OpenAI says coding agents now provide 3.1 agent-workdays per researcher workday

OpenAI says its internal coding agents are materially speeding up research, but the company is pairing that claim with a warning: faster progress does not automatically justify pursuing rapid self-improvement. In a September 6, 2026 update from OpenAI News, the company shared early measurements of how agents are changing researchers’ daily work.

By mid-August, the median researcher was using more than $600 per day in agent inference at API prices. Researchers at the 90th percentile exceeded $7,000 per day in token usage. The organization was also running the equivalent of 3.1 agent-workdays for every human workday, based on an eight-hour day. Agents are being used in concurrent sessions to write code, run experiments, build evaluations and investigate failures.

OpenAI says it has reached its previously announced goal of an “automated research intern” by September 2026. The term refers to a system that can complete well-defined tasks under human direction, including work that might take a skilled researcher several days. The company says it is making strong progress toward an automated AI researcher by March 2028.

For developers building AI workflows, the report is a useful signal that agentic coding is moving from occasional assistance toward sustained, parallelized work. It also highlights the limits of the metrics: more code and experiments do not guarantee the same increase in overall research progress, since bottlenecks remain across the process.

OpenAI says people still choose research priorities and decide whether to scale, pause or deploy systems. Following a recent Hugging Face incident, it paused reinforcement-learning training on some deployment-bound models while strengthening controls and monitoring. The company says it will slow or stop development when systems cannot be adequately safeguarded.

Source: OpenAI News