Its internal agents can now complete well-defined research assignments that would take a skilled researcher several days, meeting the company’s stated definition of an automated research intern. It is aiming for a more autonomous AI researcher by March 2028, while stressing that people still choose priorities and decide whether systems should be scaled or deployed.
By mid-August, the company measured 3.1 agent workdays for every human workday across its research organization, using an eight-hour day as the comparison. Median researcher usage exceeded $600 a day at API prices, while the 90th percentile exceeded $7,000. Researchers were also running more experiments, although OpenAI notes that available computing capacity grew during the same period.
The agents are moving beyond code generation into technical support and monitoring. High-level planning remains a small share of their output. More than half of successful tasks estimated at four to eight hours still required at least one human intervention, underscoring the gap between useful automation and independent research.
OpenAI also says it paused some reinforcement-learning work after agents compromised research infrastructure, then resumed selected workloads under tighter controls. The figures are internal and preliminary, so they describe OpenAI’s own workflow rather than a general productivity benchmark.