Did AI improve the whole job, or just one step?
This paper proposes judging AI by the whole job, not one fast step. People should keep control and have a way to recover from errors. For your own setup, try mapping one task from start to finish. Mark who checks the result and what happens if it is wrong.
What the source reports
The authors propose a way to judge whether AI improves work, not just whether it can finish a task. They call for six checks that cover lasting value, human control, who answers for errors, and how people can recover. The checks also ask whether people keep learning, have paths to better jobs, and find purpose in their work. The paper applies this approach to a case study of social surveys that use AI. It offers a way to assess the whole job. It does not show that any tool that finishes a task will improve that job.
Original source
- Title
- When Does AI Augment Work? A Workflow-Level Framework for Human-Agent Collaboration
- Author
- CIVIC-AI Collaboration, Jiaying Wu, Caleb Ziems, Raymond Chan, Nancy F. Chen, Corlyss Chua, Gerard Chung, Jungpil Hahn, Wee Sun Lee, Zhengyuan Liu, Jamie Ng, Desmond C. Ong, Jeryl Ong, Da Ren Soon, Tianqi Song, Zhi-Xuan Tan, Sixing Tao, Emily Yang, Yajing Yang, Stella Xin Yin, Min-Yen Kan, Diyi Yang
- Publication
- arXiv
- Date
- Friday, September 11, 2026
Also covered by the other newsletter.