OpenAI develops automated AI ‘research intern’ ahead of schedule: What to know

OpenAI successfully deployed an automated research intern capable of executing structured research projects autonomously including ones that would take a human researcher several days.

The company said its internal measurements demonstrated that it has now attained the objective for a more capable automated AI researcher.

Measured in standard eight-hour workdays-the research organization is now using nearly 3.1 agent-workdays of effort daily workday of human labor.

A rising share of researchers are also executing massive-scale parallel agent workflows with some running four or more agents at once up from earlier in the year.

OpenAI classified the work into autonomous coding assistants using a research and development taxonomy developed by Epoch AI-ranging from problem discovery to hypothesis testing, building code and datasets, analysis and communicating findings.

The significant shift appeared unexpectedly, internal channels where researchers used to request cross-team escalations or experiments have seen declining traffic in 2026.

The company said one team has ended open-door availability believing that agents are cushioning the impact of that support burden. Meanwhile, agentic tools are streamlining their research workflow, allowing researchers to scale, pause, and ship a system whenever they select.

Nonetheless, the self-defined research term bar in September is closer to its stated 2028 goal of a fully automated AI researcher though outside observers note that there are no proper ways to verify these claims.