OpenAI's Moonshot: Replace the Research Lab with AI

OpenAI has a new North Star, and it's audacious even by Silicon Valley standards: build a fully automated AI researcher — a system capable of independently tackling scientific, mathematical, and policy problems too large or complex for any human team to handle alone.

In an exclusive interview with MIT Technology Review, OpenAI chief scientist Jakub Pachocki laid out the company's most ambitious research vision yet, complete with a concrete timeline that the AI industry will be watching closely.

The Vision: A Research Lab in a Data Center

"I think we are getting close to a point where we'll have models capable of working indefinitely in a coherent way just like people do," Pachocki told MIT Technology Review. "Of course, you still want people in charge and setting the goals. But I think we will get to a point where you kind of have a whole research lab in a data center."

This isn't just a philosophical aspiration. OpenAI has mapped out a two-phase plan with hard deadlines:

• By September 2026: An "autonomous AI research intern" — a system that can independently take on specific research problems that would take a human researcher a few days to complete
• By 2028: A fully automated multi-agent research system capable of tackling problems at a scale no human team could manage

The intended application domains are sweeping: mathematics (new proofs and conjectures), life sciences (biology and chemistry), and even business and policy analysis. In theory, Pachocki says, "you would throw such a tool any kind of problem that can be formulated in text, code, or whiteboard scribbles — which covers a lot."

Codex as the Prototype

OpenAI's January 2026 launch of Codex — an agent-based app that can autonomously spin up code, analyze documents, generate charts, and manage complex tasks — is the early proof of concept. Pachocki says OpenAI claims most of its technical staff now use Codex in their work.

"There's a big change happening, especially in programming," Pachocki said. "Our jobs are now totally different than they were even a year ago. Nobody really edits code all the time anymore. Instead, you manage a group of Codex agents."

The key capability gap to close: making AI systems that can operate for longer periods with less human guidance. Right now, even sophisticated agents need human check-ins after hours or days. The AI research intern would need to work reliably for days at a stretch. The 2028 system would need to run open-ended research programs autonomously.

The Technical Path

Pachocki points to a clear progression in AI capability for long-horizon tasks. GPT-4, compared to GPT-3, could work on a problem for far longer without losing coherence. Reasoning models — which think through problems step by step and backtrack when stuck — brought another jump. The next leap comes from training on complex, multi-step tasks like hard math and coding competition problems, which force models to learn how to manage huge contexts and break complex problems into subtasks.

Doug Downey, a research scientist at the Allen Institute for AI (unaffiliated with OpenAI), validated the logic:

"There are a lot of people excited about building systems that can do more long-running scientific research. I think it's largely driven by the success of these coding agents. The fact that you can delegate quite substantial coding tasks to tools like Codex is incredibly useful and incredibly impressive. And it raises the question: Can we do similar things outside coding, in broader areas of science?"

Context: Competition and Urgency

This announcement comes at a moment of intense competitive pressure for OpenAI. Anthropic's Claude Code and Claude Cowork have reportedly been eating into OpenAI's enterprise market — with OpenAI's own CEO of Applications Fidji Simo reportedly calling Anthropic's growth a "wake-up call" in internal communications that leaked this week. OpenAI is simultaneously narrowing its consumer product focus while swinging for the fences on the research frontier.

The automated researcher goal also threads a needle between OpenAI's stated mission (benefiting humanity) and its commercial reality (needing revenue). If the company can build AI systems that compress years of scientific research into months, the implications for drug discovery, materials science, and engineering are enormous — and so is the commercial opportunity.

The Bigger Picture

OpenAI is not alone in this vision. Anthropic CEO Dario Amodei has described his goal as building "a country of geniuses in a data center." Google DeepMind's Demis Hassabis has talked for years about solving humanity's hardest problems with AI. But OpenAI is now putting specific timelines and measurable milestones on the table — which means there will be something concrete to evaluate come September.

For the research community and the broader public, the question isn't just whether OpenAI can build an AI research intern by fall. It's what happens to scientific institutions, academic hiring, and the nature of intellectual work if they succeed — and succeed at scale.

The North Star has been named. The race to reach it is already underway.