OpenAI reports progress towards automated research interns

OpenAI said its internal systems had reached its automated research-intern goal: handling some well-defined tasks that would take a skilled researcher several days, under human direction. The company described the result as progress in research assistance rather than fully autonomous scientific discovery.

Agents contribute code and run experiments, while human researchers choose priorities, assess results and decide which directions to pursue. Preliminary internal measurements show increases in coding and experimental activity, but those outputs do not translate directly into the same increase in overall research speed.

The account also describes the need to manage how systems are scaled and when work should be paused. More experiments create additional results that must be checked and interpreted.

Research throughput depends on identifying useful questions and recognising reliable answers. Automating the work between those decisions can help without removing the decisions themselves.