OpenAI AI Research Assistants Arrive, Fully Automated Researchers by 2028
The biggest topic in the AI world last week was OpenAI’s GPT-6 Astra, presented by the company as a major step toward AGI (Artificial General Intelligence). Now, OpenAI has revealed another milestone: AI research assistants are here, and fully automated AI researchers could arrive by 2028.
In a new blog post, OpenAI announces that it has achieved its goal of creating an automated research assistant capable of working under human supervision. The company notes that these systems can handle well-defined research tasks that would take a skilled researcher several days to complete.
OpenAI’s AI assistants are already doing more work
AI is already transforming the world in many ways, from academia to the markets. Now, it seems to be the researchers’ turn; this technology could drastically shape the future of research. OpenAI has revealed that its AI agents are already acting as additional support for its research teams.
The company indicates that researchers use coding agents throughout the day, often running multiple sessions simultaneously. These agents write more code, conduct more experiments, and take on increasingly complex tasks.
By mid-August, OpenAI’s research team was using the equivalent of 3.1 agent workdays for every human workday. The company also observed that an increasing number of researchers are running four or more agents at the same time.
Furthermore, agents are no longer asked solely to write code. OpenAI states that researchers are delegating higher-level, longer-duration tasks to them, such as research planning, drafting technical specifications, analysing experiments, overseeing training processes, and troubleshooting research infrastructure. However, the results are not perfect. OpenAI acknowledges that the agents still require significant human guidance as tasks become more complex. Over the past six months, more than half of the successfully completed tasks—estimated to require between four and eight hours of human work—involved at least one human intervention.
Fully automated AI researchers could arrive by 2028
OpenAI states that it is making “significant progress toward creating an automated AI researcher by March 2028.” The goal is to build a system capable of working on deep learning and AI alignment under human supervision, thereby helping to advance AI research itself.
If this happens, AI would no longer be limited to helping humans write code or analyse experiments. It could increasingly participate in the research cycle itself, generating ideas, developing them, running experiments, and analysing the results.
OpenAI believes that automated research could also contribute to AI safety and alignment, including the development of better defences against increasingly capable AI systems.
However, the company does not mean to suggest that the path ahead is free of risk. It notes that more advanced systems may be harder to oversee and that safety and alignment efforts must evolve alongside capabilities. Furthermore, OpenAI states that it is prepared to slow down or halt development if a system’s safety cannot be sufficiently guaranteed.
For now, it is humans who decide which research is relevant and which ideas are worth exploring. However, if OpenAI achieves its goal for 2028, the definition of an AI researcher could change radically by the end of the decade.