Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce
Shao, Y., Zope, H., Jiang, Y., Pei, J., Nguyen, D., Brynjolfsson, E., & Yang, D. (2025). Stanford University. arXiv:2506.06576v2
Summary
A large-scale audit of worker preferences and AI capabilities across 844 occupational tasks spanning 104 occupations. Introduces the Human Agency Scale (HAS) as a shared vocabulary for quantifying the automation-augmentation spectrum, and constructs the WORKBank database from 1,500 domain worker surveys and 52 AI expert assessments.
Why It Matters
This paper reframes the AI-and-work conversation from “what can AI do?” to “what do workers actually want AI to do?” It surfaces a significant disconnect between where AI investment is flowing and where workers want help.
Key Contributions
- Human Agency Scale (H1-H5): A human-centered framework for describing levels of AI involvement in tasks
- WORKBank Database: First large-scale audit combining worker desires and expert capability assessments
- Four-Zone Landscape: Categorizes tasks by desire × capability into actionable zones
- Skills Shift Evidence: Documents early signals of core competency changes (information → interpersonal)
Key Statistics
- 46.1% of tasks: workers express positive attitude toward AI automation
- 69.38% of pro-automation responses: primary reason is “freeing time for high-value work”
- 45.2% of occupations: H3 (equal partnership) is the dominant worker-desired level
- 41.0% of Y Combinator companies: mapped to Low Priority or “Red Light” zones
- 1.26%: Claude.ai usage share from top 10 occupations with highest automation desire
Methodological Notes
- Audio-enhanced survey interface allowing workers to articulate reasoning
- Tasks sourced from O*NET database (U.S. Department of Labor)
- Expert panel of 52 AI researchers and practitioners
- January–May 2025 data collection window