OpenAI Targets Automated Researcher by 2028

OpenAI’s metrics suggest agentic coding is becoming core research infrastructure, but “agent-workdays” measure activity—not equivalent scientific progress.

• –Researchers are running more concurrent agents, writing more code, and conducting more experiments.
• –Human researchers still set priorities, assess results, and decide what advances into core training runs.
• –As routine implementation accelerates, compute, research judgment, and experiment bottlenecks become more important.
• –The 2028 target is ambitious, but the report offers an unusually concrete look at how frontier labs are operationalizing automated research.