{"slug": "turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning", "title": "Turing Award Winner Rich Sutton Launches Oak Lab to Build Continuously Learning AI Agents", "summary": "Turing Award winner Richard Sutton and co-founder Khurram Javed left John Carmack's Keen Technologies to launch Oak Lab, a Canadian AI startup building continuously learning AI agents based on Sutton's OaK architecture. The company aims to create agents that learn in real time without storing or replaying data, with a long-term goal of a trillion-parameter agent running on 20 watts.", "body_md": "# Turing Award Winner Rich Sutton Launches Oak Lab to Build Continuously Learning AI Agents\n\n- Turing Award winner Richard Sutton and co-founder Khurram Javed left John Carmack's Keen Technologies to launch Oak Lab, a Canadian AI startup\n[[1]](https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/) - The company will build on Sutton's OaK (Options and Knowledge) architecture, which aims to create agents that learn in real time without storing or replaying data\n[[2]](https://thelogic.co/briefing/ai-pioneer-richard-sutton-launches-startup-to-build-always-learning-agents/) - Sutton has called current deep learning methods \"weak and inefficient\" and argues they need \"fundamentally new ideas\" rather than incremental improvements\n[[1]](https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/) - The long-term goal is an agent with a trillion parameters that learns and plans in real time using just 20 watts of energy\n[[1]](https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/) - No funding amount or investors have been disclosed; the team has been described as small and focused\n[[3]](https://digg.com/tech/5p0hdvmr)\n\nRichard Sutton, the 2024 Turing Award winner widely credited as the father of modern reinforcement learning, has founded a new AI startup called Oak Lab. Sutton and co-founder Khurram Javed, a former PhD student, both departed John Carmack's Dallas-based Keen Technologies last month to launch the Canadian-incorporated company [[1]](https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/) [2].\n\nOak Lab will pursue what Sutton considers a fundamentally different approach to building intelligent systems — one rooted in his OaK (Options and Knowledge) architecture, which he has presented at conferences including RLC 2025 and MIT CSAIL's Dertouzos Distinguished Lecture series [[4]](https://galtay.github.io/blog/sutton-on-oak-at-rlc-2025/) [5]. The architecture is designed to produce agents that learn continuously from their own experience rather than from pre-collected datasets, a sharp rebuke of the large-scale pretraining paradigm that dominates commercial AI today.\n\nSutton, a University of Alberta computing science professor who helped pioneer temporal difference learning and policy gradient methods, has grown increasingly vocal in his criticism of the current AI industry. He has argued that generative AI excels at imitation but cannot independently evaluate its outputs, limiting its usefulness for scientific discovery [1].\n\n## The Startup\n\nOak Lab represents Sutton's return to independent research after a turbulent few years. Google DeepMind closed the Edmonton lab he helped lead in 2023, after which he joined Carmack's Keen Technologies in September of that year [2]. His departure from Keen — along with Javed — signals a philosophical split: while Oak Lab shares Keen's belief in reinforcement learning and runtime experience, Sutton has said current deep learning methods require \"fundamentally new ideas and a thorough reworking\" for ambitious AI goals\n\n.\n\n[[1]](https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/)The company has not disclosed a funding amount, specific investors, or a precise team size beyond describing itself as a \"small, focused team.\" Sutton and Javed have promised to publish technical papers \"in the coming weeks and months\" [3].\n\n## The OaK Architecture\n\nAt the core of Oak Lab's work is the OaK architecture — Options and Knowledge — originally developed as part of The Alberta Plan for AI Research by Sutton, Mike Bowling, and Patrick Pilarski [4]. The system is built around three principles: domain generality (containing nothing domain-specific), experiential learning (growing from runtime experience rather than design-time engineering), and open-endedness (unlimited sophistication within computational constraints).