OpenAI chief scientist Jakub Pachocki says the company is approaching AI systems that can "work indefinitely in a coherent way just like people do," signaling a shift from code assistants to autonomous research capabilities.1
"I think we will get to a point where you kind of have a whole research lab in a data center," Pachocki told MIT Technology Review.1 The prediction reflects advances in reasoning models that extend AI work sessions without human oversight.
Pachocki attributes the progress to improved base capabilities rather than architectural breakthroughs. "Simple boost in all-round capability leads to models that can work longer without help," he explained.1
The infrastructure market is splitting to support this evolution. NVIDIA's RTX PRO 6000 Blackwell targets high-end training workloads, while new technologies like DiLoCoX-107B distributed training and RL-KPI reinforcement learning frameworks optimize inference efficiency.2 Palantir shares rose 6% as investors positioned for AI infrastructure demand.
The autonomous research vision raises deployment concerns. Pachocki recommends "very powerful models should be deployed in sandboxes cut off from anything they could break or use to cause harm."1 He acknowledged regulatory gaps: "I think this is a big challenge for governments to figure out."1
The technology shift occurs as startup funding slows 23% in March 2026, pressuring AI companies to demonstrate practical value beyond code completion.3 Autonomous research systems represent a higher-margin use case than developer tools, potentially justifying infrastructure investments in specialized hardware and frameworks.
Industry observers note the transformation creates bifurcated opportunities: established infrastructure players benefit from training demand, while specialized vendors target inference optimization. The regulatory uncertainty around concentrated AI capabilities remains unresolved as technical progress accelerates.