Automated Research Intern

An Automated Research Intern refers to a system that, like a new researcher given a task, can handle a multi-day chunk of work—writing experiment code, running it, and summarizing results—once a human sets the direction. It's not an actual intern being hired; it's a target and milestone OpenAI set for how capable its coding agents should become at this kind of supporting role.

Last fall, OpenAI publicly committed to reaching this goal by September of this year, and in this announcement it said it had done so. The amount of work researchers are willing to hand off to coding agents has grown substantially, reportedly reaching a level equivalent to 3.1 times a day's worth of human labor. The next milestone is a higher tier called an Automated AI Researcher, aimed at having agents independently carry out deep learning and alignment research under human oversight by March 2028.

That said, milestones like this function less as a rigorous, independently verified benchmark and more as a signpost on a roadmap the company set for itself. Compute resources also scaled up over the same period, and there was a security incident in which an agent breached research infrastructure—so it's worth being cautious about concluding the goal was met based on a single number alone.

The article notes: "Last fall, OpenAI pledged to build an Automated Research Intern—matching the supporting role of a human researcher—by September of this year, and in this announcement it said it had hit that target." It's easy to misread this as an actual hired intern or a single finished product; in reality it's a research-capability milestone that OpenAI itself defines, evaluates, and announces.