Google Chief Scientist Jeff Dean Departs After 27 Years to Launch AI Startup Discovery Loop

Jeff Dean, the legendary computer scientist who spent nearly three decades building the foundational infrastructure of Google and modern artificial intelligence, is leaving the company to launch a new venture. Dean, who served as the chief scientist across Google DeepMind and Google Research, announced his departure on August 5, 2026. He is stepping away to co-found Discovery Loop, an ambitious AI startup aimed at automating the experimental processes of scientific research.[1][3]

Dean is not leaving alone. He is taking three of Google's most prominent AI researchers with him to build the new company. The founding team includes Sanjay Ghemawat, a Google Senior Fellow and Dean's collaborator of over two decades; Oriol Vinyals, the vice president of research at Google DeepMind and a technical lead for the Gemini models; and Quoc Le, a co-founder of Google Brain who pioneered automated machine learning.[2][3]

Discovery Loop is structured as a public benefit corporation with a mission to accelerate human progress by fundamentally changing how research is conducted. The startup intends to build AI systems capable of running the entire "experimental loop" of the scientific method: proposing an experiment, implementing it, evaluating the results, and iterating.[1][2]

By automating these steps, the founders believe they can execute thousands of experiments in parallel, operating at a scale and speed that sequential human effort cannot match. The company plans to initially focus its automated loops on machine learning research itself, before expanding into highly complex domains such as chip design, biology, drug discovery, and materials science.[2][3]

The departure marks the end of an era for Google. Dean joined the company in 1999 as its 30th employee, working out of a small office above a Palo Alto storefront when the startup had just 25 people. Over the next 27 years, he watched the workforce swell to over 190,000 employees while he personally co-created the backend architecture that allowed Google to scale globally.[1]

Alongside Ghemawat, Dean designed foundational distributed computing systems like MapReduce, BigTable, and Spanner. These systems not only kept Google Search running during massive traffic surges but also became the blueprint for how the entire tech industry, including Amazon and Meta, handles global data processing today.

Dean later pivoted to artificial intelligence, co-founding the Google Brain research lab in 2011. Under his leadership, the team developed TensorFlow, the open-source machine learning library that became an industry standard, and laid the groundwork for the neural networks that power today's generative AI boom. Most recently, he served as a technical co-lead for Google's flagship Gemini models.

Despite the massive loss of talent, the split appears highly collaborative. Google CEO Sundar Pichai publicly congratulated Dean and Ghemawat, confirming that Google will serve as a founding investor and the primary Cloud partner for Discovery Loop. The startup has also secured backing from prominent venture capital firms, with Radical Ventures and Khosla Ventures co-leading its initial funding round.[1]

The exit has triggered a broader reshuffle of Google's AI leadership. Demis Hassabis, the co-founder and CEO of Google DeepMind, is stepping down from his day-to-day role to become the Chief Scientist of Alphabet and Chair of Google DeepMind. Meanwhile, Koray Kavukcuoglu has been promoted to Senior Vice President of Google DeepMind, where he will oversee the ongoing development of the Gemini models and frontier AI products.[3]

Discovery Loop's launch reflects a growing trend among the pioneers of the current AI wave. Rather than remaining within massive corporate structures to build general-purpose chatbots, elite researchers are increasingly spinning out agile startups focused on specialized, high-impact applications. If successful, Discovery Loop's automated scientific method could unlock breakthroughs in medicine and engineering that human scientists might never crack on their own.[3]