more follow‑up actions, status updates, and tracking tables, increasing management density.

Raising the baseline of normal output

AI changes what is considered “normal.” Previously, a report without an abstract was acceptable; now AI‑generated abstracts are expected. A single proposal version was enough before; now multiple AI‑generated versions become the default. This hidden pressure raises both the upper limit of capability and the minimum expectation.

Managing work boundaries, not the tool

To prevent AI from merely adding work, organizations should:

Cancel unnecessary meetings instead of generating more minutes.

Consolidate low‑value reports rather than expanding weekly reports into daily ones.

Reduce redundant approvals and status updates instead of maintaining extra tracking sheets.

Explicitly protect the time saved rather than immediately filling it with new tasks.

Without deliberate redistribution, efficiency gains flow into higher output demands, making AI a pressure tool rather than a relief.

Conclusion

AI accelerates individual tasks but does not automatically reduce total work; it raises work density. When organizations do not redefine boundaries, the extra capacity is consumed by additional expectations, leaving employees no breathing space despite higher apparent efficiency.