A new experiment suggests AI can make people better and faster at many common work tasks, but not all of them.
Researchers at the AI Security Institute (AISI) ran a pilot study to measure a simple question: does access to a modern large language model actually boost productivity on everyday job activities?
To test the hypothesis, they turned to O*NET, a U.S. Department of Labor database that breaks jobs into common “work activities” shared across professions. Instead of focusing on one career, they picked four broad tasks that show up in many jobs: monitoring processes, drafting technical material, organizing and planning work and interpreting information for others.
The team recruited 500 participants and used a randomized controlled trial. One group had access to a state-of-the-art AI model released in early 2025. The other group did not. Because the two groups were otherwise comparable, any difference in results could be linked to AI access.
Participants were graded on three things: quality of their output, time taken, and “points per minute,” a combined measure of productivity. The results were uneven.
On Task 1, which involved monitoring processes or materials, AI users scored 22% higher in quality. But they did not finish faster.
On Task 2, drafting and specifying technical devices or parts, AI again improved quality by 23%. Time and overall productivity did not meaningfully change.
Task 3 was different. The task asked participants to organize, plan and prioritize work, producing more open-ended, forward-looking proposals. Here, AI made no statistically significant difference in quality, time, or productivity.
Task 4 flipped the pattern. When participants were asked to interpret information for others, AI did not improve quality, but it cut completion time by 42% and boosted points per minute by 102%.
In simple terms, AI helped most when the work involved structured analysis, synthesis, and interpreting data. It struggled more when the task required subjective judgment and strategic thinking.
Researchers say the results show the “jagged capabilities” of AI systems. They excel in some domains and lag in others.
But authors warn that the sample size was limited and the findings are preliminary. Still, the pattern lines up with other research: AI can meaningfully improve performance on many structured tasks, but it is not yet a universal productivity engine across all types of work.
Disclaimer: Opinions expressed at CapitalAI Daily are not investment advice. Investors should do their own due diligence before making any decisions involving securities, cryptocurrencies, or digital assets. Your transfers and trades are at your own risk, and any losses you may incur are your responsibility. CapitalAI Daily does not recommend the buying or selling of any assets, nor is CapitalAI Daily an investment advisor. See our Editorial Standards and Terms of Use.