Jeff Dean and Three Top Google AI Researchers Left to Automate Science. The Startup Is Called Discovery Loop.

What Discovery Loop Aims to Do

Discovery Loop is structured as a public benefit corporation (PBC) . Its mission is to build AI systems that can automate the full experimental loop of scientific research: forming hypotheses, designing experiments, running them at massive scale, analyzing results, and iterating — all without constant human intervention .

The startup plans to use "high-octane algorithms" to initiate and iterate thousands of experiments simultaneously, accelerating discovery in fields like drug development, materials science, and chip design . Dean has described the vision as fully computerizing the scientific method in domains where the loop can be digitized, moving from "answering questions to discovering new knowledge" .

The founders wrote on their website: "Imagine a future where a handful of people can conduct scientific research and engineering tasks much more rapidly, and with higher quality, than massive teams of scientists and engineers do today" .

A Founding Investor You Wouldn't Expect: Alphabet

Despite losing its four top builders, Alphabet — Google's parent company — is a founding investor in Discovery Loop and is supplying cloud computing resources for at least the first year . The arrangement means Google retains a financial stake and a cloud revenue stream even as it loses the talent. Sundar Pichai publicly congratulated Dean on "an incredible 27-year run" and framed the exits as amicable .

The seed round is co-led by Radical Ventures and Khosla Ventures, with additional participation from Lightspeed, Kleiner Perkins, and Doerr Capital .

Who Left and Why It's a "Rolling Stones-Level" Loss

The four founders collectively represent a combined century of experience building the foundational infrastructure and AI systems that power modern Google .

• Jeff Dean joined Google in 1999 as roughly the 30th employee and helped build MapReduce, Bigtable, TensorFlow, and the Gemini model family — the foundational infrastructure of modern Google and much of the internet .
• Sanjay Ghemawat is a Senior Fellow and co-creator of MapReduce, Google File System, Bigtable, and Spanner — systems that underpin Google's core operations .
• Oriol Vinyals is a co-lead of the Gemini model family and a key architect of Google's modern AI capabilities .
• Quoc Le is a leading AI researcher and a founding member of Google Brain who helped develop sequence-to-sequence learning and many of Google's deep learning breakthroughs .

The four had worked together for between 14 and 30 years . Bloomberg reported that the departure jolted morale inside Google's AI ranks and raised new questions about the company's ability to retain top talent as the AI talent war intensifies .

Will It Work?

Discovery Loop's wager is ambitious: compress the iterative cycle of the scientific method — hypothesize, experiment, evaluate, repeat — from months or years into hours or days by running loops in parallel at machine speed . The startup says it will initially focus on automating machine learning research and engineering before expanding to other sciences .

Critically, the founders acknowledge that not all science can be digitized yet. Their plan is to start with domains where the experimental loop can be captured and run computationally, then expand outward .

As one coverage piece summarized: "The startup is also interested in using AI to help create more powerful AI (a process known as recursive self-improvement), which would cut human iteration out of the loop entirely" .