\n\nTechnically, OaK uses a hierarchical reinforcement learning framework where an \"option\" consists of a policy mapping states to action probability distributions and a termination condition. The architecture runs parallel processes for learning policies and value functions, generating new state features, creating subproblems, learning transition models, and planning [4]. All components learn continually, and each learned weight has a dedicated step-size parameter that is meta-learned using online cross-validation.\n\nSutton's ultimate vision is ambitious: \"an agent with a trillion parameters that learns and plans in real time with 20 watts of energy\" — roughly the power consumption of a human brain [1]. The near-term plan is more modest: demonstrate the limitations of existing methods in simple settings, develop domain-independent algorithms, then scale\n\n.\n\n[[3]](https://digg.com/tech/5p0hdvmr)## Why It Matters\n\nSutton's critique carries unusual weight in the AI research community. His 2019 essay \"The Bitter Lesson\" — which argued that general methods leveraging computation ultimately outperform approaches that try to encode human knowledge — became one of the most cited pieces of AI philosophy in the deep learning era. His new venture effectively argues that today's LLM-centric industry has ignored the essay's deeper implications.\n\nThe launch also highlights a growing counter-movement within AI research. While companies like OpenAI, Anthropic, and Google DeepMind have invested billions in scaling transformer-based models trained on massive static datasets, a cohort of researchers — including Sutton — argues this paradigm hits fundamental limits for producing agents that can reason, plan, and adapt in real time [1].\n\nSutton has described the goal as building something closer to \"a baby learning in its first year\" than current AI systems, which he characterizes as sophisticated pattern matchers that lack genuine understanding of cause and effect [3].\n\n## What's Next\n\nOak Lab plans to build a prototype of the complete OaK architecture \"within the next few years,\" according to statements from the founders [3]. The immediate focus is on publishing research that demonstrates the shortcomings of existing methods in controlled environments before scaling to larger systems.\n\nThe startup enters a competitive landscape for RL-focused AI companies, though few have Sutton's academic pedigree. His departure from Keen Technologies also raises questions about Carmack's venture, which has not publicly commented on the split [2]. Sutton remains a professor at the University of Alberta and a fellow of the Alberta Machine Intelligence Institute (Amii)\n\n.\n\n[[6]](https://www.amii.ca/people/richard-s-sutton)## Further sources\n\n[[1] The Decoder — Turing Award winner Rich Sutton founds Oak Lab to build AI agents… ↗](https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/)\n\n[[2] The Logic — AI pioneer Richard Sutton launches startup to build always-learning… ↗](https://thelogic.co/briefing/ai-pioneer-richard-sutton-launches-startup-to-build-always-learning-agents/)\n\n[[3] Digg — Oak Lab startup details and founder statements ↗](https://digg.com/tech/5p0hdvmr)\n\n[[4] Blog summary — Richard Sutton on OaK at RLC 2025 ↗](https://galtay.github.io/blog/sutton-on-oak-at-rlc-2025/)\n\n[[5] MIT CSAIL — Dertouzos Distinguished Lecture: The Oak Architecture ↗](https://www.csail.mit.edu/event/dertouzos-distinguished-lecture-oak-architecture-vision-superintelligence-experience)\n\n[[6] Alberta Machine Intelligence Institute — Richard S. Sutton profile ↗](https://www.amii.ca/people/richard-s-sutton)\n\nThe stories that matter, in one email. Free — unsubscribe anytime.", "url": "https://wpnews.pro/news/turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning", "canonical_source": "https://mlq.ai/news/turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning-ai-agents/", "published_at": "2026-07-14 16:48:17.521044+00:00", "updated_at": "2026-07-14 16:48:19.666197+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-startups", "ai-research"], "entities": ["Richard Sutton", "Khurram Javed", "John Carmack", "Keen Technologies", "Oak Lab", "Google DeepMind", "University of Alberta", "MIT CSAIL"], "alternates": {"html": "https://wpnews.pro/news/turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning", "markdown": "https://wpnews.pro/news/turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning.md", "text": "https://wpnews.pro/news/turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning.txt", "jsonld": "https://wpnews.pro/news/turing-award-winner-rich-sutton-launches-oak-lab-to-build-continuously-learning.jsonld"}}