
Google "Foundation Veteran" Jeff Dean Leaves to Launch Startup, New Company Targets AI4S
Author: Hualin Wuwang
In 1999, when Jeff Dean joined Google, the company had only 20 people, and the office was upstairs from a T-Mobile store in Palo Alto. 27 years later, he is Google's Chief Scientist, behind him are MapReduce, Bigtable, Spanner, TensorFlow, TPU, Gemini—almost every piece of core infrastructure supporting Google's operations bears his name.
On August 5, Jeff Dean announced his departure from Google to co-found Discovery Loop with three long-time collaborators. On the day the news was announced, Alphabet's stock price fell by approximately 5%.
The market's reaction was honest. What left was not an executive, but a load-bearing pillar.
01 Building Infrastructure for AI Research
The goal of Discovery Loop, in the founding team's own words, is to "automate the experimental loop of machine learning, science, and engineering."
This statement needs to be broken down to be understood. The core process of scientific research is actually a loop—propose a hypothesis, design an experiment, execute the experiment, verify results, then propose new hypotheses based on the results, and start another round. Humans have done science for hundreds of years relying on this loop. The problem is that this loop relies heavily on manual labor, and speed is limited by human energy and time.
What Discovery Loop aims to do is to use AI models and large-scale computing power to automate this loop, running thousands of experiments in parallel.

Discovery Loop team, all from Google tech experts | Image source: x
The specific roadmap is divided into three steps:
First step, start with machine learning research itself—use AI to automatically conduct AI research, automatically propose experimental plans, automatically run evaluations, and automatically iterate models. This is the most pragmatic starting point because the feedback loop for ML research is the shortest, evaluation metrics are the clearest, and data is most readily available.
Second step, use this capability to optimize their own tech stack, making themselves their own first customer.
Third step, expand to broader science and engineering fields, including hardware design, drug discovery, and clean energy.
The company is registered as a Public Benefit Corporation, following the same governance path as OpenAI, Anthropic, xAI, and Ilya Sutskever's SSI.
The company's seed round was jointly led by Radical Ventures and Khosla Ventures, with Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet itself participating in the investment. The amount and valuation were not disclosed.
It is worth noting that Google not only invested money but will also serve as a cloud partner, providing at least the first year of computing power support for Discovery Loop.
This detail is significant—it is not common in Silicon Valley for a former employer to willingly provide computing resources to support an external company founded by its core employees.
02 "Infrastructure Living Fossil"
The founding team of Discovery Loop consists of only four people, but these four people combined are almost a brief history of modern computing and AI infrastructure.
Jeff Dean, Google's 30th employee, a 27-year veteran. Together with Sanjay Ghemawat, he created MapReduce, Bigtable, Google File System, and Spanner—these are not fancy application-layer products, but the underlying "operating system" that enables Google to process global-scale data. Without these things, there would be no Google Search, no Gmail, no Google Ads. Later he led Google Brain, drove the development of TensorFlow and the design of TPU chips, and became Chief Scientist after the merger of Google Brain and DeepMind in 2023.
Sanjay Ghemawat, Google Senior Fellow, Dean's partner for over 20 years. Inside Google, Dean and Ghemawat are a legendary combination—the two almost always appear together on the most important papers and projects. Some joke that they share a brain.
Oriol Vinyals, Google DeepMind Vice President of Research, one of the technical leads for Gemini. During his time at DeepMind, he participated in landmark projects such as AlphaStar (the AI that defeated StarCraft professional players) and AlphaCode (code generation).
Quoc Le, co-founder of Google Brain, one of the pioneers of AutoML, made foundational contributions in the field of sequence-to-sequence learning.

Jeff Dean's departure announcement posted on x | Image source: x
Content of Jeff's departure announcement:
Tomorrow will be my last day at Google, having spent 27 years here, witnessing it grow from 25 people to over 190,000, this has been an amazing journey. Below is a note I shared with many people inside Google today. Excerpts follow:
It is absolutely an honor to work with you and help build some of the most widely used and impactful products in history. When I was a child, I dreamed of helping build software used by many people, and now Google has 13 products used by over a billion people (too incredible!).
