Openai's internal AI agent surpasses human efficiency in research tasks

OpenAI has announced that its internal AI agent has achieved a significant milestone by completing complex research tasks independently, surpassing human efficiency. The system, operating under human supervision, has demonstrated the ability to perform tasks that typically require senior researchers several days to complete. As of mid-August 2026, the ratio of AI agents to human researchers within OpenAI's core research organization has reached 3.1 to 1. This means that for every human workday, the research organization utilizes 3.1 agent workdays. Kevin Liu, an OpenAI employee, emphasized that recursive self-improvement is likely to be a key driver of AI advancement. He called for greater industry transparency to enable public discourse on controlling future model development. The 3.1:1 ratio reflects the actual runtime and workload of agents, not direct productivity equivalence. This shift highlights the integration of parallel working models within OpenAI's research ecosystem. Liu urged other AI companies to adopt similar transparency, ensuring the broader community can engage in informed discussions about AI progress and regulation.