METR Survey of 349 Technical Workers and Task-Substitution Model Separate AI Speed From Economic Uplift

Economists and empirical researchers at Berkeley-based institute METR released a pair of workforce studies on Sunday, October 18, examining how frontier artificial intelligence tools reshape technical productivity and task selection. Authored by METR researcher Joel Becker alongside economic researchers Tom Cunningham and Parker Whitfill on Shattuck Avenue in Berkeley, the reports combine survey data with formal economic modeling. Technology council directors, corporate R&D departments, and workforce development boards in San Francisco, Walnut Creek, San Ramon, and Oakland are using the findings to benchmark enterprise AI investments.

Becker published results from a detailed February through April 2026 survey of 349 technical workers—comprising 87 software engineers, 71 AI researchers, 129 academics and PhD students, and 48 founders and engineering managers—measuring self-reported productivity gains from frontier AI tools. The study carefully distinguishes between raw task speed, meaning how much faster an isolated task is completed, and net economic value created across a project. Becker noted that in METR's earlier 2025 randomized developer trial, participants overestimated their AI time savings by an average of 40 percentage points.

In a companion economic analysis titled Task Substitution and Uplift, Cunningham and Whitfill formalize three distinct productivity metrics comparing pre-AI (2021-style) task portfolios against post-AI task portfolios. Their mathematical model tracks what happens when automated tools sharply lower the cost of specific engineering activities while leaving other analytical steps unchanged. Regional economic commission officials noted that understanding task substitution is vital for forecasting tax revenues and $150,000-plus technical salaries across Bay Area innovation districts.

Through their formal framework, Cunningham and Whitfill prove the fundamental inequality that AI uplift on old tasks is less than or equal to uplift in total economic value, which in turn is less than or equal to uplift on new tasks. This ordering holds because engineers and researchers systematically shift their workdays toward AI-leveraged tasks such as rapid prototyping, exploratory simulations, and interactive data tooling that would have been too slow to attempt in 2021. Consequently, measuring only legacy benchmarks misses the value generated by newly feasible workflows.

Enterprise engineering leaders, labor economists, and municipal workforce program planners across Contra Costa and Alameda counties can download both the 349-worker survey report and the Task Substitution paper online at metr.org. Questions regarding workforce survey methodology or institutional research funding may be directed to METR at 2150 Shattuck Avenue in Berkeley or by calling (510) 982-3130 between 8:30 AM and 5:00 PM Monday through Friday. Additional economic seminar schedules are listed on the institute's portal.