Jeff Dean, Google's chief scientist and the 30th person ever hired by the company, is leaving after 27 years to start a new company called Discovery Loop. The company will use artificial intelligence to speed up scientific research. Dean will be the CEO. CNBC WSJ

Dean started Discovery Loop with three other senior Google AI researchers: Sanjay Ghemawat, Quoc Le, and Oriol Vinyals. The company is set up as a public benefit corporation, meaning it is legally required to serve a public good, not just make money for investors. In this case, that public good is accelerating scientific research. The company's goal is to use AI to run thousands of experiments at the same time, partially automating the work that scientists do. TechCrunch

Discovery Loop also plans to explore an idea called recursive self-improvement, which means using AI to help build better AI systems. The company's first funding round is co-led by Radical Ventures and Khosla Ventures, with Kleiner Perkins, Lightspeed, and Doerr Capital also participating. Alphabet, Google's parent company, has also provided financial backing. TechCrunch

Dean confirmed the news on his official X account. The startup uses the handle @DiscoLoopAI. X / Jeff Dean

Google's CEO acknowledged the departure in a blog post, confirming that Dean is leaving after 27 years and saying that he wants to try something new and that Google is excited to support him. Google Blog

The departure is part of a bigger shuffle of AI leadership at Google. On the same day, Demis Hassabis stepped down as CEO of DeepMind, Google's AI division, to take on a chief scientist role. Reuters CNBC

Dean joined Google in 1999 as the company's 30th employee. His work spans the architecture of Google Search itself, the systems that crawl the web and organize results, through to the AI models behind Google Gemini. He played a leading role in Google's early AI research, helping build the foundation that grew into Google DeepMind's current lineup of AI models. TechCrunch

His interest in AI-driven scientific discovery predates Discovery Loop. Dean was an early investor in Profluent, a company using AI for drug discovery, backed in 2024. He also invested in Periodic Labs, a startup founded by OpenAI and Google Brain researchers that raised a $300M venture round in 2025. Earlier, in 2025, Dean was linked to a $100M fund for AI researchers, pledged by Databricks and Perplexity co-founders, that would award $3 million annually over five years, including for AI applied to scientific discovery. TechCrunch (Profluent) TechCrunch (Periodic Labs) TechCrunch ($100M fund)

Dean also co-authored a 2022 study that pushed back against the idea that AI is inherently bad for the environment. TechCrunch

The four co-founders bring different skills to the table. Ghemawat is a longtime Google engineer who worked with Dean on the core systems behind Google Search. Quoc Le was a key contributor to a technique called sequence-to-sequence learning, which teaches a computer to take one series of items, like words in a sentence, and produce another series, like a translation. This work is the basis for modern machine translation. Oriol Vinyals contributed foundational research in how computers learn through trial and error with rewards, a method similar to how you might train a dog, during his time at DeepMind. Together, the group covers the range of skills that Discovery Loop's vision of running many experiments at once would require. TechCrunch

The technical ambition is large. Running thousands of experiments at the same time under AI control implies a system that can come up with ideas to test, design the steps to test them, run those steps (probably through automated lab equipment or computer simulations), check the results, and feed those results back to improve the AI, all at machine speed. The recursive self-improvement part adds a second layer: the AI systems getting better at improving the AI systems that drive the research. If this works, it could shrink research cycles that currently take months or years down to potentially hours.

The broader context here is a shift in where the top people in AI are choosing to work. Dean's departure from Google, along with the DeepMind leadership change the same day, leaves Google's AI organization in a materially different shape than it was a week ago. Alphabet's decision to back Discovery Loop financially suggests that Google sees more value in keeping a relationship with this team as an outside company than in holding onto them as employees. That choice may reflect the limits that large companies face when pursuing research that may not pay off for a long time, or the difficulty of justifying the high computing costs of running thousands of experiments within a publicly traded company's budget.

For those watching the AI field, Discovery Loop's public benefit corporation structure is worth noting. It places a legal obligation on the company to balance investor returns with a defined public benefit, in this case accelerating scientific research. Whether that legal structure actually guides decisions when expensive infrastructure and investor expectations come into play is an open question. But it does put Discovery Loop in a different category from a typical startup, and it follows a pattern that several AI safety and research organizations have adopted.

The funding lineup, Radical Ventures, Khosla Ventures, Kleiner Perkins, Lightspeed, Doerr Capital, plus Alphabet, is a notably wide group for a company just starting out. Alphabet's participation is the detail that stands out. A company losing its chief scientist to a startup and then investing in that same startup is an unusual sequence. It suggests that Google's leadership decided Discovery Loop's research direction was better pursued outside the company, with Google positioned as an investor rather than an employer.

Dean's career at Google covers the full arc of modern computing: from the systems that made web search work at a global scale, through the TensorFlow era that made deep learning accessible to everyday developers, to the Gemini AI models currently competing with the best in the field. The question now is whether that same instinct for building large-scale systems, applied to automating scientific research instead of search, produces a comparable leap forward.