Hey there! 👋
Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters, now also live at savvymonknewsletter.com. Google's chief scientist is stepping away after twenty seven years to build a company that tries to automate scientific research itself, and he is not going alone.
Let's get into it.
TODAY'S DEEP DIVE
Jeff Dean's Startup Is Built to Run Thousands of Experiments a Day Without a Human in the Loop
On 5 August 2026, Google announced that Jeff Dean, its chief scientist and thirtieth employee, is leaving the company he joined in 1999 to co-found Discovery Loop, a startup built to automate the process of scientific and engineering research. The announcement came bundled with a wider leadership reshuffle at the top of Google's AI organisation, and together the two pieces of news reset who runs what across the company's entire AI stack.
The Departure
Dean spent nearly three decades at Google, where he helped build the crawling and indexing systems behind search, co-founded Google Brain, and later played a leading role in the development of Gemini's multimodal models. He is fifty eight.
Speaking at the University of Washington's computer science commencement in June, Dean described getting the itch to join a startup back in 1999, when Google itself had roughly twenty employees.
Nearly three decades on, he said he has the itch again. At Discovery Loop he will serve as chief executive.
Who Is Going With Him and other Google Veterans
Three other long-serving Google veterans are joining Dean as co-founders. Sanjay Ghemawat, a Google senior fellow, co-created MapReduce, the Google File System, Bigtable, and Spanner, the infrastructure that underpinned internet-scale computing for two decades.
(From left) Oriol Vinyals, Sanjay Ghemawat, Jeff Dean, and Quoc Le | Pic from Discovery Loop
Quoc Le, a founding member of Google Brain, co-invented sequence-to-sequence learning, a technique that sits behind much of modern language model architecture.
Oriol Vinyals, a vice president of research at Google DeepMind and a technical co-lead on Gemini, is also known for building AlphaStar, the system that beat professional players at the video game StarCraft II. Between them the four founders have worked together anywhere from fourteen to thirty years, and Ghemawat has collaborated directly with Dean for more than two decades.
What Discovery Loop Is Building
Discovery Loop is structured as a Delaware public benefit corporation based in Palo Alto, and its goal is to automate what the founders call the experimental loop, the repeated cycle of proposing a hypothesis, running an experiment, evaluating the result, and iterating again.
Ghemawat has said the team wants infrastructure built around different requirements than what Google maintains for its consumer and advertising products. At launch the company plans to focus on automating machine learning research and engineering itself, before expanding into areas including hardware design, drug discovery, and clean energy.
Our mission is straightforward: we are building AI solutions that can automatically solve important problems in machine learning, science, and engineering. By advancing the pace at which we conduct engineering and scientific discovery, we can bring the benefits of science and technology to the world much faster. Ultimately, our goal is to build AI systems that act as a deeply positive, empowering force for humanity, delivering technology solutions that improve people's lives on a global scale.
The founders have also pointed to recursive self improvement, the idea of using AI to help build more capable AI, as a longer term direction. In a joint statement, the team framed the aim as moving AI beyond answering questions toward making discoveries, with the goal of delivering the benefits of new technologies to the world sooner than the current slower, human-paced research cycle allows.
The Money Behind It
Google is a founding investor in Discovery Loop and will supply cloud computing power for at least the first year. The startup's initial funding round is being co-led by Radical Ventures and Khosla Ventures, with additional participation from Kleiner Perkins, Lightspeed, and Doerr Capital. Alphabet is also among the investors backing the round.
The Reshuffle at Google
The same announcement confirmed that Demis Hassabis is stepping back from his role as chief executive of Google DeepMind to become the unit's chairman and Alphabet's chief scientist, a new title at the parent company level.
Hassabis will continue to lead Isomorphic Labs, Alphabet's AI-driven drug discovery spinoff, alongside his new role. Taking over Hassabis's operational responsibilities at DeepMind is Koray Kavukcuoglu, currently the unit's chief technology officer, who is being promoted to senior vice president and will report directly to Google chief executive Sundar Pichai.
What It Means for Google
The market read the news as a genuine loss. Alphabet's stock fell by roughly five per cent during trading on the day of the announcement, a drop worth an estimated one hundred sixty to two hundred billion dollars in market value. It is also the most senior departure of its kind Google has faced in this AI cycle, and the first time a Google AI leader of Dean's stature has left to start a company rather than join a rival lab.
Sundar Pichai reiterated that Google would continue supporting Discovery Loop through its cloud partnership even as the company loses four of its most experienced technical leaders in one move.
The Bottom Line
Dean is not leaving Google for a competitor. He is betting that the next big unlock in AI is not a better chatbot but a system that can run science itself, faster and at a scale no team of human researchers could match.
Google gets to keep a stake in that bet as an investor, but it also loses the person who spent twenty seven years building much of the infrastructure the company still runs on, along with three of the collaborators who built it alongside him.
AI PROMPT OF THE DAY
Category: Research and Strategy
"Act as a research strategist helping me think through how automation could speed up experimentation in [describe your field or project]. Walk me through which parts of my current process are the slowest to iterate on, which of those steps could realistically be automated with current AI tools versus which still need human judgement, and what a pilot version of a faster experimentation loop would look like for my specific work."
ONE LAST THING
Every few years, a departure like this becomes a marker of how seriously the industry is taking a particular bet. Dean spent twenty seven years building the infrastructure other people's AI systems now run on, and he chose to leave that behind for a much narrower, much harder problem, teaching machines to run the scientific method with no one holding the loop steady.
Whether that bet pays off will not be clear for years. But the fact that three of his closest collaborators walked out with him is its own kind of signal.
Hit reply, I read every response.
See you in the next one.
— Vivek
P.S. Know a developer or founder who follows AI leadership moves closely? Forward this to them.