It all started with AI chatbots that could answer questions and help with homework or research. Then came AI agents that can carry out tasks on their own with minimal input. So what’s the next step? For OpenAI, its “North Star” for the next few years is to build a fully automated, agent-based AI researcher. According to the company, this autonomous AI system would be capable of handling complex problems independently, without needing step-by-step human guidance.
In simple terms, OpenAI is reportedly working on an AI system that can plan its own work, analyse information, test different ideas, and keep working on a problem for hours or even days, instead of simply replying to prompts. In a conversation with MIT Technology Review, OpenAI chief scientist Jakub Pachocki described the project as the company’s new long-term goal.
Pachocki said the company aims to bring together multiple research strands, including reasoning models, autonomous agents, and interpretability, into one unified system capable of tackling large and complex problems in fields such as mathematics, physics, and the life sciences, with minimal human intervention.
According to the report, OpenAI’s goal behind this ambitious project is to move beyond chatbots and coding assistants and build an AI system that behaves more like a real researcher. Instead of answering one question at a time, the system would be able to plan tasks, analyse information, test ideas, and work on complicated problems for long periods on its own. OpenAI internally calls this goal its “North Star”, meaning it will guide most of the company’s research over the coming years.
First step is to build an AI research intern
To begin with, Pachocki said the first step will be creating what OpenAI describes as an AI research intern. This would be an autonomous agent capable of taking on smaller research assignments that would normally take a human several days to complete. Over time, the company plans to scale this into a larger multi-agent system, where multiple AI programs work together inside data centres to handle bigger and more complex projects. In theory, such a system could be used for scientific research, software development, business decisions, or even policy analysis.
Interestingly, OpenAI’s plans to build an autonomous researcher, comes at a time when the race to build autonomous AI systems is heating up. Companies such as Google DeepMind and Anthropic are also developing advanced reasoning models and AI agents. OpenAI says recent progress in coding agents and reasoning-based AI shows that machines are already improving at working for longer without human help. Some internal tools are now also showing the ability to complete tasks that earlier took days of manual work. The company is seeing this as an early sign that fully automated research systems could soon become a reality.
Meanwhile, the project also raises serious concerns. Systems that can run independently for long periods could make mistakes, misunderstand instructions, or be misused if not properly controlled. OpenAI says it is already experimenting with ways to monitor how AI reasons while it works, and may keep highly capable systems in restricted environments to reduce risks.
Pachocki added that fully human-level intelligence, often referred to as AGI, may still be far away, but even partially autonomous AI could be powerful enough to change how research, technology, and businesses operate, making the race to build such systems one of the biggest challenges in the AI industry today.
