OpenAI Hits "Automated Research Intern" Milestone, Agents Log 3.1x Human Workdays

rohanpaul_ai · x · 2026-09-07

OpenAI says it has reached its "automated research intern" milestone: a human-supervised system able to complete well-defined tasks that would take a skilled researcher days.

Internal figures show how deeply agents are embedded in research: as of mid-August, the research org runs 3.1 agent-workdays of effort per workday of human labor. Cited data also suggests OpenAI engineers now change roughly 7x more code per contributor than the pre-2025 baseline.

Caveat: the 3.1:1 ratio measures agent runtime rather than equivalent productivity, but it captures how parallel agent work has become core to OpenAI's research process.

Related event: OpenAI Says AI Research Intern Goal Met, Fully Automated Researcher by March 2028(29 posts)→

Original post →

More from coding & agent

coding & agent channel →