Reinforcement learning pioneer Richard Sutton, 68, announced on July 14 via social platform X that he and collaborator Khurram Javed have departed Keen Technologies—the AI company founded by John Carmack—to establish a new firm called Oak Lab, registered in Canada. The startup is dedicated to building an AI system fundamentally different from current LLMs: an agent capable of continuous learning from its own experience and evolving in real time. Sutton stated bluntly that current deep learning methods suffer from fundamental deficiencies requiring complete conceptual reconstruction. Oak Lab's founding motivation connects to Sutton's 2019 essay "The Bitter Lesson," in which he reviewed 70 years of AI research and argued that general methods leveraging massive computation ultimately surpass approaches embedding human domain knowledge. Sutton rejects the interpretation that LLMs embody this lesson, arguing in a September 2025 interview that LLMs learn from training data, not from experience, and that pouring vast quantities of human text into a model is still a path dependent on human knowledge.