AI obviously makes some tasks faster.
Whether that adds up to real, measurable productivity for a business or an economy is a much harder question, and the honest answer is: sometimes, unevenly.
TL;DR
- Customer support agents using an AI assistant were about 14% more productive, with the largest gains, around 34%, going to the least-experienced workers.
- Yet 95% of enterprise generative AI pilots delivered no measurable profit-and-loss impact.
- In one trial, experienced developers were actually 19% slower with AI, while believing they were faster.
The productivity findings here come from peer-reviewed field studies, the International AI Safety Report, and the Stanford AI Index.
What the evidence actually shows
| wdt_ID | wdt_created_by | wdt_created_at | wdt_last_edited_by | wdt_last_edited_at | Setting | Finding |
|---|---|---|---|---|---|---|
| 1 | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | Call center agents (QJE) | 14% overall, +34% for novices |
| 2 | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | Software developers (Microsoft, Accenture) | 26% tasks completed |
| 3 | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | Experienced developers (METR trial) | 19% slower |
| 4 | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | emmanuel-ashemiriogwa | 04/08/2026 10:57 AM | Enterprise pilots (MIT NANDA) | 95% show no P&L impact |
A review of task-level studies found productivity gains of 20 to 60% in controlled settings, but only 15 to 30% in most real-world work.
The cleanest and most durable finding is that AI levels up the bottom: it helps less-experienced workers most and barely moves top performers.
Why task gains don’t automatically become business gains
A 14% boost on one task does not mean a 14% more productive company.
Enterprise tools often stall in pilots because they cannot retain context or fit a specific workflow, which is why so many show no bottom-line impact.
And AI has a “jagged frontier”: it lifts performance on tasks within its capability and quietly degrades it on tasks just beyond, which workers struggle to detect.
Where the U.S., UK, and Canada sit
At the economy level, the link is still emerging. U.S. labor productivity growth has climbed to an annualized 1.8 to 2.7%, above the 1.4% average of the previous decade, with economists attributing much of it to early AI capital investment rather than direct worker gains.
For U.S., UK, and Canadian firms, the practical lesson is that AI reliably speeds up structured, checkable work and helps newer staff, but turning that into measured productivity requires redesigning the work, not just buying the tool.
ELI5
AI helps people do certain jobs faster, especially people who are still learning. But it does not automatically make a whole company more productive, and sometimes it even slows down experts. The benefit depends a lot on the task.
Sources
