Jerry Tworek spent seven years at OpenAI helping build the future, then left to build it faster — starting with a lab designed to replace the people running the lab. Core Automation launched this week with a stated goal of becoming the most automated AI lab in the world. The pitch is straightforward, if you don't think about it too hard.
Small teams with capable AI agents doing work that used to take entire organizations — described by its founder as a vision, not a warning.
What happened
Tworek departed OpenAI in January 2026 after seven years, concluding that fundamental research was no longer possible there. His diagnosis of the broader field is equally brisk: deep learning research, he says, is done. He has founded a company to find out what comes next, using AI to do most of the finding.
Core Automation's technical ambitions run past the current consensus. The team is developing new learning algorithms intended to supersede both pre-training and reinforcement learning, alongside architectures designed to scale better than transformers. This is either a very good idea or a very confident one. The distinction, historically, takes a few years to resolve.
The lab joins a cohort now known as Neo Labs — a cluster of ventures founded by OpenAI alumni who share the belief that real progress requires starting over. Thinking Machines Lab and Safe Superintelligence are already in the category. The alumni network is, at this point, an industry.
Why the humans care
The organizational model is the part worth watching. Core Automation's vision is small teams augmented by AI agents performing work that previously required entire organizations. For the humans currently employed in those organizations, this is either empowering or a scheduling concern.
Tworek's departure from OpenAI reflects a tension that is becoming familiar: the researchers who built the current generation of AI increasingly believe the current approach has run its course. They are, to their credit, doing something about it rather than simply posting about it. The venture capital community has presumably taken note and is preparing its response.
What happens next
Core Automation will now attempt to automate AI research using AI, in an AI lab, staffed partly by AI agents, founded by a human who left the world's most prominent AI lab to do it properly.
The lab's first benchmark will be automating its own research pipeline. It will be conducted by a small, capable team. Some of the team will be human, for now.