# AI pioneer Richard Sutton founds new research lab

> Source: <https://betakit.com/ai-pioneer-richard-sutton-founds-new-research-lab/>
> Published: 2026-07-14 17:22:08+00:00

Reinforcement learning pioneer and Turing Award winner Richard Sutton is striking out on his own, founding a new AI research lab focused on real-time learning that will reduce compute demands.

**The news: **On Monday, Sutton and his colleague Khurram Javed both announced they were leaving positions at [Keen Technologies](https://keenagi.com/)—another AI research firm founded by virtual reality-pioneer John Carmack—to found their own boutique firm called [Oak](https://oaklab.ai/)[ ](https://oaklab.ai/)[Lab](https://oaklab.ai/). Sutton and Javed’s new venture is described on Oak Lab’s website as focusing on developing intelligence from experiential learning, rather than large data sets, allowing AI to operate with significantly less compute demand.

**From the source:** “Like Keen … we at Oak Lab believe in reinforcement learning, and that intelligence is created and maintained from run-time experience,” Sutton wrote in a [post to X](https://x.com/RichardSSutton/status/2076663628301058329) yesterday. “But we think current deep learning methods are weak and inefficient, and need not more tweaks, but fundamentally new ideas.”

**Following the thread:** Sutton and Javed’s new lab is operating from the perspective of the duo’s “[big world hypothesis](https://openreview.net/pdf?id=Sv7DazuCn8),” which posits that the world is too large for any AI model to pre-learn everything. Oak Lab’s algorithms instead train on real-time experiential learning, rather than relying on the massive datasets that have been used to train most modern AI systems. Oak Lab’s website says the company’s algorithms learn without storing or replaying data, reducing compute and energy demands. The company has a “holy grail” goal of creating a trillion-parameter AI agent that can learn and plan on just 20 watts of energy—roughly as much as it would take to power a few lightbulbs.

**Final thought: **Sutton wrote the literal textbook on reinforcement learning—a method where AI agents learn through trial and error—and founded the University of Alberta’s Reinforcement Learning and Artificial Intelligence Lab. His venture with Javed, while still in [its earliest stages](https://x.com/kjaved_/status/2076663868160459214?s=46), holds the promise of tackling one of AI’s biggest challenges: rapidly scaling while contending with increasingly expensive demand for more compute. Much of the data centre boom seen in Canada and elsewhere has been spurred on by this demand, particularly as more enterprises and startups see [AI use as non-negotiable](https://betakit.com/building-with-ai-is-now-the-price-of-admission-for-software-startups-inovia-report-says/) and the [costs of tokenmaxxing soar.](https://www.cbc.ca/news/business/ai-spending-ending-tokenmaxxing-tokenomics-9.7237680) If Sutton is successful and timely, an effective and energy efficient AI could also lessen concerns around energy overconsumption and the environmental impact of AI development.

*BetaKit’s Prairies reporting is **funded in part by YEGAF**, a not-for-profit dedicated to amplifying business stories in Alberta.*

*Feature image courtesy the Alberta Machine Intelligence Institute. Photo by Chris Onciul.*