
What happened
Jeff Dean, Google's longest-serving AI executive, is stepping down to launch Discovery Loop, a public benefit corporation co-founded with fellow Google researchers Sanjay Ghemawat, Quoc Le, and Oriol Vinyals. Dean will serve as CEO. The startup aims to use AI to automate and accelerate scientific research by running thousands of experiments simultaneously.
Why it matters
Discovery Loop plans to remove what it calls a "significant bottleneck" in science—the slow, sequential human iteration that has historically driven progress. By automating the experimental loop, the company seeks to deliver both higher quantity and higher quality experiments, potentially speeding the path to scientific breakthroughs. Dean told the New York Times he believes AI can "more fully automate what has traditionally been a very human-intensive experimental loop."
What to watch
The startup has received backing from Alphabet (Google's parent company), with the initial funding round co-led by Radical Ventures and Khosla Ventures; Kleiner Perkins, Lightspeed, and Doerr Capital also participated. The company is also exploring recursive self-improvement—using AI to help create more powerful AI—which would cut human iteration out of the research loop entirely.
Summaries like this, in your inbox every morning.
Jeff Dean's departure marks a significant moment in Google's AI leadership hierarchy. Dean joined Google in 1999 as the company's 30th employee and has been instrumental in building the infrastructure that powers Google Search—including its crawling, indexing, and query-serving systems—as well as playing a leading role in Google Gemini's multimodal models and the company's early AI research. His exit to launch a startup suggests that even within the world's largest AI lab, there is perceived opportunity in a more focused, specialized application of AI to scientific discovery.
Discovery Loop's core thesis addresses a long-standing friction in scientific and engineering work: the reliance on slow, sequential human-driven iteration. By automating experimental loops and running thousands of experiments simultaneously, the startup believes it can compress the timeline between hypothesis and breakthrough. The company's interest in recursive self-improvement—where AI generates better AI without constant human input—points to an even more ambitious vision of reducing the human role further. Dean's quote to the New York Times that AI can deliver "both a higher quantity and a higher quality of experiments" suggests he sees this not as a replacement for human judgment, but as a way to multiply the researcher's leverage across a much larger experimental surface.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Dell leads day 1 of the AI Cybersecurity and Resilience Summit on Oct
RapidCanvas raised $29 million in a Series B led by Vertex Ventures SEA & India, with Hitachi Ventures, Accel…
Veeam announced Veeam Intelligent ResOps (iResOps), with Microsoft 365 as the first generally available worklo…
NDC Minister Yeh Chun-hsien said Taiwan's economy is projected to grow 6.04% in 2027, as global AI-related cap…

CoreWeave shares tumbled almost 25% this week as its bond yields surged, then fell another 12% Friday, even af…

AIBotics completed its acquisition of NovaCore Labs' operating assets from PlayJamaica Limited, gaining 10 ser…
