Google’s Jeff Dean Leaves After 27 Years to Found ‘Discovery Loop’ to Automate Science

Some of Google’s most influential minds are striking out on their own. Jeff Dean, chief scientist at Google DeepMind, announced on X that he has co-founded Discovery Loop with longtime collaborators Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. The company’s mission is to automate machine learning, science, and engineering to accelerate discovery. Dean is leaving Google after 27 years.

The founding team reads like a history of modern computing. Dean and Ghemawat built the foundations of large-scale distributed systems — MapReduce, Bigtable, Spanner, and the Google File System — while Vinyals worked on AlphaStar and Gemini at DeepMind, and Le, a founding member of Google Brain, led seminal work on sequence-to-sequence learning and AutoML. The four have worked together for 14 to 30 years.

Discovery Loop’s idea is to automate the research loop itself: using high-octane algorithms to launch and iterate thousands of experiments at once, partially automating the process of ML, science, and engineering research. Dean is expected to serve as CEO.

Notably, Discovery Loop is structured as a Public Benefit Corporation (PBC) — the same governance path taken by frontier labs including OpenAI, Anthropic, xAI, and Ilya Sutskever’s SSI. Its seed round has not yet closed; it is co-led by Radical Ventures and Khosla Ventures with participation from Lightspeed, Kleiner Perkins, Doerr Capital, and — notably — Alphabet, the very company Dean is leaving. The startup declined to disclose the amount or valuation.

A rush of star-researcher labs, now aimed at automating discovery

Elite researchers spinning out of Big Tech to launch their own labs is a defining trend of this AI cycle — from Mira Murati’s Thinking Machines to Ilya Sutskever’s SSI. Discovery Loop targets one of the hottest niches within it: automating research itself, alongside a wave of “AI that does AI and science research” startups. With four names etched into computing history now betting on automated discovery, the idea has moved firmly into the frontier’s next front.