Our work has had a huge impact on the world, and I have had the privilege to collaborate and build friendships with many colleagues whom I deeply admire, respect, and enjoy being with. Every time I see people around the world using our products to find information, process emails, translate documents, watch videos, learn new things, navigate and understand the physical world, browse the web, use mobile phones, run large-scale computations on our infrastructure, ride in autonomous vehicles, or perform complex tasks with the help of our AI systems, I still feel immense joy. I hope you can all share this joy, because this is a shared achievement! Thanks to all colleagues at Google over the years!
Now I am excited to start @DiscoLoopAI with my old friends and colleagues @Sanjay_Ghemawat, @OriolVinyalsML, and @quocleix.
The four individuals have collaborated for periods ranging from 14 to 30 years. The introduction on the Discovery Loop official website states that this team includes three of the most cited researchers in the AI field, and two of the most cited researchers in the distributed systems field.
If you look closely at this resume, you will find a keyword running through Dean's entire career—infrastructure. What he did at Google was never user-facing products, but the layer that allows others' work to be systematically amplified. MapReduce amplified engineers' ability to process data, TensorFlow amplified researchers' ability to train models, and TPU amplified the computing power supply for the entire industry.
Now, what he wants to amplify is scientists' ability to conduct research. The tools have changed, but the role has not. The essence of Discovery Loop is "infrastructure for scientific discovery."
03 AI Migration Wave
Jeff Dean's departure is not an isolated event.
Just this summer, Google DeepMind experienced an unprecedented wave of talent loss. In June, Nobel Prize winner John Jumper (core of the AlphaFold team) left to join Anthropic, and Transformer paper co-author Noam Shazeer also left shortly before. In just one week in June, five DeepMind senior researchers went to Anthropic and OpenAI respectively, and Alphabet's market cap once evaporated by approximately $270 billion.
Now, even the person who built MapReduce has left.
And this is not just Google's problem. Mira Murati left OpenAI in 2024 to found Thinking Machines Lab, raising $2 billion in 5 months with a valuation of $12 billion. Ilya Sutskever founded SSI after leaving OpenAI, focusing on superintelligence safety. Andrej Karpathy also joined Anthropic this year.
A trend worthy of deep thought is emerging—the most core technical talent in the AI industry is systematically migrating from large companies to startups.
The reason behind this is not just money. Anthropic and OpenAI's pre-IPO equity indeed has huge appeal, but Dean is not short of money at Google. The reason he gave himself is more worthy of attention—the quarterly financial report pressure of listed companies limits the freedom of scientific decision-making. The subtext of this statement is that in a system that needs to account for quarterly income to Wall Street, it is difficult to conduct basic research on a ten-year scale.
Discovery Loop registered as a Public Benefit Corporation, essentially solving this problem at the institutional level. The PBC structure allows the company to pursue commercial returns while maintaining a legally protected mission commitment, without being forced to give way long-term scientific goals to short-term monetization due to shareholder pressure.
Google obviously also realized this contradiction. Alphabet itself invested in Discovery Loop and also provided the first year of computing power—this is almost an admission that some things cannot be done within itself, but it does not want to completely lose these people. This is a pragmatic choice for a large company facing talent outflow, but it also exposes a structural dilemma.
At the same time, Google is also making internal adjustments. In the restructuring announced on the same day as Dean's departure, Demis Hassabis stepped down from daily management of Google DeepMind, transferring to Chairman and Alphabet Chief Scientist; CTO Koray Kavukcuoglu was promoted to Senior Vice President, taking over Gemini model development. Pichai said in an internal letter that this is to inject new momentum into Google's AI work.
But the signal read by the market may be exactly the opposite.
Analysts noticed that the number of iconic figures leaving Google's AI department this summer exceeded any comparable period in the company's history. Those leaving are not "people who train models on infrastructure," but "people who build the infrastructure itself." This is a qualitative change.
Jeff Dean said a sentence at the University of Washington graduation ceremony in June this year, which looks like a preview in retrospect. He told computer science graduates that one of the problems worth solving is "developing tools to accelerate scientific discovery and engineering."
He also said another sentence, which probably better explains why he is leaving.
"The beauty of software is that a small group of people can build something that has a huge impact on the world."
27 years ago, he proved this in a 20-person startup. Now, at 58 years old, he wants to prove it once again.













