Jeff Dean Leaves Google to Automate the Scientific Method With Discovery Loop

Jeff Dean, Google's chief scientist, is leaving the company after nearly 27 years to co-found Discovery Loop, a public benefit corporation aimed at automating the experimental loops of the scientific method. The startup, announced on August 5, 2026, will be backed by Google as a founding investor and Cloud partner, and will initially focus on automating machine learning research before expanding into chip design, biology, drug discovery, and materials science. Co-founders include Sanjay Ghemawat, Oriol Vinyals, and Quoc Le, all senior Google AI figures.

AI Models & Platforms https://www.unite.ai/series/artificial-intelligence/ Jeff Dean Leaves Google to Automate the Scientific Method With Discovery Loop Add Unite.AI to your preferred sources on Google https://www.google.com/preferences/source?q=unite.ai Jeff Dean, Google’s chief scientist and one of the central figures in modern AI infrastructure, is leaving the company after nearly 27 years to co-found Discovery Loop https://www.discoveryloop.com/ , a startup whose stated goal is to automate the experimental loops of scientific research itself. The departure was announced on August 5, 2026 https://blog.google/company-news/inside-google/message-ceo/next-chapter-ai-momentum/ as part of a broader reshuffle of Google’s AI leadership, with the company confirming it will back the new venture as a founding investor and Cloud partner. Dean is not leaving alone. Three other senior Google AI figures are co-founding the company with him: Sanjay Ghemawat, a Google Senior Fellow and Dean’s collaborator on much of the company’s foundational computing infrastructure; Oriol Vinyals, VP of research at Google DeepMind and a technical lead on Gemini; and Quoc Le, a co-founder of Google Brain and the scientist behind Google’s AutoML-Zero project. Wired’s Steven Levy, who interviewed the four founders ahead of the announcement https://www.wired.com/story/jeff-dean-google-discovery-loop-startup/ , reports the idea came together only a few weeks ago. The company is incorporated as a public benefit corporation. Its mission, as described on its own site, is to build systems that can run the full experimental loop of the scientific method: proposing an experiment, implementing and running it, evaluating the results, and iterating, at a scale and speed that sequential human effort cannot match. The founders’ plan is to run thousands of such loops in parallel, initially pointed at machine learning research itself before expanding into domains including chip design, biology, drug discovery, and materials science. What Discovery Loop actually plans to build The mechanism, per the company’s own description, has three stages. First, it will focus on automating machine learning research and engineering, using frontier AI models and large-scale compute to propose, run, and learn from evaluations. Second, it will act as its own first customer, using those automated capabilities to optimize its own technology stack. Third, it intends to generalize to any learning loop with measurable outcomes in science and engineering, citing the National Academy of Engineering’s Grand Challenges as its eventual target class, including engineering better medicines, advancing health informatics, and making solar energy economical. Le, who pioneered techniques for automating model design at Google, framed the ambition in terms of what automated loops might find in AI itself. “I’m very excited about automating machine learning,” he told Wired. “It might be that we will discover a different transformer architecture.” Vinyals pointed to the specific capability gap the effort rests on: current models are not strong at generating genuinely new ideas to test. “One of the things that we’ll be obviously very focused on is how these models come up with new ideas to try,” he said. “That’s not something that currently they’re super strong at.” The founders told Wired that early systems will co-develop ideas with humans, with deeper automation as the goal. The founders’ record, and Google’s stake in what comes next The founding team’s collective track record is the core asset. Discovery Loop’s site describes the four as representing three of the most-cited researchers in AI and two of the most-cited in distributed systems, with contributions spanning Google Search, Google Translate, the Google File System, MapReduce, BigTable, Spanner, TensorFlow, Pathways, TPUs, AlphaChip, AlphaStar, AlphaCode, AlphaFold, Gemini, sequence-to-sequence models, and chain-of-thought reasoning, among others. Dean and Ghemawat built much of the infrastructure that Google’s early search and advertising businesses ran on; Vinyals and Le were central to the neural network era that followed. Unite.AI has tracked Google’s recent AI model releases https://www.unite.ai/google-ships-three-gemini-flash-models-as-its-flagship-slips/ , a line of work Vinyals was directly helping to lead. The departures land inside a wider reorganization of Google’s AI leadership, detailed in messages from CEO Sundar Pichai and DeepMind co-founder Demis Hassabis. Hassabis is stepping back from day-to-day leadership of Google DeepMind to become its chair and Alphabet’s chief scientist, while Koray Kavukcuoglu, the unit’s CTO and Google’s chief AI architect, takes over as SVP of Google DeepMind, overseeing Gemini model development, frontier AI research, and the Gemini app and developer teams. Pichai’s message disclosed that the Gemini app has passed 950 million monthly users and that Gemma models have surpassed 900 million downloads. Google’s relationship with Discovery Loop goes beyond well-wishes. The company is a founding investor, will serve as the startup’s Cloud partner, and plans to collaborate on a research framework for ML systems and infrastructure. Wired reported that Google will also supply compute power for the venture’s first year. “Over 27 years, Jeff and Sanjay helped to drive some of the most significant technology transitions, from our early search infrastructure to the neural networks that helped create the modern AI era,” Pichai said in a statement. “We’ll continue to work with Discovery Loop as a founding investor and Cloud partner, and collaborate on a research framework for ML systems and related infrastructure advances.” On the funding side, Wired reported that Khosla Ventures and Radical Ventures are among the backers, with Radical managing partner Jordan Jacobs joining the board; the founders are not disclosing the round’s size or valuation. The venture thesis, as Khosla put it to Wired, is a shift in what AI is for in research: humans have been using AI to do research, he said, while the premise here is that AI is the researcher. What happens next The near-term observables are concrete. Discovery Loop is recruiting a lean, in-person founding team through its open roles page https://jobs.ashbyhq.com/Discovery-Loop , having started with no hires and no office. Its first executable milestone is the automated machine learning loop it plans to turn on its own stack, and its first year of operation runs on Google-provided compute under the Cloud partnership. The stated expansion path, from ML research into chip design, biology, drug discovery, and materials, will be measurable against the company’s own criterion: learning loops with outcomes that can be scored. Whether the approach produces a result the field recognizes as a discovery, rather than an optimization, is the question the company’s structure is built to answer.