OpenAI announced that it had achieved its goal of developing an “automated research intern” by September 2026. The company defines this system as one that, under human guidance, carries out well-defined research tasks and completes work that could take a skilled researcher several days. Research priorities and scaling decisions are still made by humans. Sam Altman announced the goal in October 2025; the next target is to reach the level of an “automated AI researcher” by March 2028.
Coding agents, which were used on a limited basis at the beginning of 2026, became part of researchers’ daily workflows in mid-August. The average researcher’s daily inference cost for the agents they used, calculated at API prices, rose above $600; there were approximately 3.1 agent workdays for every human workday. Before June 2026, this figure was below human labor. The agents take on coding, experiment analysis, infrastructure issues, and technical support tasks. In August 2026, the number of experiments per researcher reached its highest level since January 2025. Human intervention was required for more than half of the agents’ 4- to 8-hour tasks. In the long term, OpenAI aims for AI to contribute to the development of new AI systems.
Why it matters
This development shows that AI research tools are moving beyond experimental assistants and becoming part of researchers’ regular workflows. For research teams, the shift is not limited to the division of tasks; the agents’ high daily usage cost means this way of working must be evaluated alongside resource and scaling decisions. Although the increase in the number of experiments points to potential productivity gains, the need for intervention in most 4- to 8-hour tasks shows that the systems do not yet operate like independent researchers. The open question, therefore, is at which stages human oversight will be maintained and to what extent responsibility for setting research priorities will remain with humans.
Background
OpenAI is not a new name in the FikirPilot archive: we have published 25 articles mentioning it in the past 90 days, the latest dated September 7, 2026.
Term: agent
An AI agent is software that calls tools and carries out multi-step tasks to achieve a goal, rather than producing a single response.