← All papers
The Quarterly Journal of Economics
Generative AI at Work
Research summary
A staggered rollout to 5,172 customer-support agents increased issues resolved per hour by 15% on average. Less experienced workers gained most, while the most experienced saw smaller gains and slight quality declines.
Evidence strengths
- 5,172 workers
- Real-world deployment data
- Measures productivity, quality, learning, and worker experience
Limitations
- Single firm and occupation
- Staggered rollout is less controlled than a randomized experiment
Topics
- Human-AI Interaction
- Artificial Intelligence
- Learning
- Social Intelligence
Connected claims
- Supports
Human-AI collaboration does not reliably outperform the better standalone human or AI; results depend strongly on the task and interaction design.
AI assistance improved productivity in a specific customer-support setting, with heterogeneous effects by worker experience.