OpenAI details its agent-powered research progress
OpenAI coding agents now power 3.1 workdays per human day, boosting research speed See how token spending, multitasking, and new drug data are reshaping AI use
OpenAI has published a look at how its coding agents are being used inside the company, offering a rare view into its internal research workflow. The report says agents are now doing the equivalent of 3.1 workdays for every one human workday, and the company says it has reached the “automated research intern” milestone Sam Altman outlined for September.
According to the report, OpenAI researchers now spend more than $600 a day in agent token usage on average, while the 90th percentile exceeds $7,000. The company also said token output has risen sharply since December, and that about 80% of researchers are using four or more agents at the same time.
The broader newsletter also highlighted a new Naturepublished study on Insilico Medicine’s AIdesigned drug rentosertib, which showed early signals in a small sample that patients may appear biologically younger on several aging measures. Another section reviewed NBC News polling showing that many U.S. voters remain uneasy about AI, even as more people report using it.