AI and Productivity: What the Evidence Shows So Far

What studies of AI in real workplaces have found about productivity, why economy-wide effects take time, and the range of forecasts.

Artificial intelligence is often described as a technology that could transform the economy. But how much has it actually raised productivity so far, and what do economists expect?

Evidence from workplaces

Studies of specific tasks and firms have found meaningful gains:

A study of customer support agents found AI assistance increased issues resolved per hour by about 14 percent on average.
Experiments with professional writing tasks found AI helped people finish faster and improved quality, especially for weaker writers.
Studies of software developers using AI coding assistants found faster completion of some programming tasks.

A common finding is that AI often helps less experienced workers most, narrowing gaps with experts.

Economy-wide effects

Gains in individual tasks do not automatically add up to large economy-wide gains. Economists expect effects to depend on:

How many tasks AI can do well and cheaply.
How quickly businesses adopt it and reorganise work.
Complementary investments in data, skills and new processes.

A range of forecasts

Forecasts vary widely. Some, including a 2023 Goldman Sachs analysis, suggested AI could raise global GDP substantially over a decade. Others are more cautious. Daron Acemoglu estimated in 2024 that AI might raise U.S. productivity only modestly over the following decade, because the share of tasks it could profitably automate in that time was limited.

Why the task gain is not the economy gain

Suppose AI doubles the speed of writing first drafts, which takes 10 percent of a lawyer's time. The lawyer's total productivity rises by only about 5 percent, since the other 90 percent of their work is unchanged. Across the economy, gains depend on how many tasks AI affects and by how much.

The productivity paradox again

In the past, technologies like electricity and computers took years to show up in productivity statistics. Many economists expect AI’s effects to follow a similar pattern, with gains growing as businesses learn how to use it.

Judging AI's impact by headlines

Dramatic predictions and dramatic dismissals both circulate widely. The evidence so far shows real but uneven gains in specific tasks, with the economy-wide effect still uncertain.

Key takeaways

Workplace studies find real productivity gains from AI in tasks like customer support, writing and coding.
AI often helps less experienced workers most.
Economy-wide effects depend on how many tasks AI affects and how quickly firms adapt.
Forecasts range from modest to large gains, and effects may take years to appear.