Jerry Tworek Launches Core Automation AI Lab
Former OpenAI researcher Jerry Tworek announced the creation of Core Automation on April 22, 2026. This new AI lab sets out to establish "the most automated AI lab in the world." The initial step involves automating the lab's own research activities.
Core Automation moves away from the pattern of building ever-bigger models that require massive amounts of training data. Company representatives state they plan to create fresh learning algorithms. These go past standard pre-training and reinforcement learning techniques. They also intend to design architectures that scale more efficiently than transformers.
Building a Specialized Team
The Core Automation group assembles specialists in frontier models, optimization, and systems engineering. Their goal centers on enabling small teams paired with advanced AI agents. Such setups would handle tasks that once demanded full organizations.
OpenAI, founded in 2015 as a nonprofit AI research entity, has long led efforts in developing large language models like GPT series. Tworek contributed there for seven years before departing. His experience includes work on key projects during OpenAI's growth phase.
Reasons for Leaving OpenAI
Tworek exited OpenAI in January 2026. He spent seven years at the organization. He believes fundamental research of that nature became impossible there. In his words, deep learning research "is done."
This departure fits a larger pattern among OpenAI staff. Several high-profile figures have left to pursue independent paths. The organization faced internal shifts, including leadership changes and a pivot toward commercialization.
Part of the Neo Labs Movement
Core Automation enters a wave of new ventures known as Neo Labs. These come from OpenAI alumni and similar figures. Examples include Thinking Machines Lab, headed by OpenAI's former CTO. Another is Safe Superintelligence, under the guidance of the ex-chief scientist.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
These labs hold a common position. They argue that true AI advances now require entirely new methods. Scaling existing approaches hits limits, so innovation demands different foundations.
Safe Superintelligence, for instance, launched in 2024 with backing from prominent investors. It focuses solely on building safe superintelligence. Thinking Machines Lab draws talent from top AI circles to rethink machine intelligence fundamentals.
Tworek's Core Automation aligns with this trend. It emphasizes automation within research itself. Public details come from CoreAuto, the lab's source.
The AI field sees rising interest in agent-based systems and self-improving labs. Core Automation positions itself at that intersection. By automating internal operations first, it aims to demonstrate practical gains from new algorithms and architectures.
OpenAI alumni have founded multiple startups since 2023. This includes ventures in AI safety, hardware, and novel paradigms. Core Automation adds to that list with its automation focus.
Broader Context in AI Research
Current AI progress often relies on transformers, introduced in 2017. These power models like those from OpenAI. Yet concerns grow over data limits and compute costs for further scaling.
New labs like Core Automation seek alternatives. They target better scaling laws and agent capabilities. Small teams using AI tools could accelerate discoveries.
Tworek's background equips him for this. His OpenAI tenure covered advanced model training and deployment. Now, he applies that to build automated research environments.
As of April 22, 2026, Core Automation begins operations. It promises updates on its automation efforts and algorithm developments.