OpenAI Discloses Progress on Recursive Self-Improvement: AI Agent Working Hours Hit 3.1 Times That of Humans.

Beating AI Express (Insight) – OpenAI set two goals for itself last year: to develop an "AI research intern" by September 2026, and a fully autonomous AI researcher by March 2028. The first goal has now been achieved. New internal data from OpenAI shows that when humans provide clear research questions, Agents can now independently complete tasks that would normally take skilled researchers several days. Researchers are increasingly delegating tasks to Agents. As of mid-August, based on the total workload of the research department, for every 8 hours of human work, multiple Agents collectively ran approximately 24.8 hours, equivalent to 3.1 Agent workdays. Prior to June this year, total Agent runtime was still lower than human work hours. It should be noted that this figure counts runtime, not a direct 3.1x increase in R&D efficiency. Agents are also taking on more demanding tasks: at the start of the year, they mainly handled research and infrastructure coding; now they are undertaking longer, more complex work such as experiment monitoring. In August, the number of experiments run per active experimenter also reached a new high since OpenAI began tracking this metric. However, research directions are still determined by humans. Agents rarely take charge of high-level research planning, and over the past six months, even for successfully completed 4-8 hour tasks, more than half required at least one human intervention. OpenAI’s next target is to upgrade the "intern" to an autonomous AI researcher capable of completing more comprehensive research work by March 2028.