This article is crafted based on publicly accessible information and serves purely as an exchange of insights, not as any form of investment advice.

On August 5, Google unveiled two significant announcements. Demis Hassabis relinquished his role as CEO of Google DeepMind, transitioning to the position of President of the division and Chief Scientist of Alphabet. Jeff Dean, Google’s 30th hire, officially departed after 27 years with the company. Accompanying him were Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Together, they co-founded a new venture named Discovery Loop.

On the day of the announcement, Google's stock price dipped by over 4%. On that same day, Google proclaimed that the Gemini app had amassed more than 950 million monthly active users. Yet, the media added a cautionary note in the same report: the development of Gemini 3.5 Pro lags several months behind schedule.

The market's focus clearly isn't on those 950 million users; it's on who left and the reasons behind their departure.

Jeff Dean's 27-year tenure at Google nearly mirrors the company's entire technical evolution. He was the architect behind MapReduce, BigTable, and Spanner—three pivotal systems that enabled Google's search engine to manage global traffic and subsequently became foundational infrastructure for the entire tech industry. Dean and Ghemawat are the only two individuals at Google to have been honored with the highest technical accolade, "Senior Fellow." It's no overstatement to say Dean embodies Google's technical essence.

However, the narrative he's weaving as he exits no longer aligns with Google's current trajectory.

On July 25, Dean shared his vision of an automated scientific method with six thousand aspiring entrepreneurs at Y Combinator's Startup School: a model that formulates experimental hypotheses, constructs necessary tools, conducts experiments, evaluates outcomes, and then autonomously designs the next round of experiments. After thousands of iterations, the system might independently uncover breakthroughs in biology or chip design. The audience that day was unaware that he was describing the very company he had just discreetly founded.

Discovery Loop's mission is unequivocal: to empower AI to autonomously navigate the entire scientific research process. This isn't a novel concept. Dario Amodei, Sam Altman, and Jensen Huang have all echoed this sentiment—AI will serve as the driving force behind scientific discovery, tackling challenges beyond human capability.

Yet, Dean and his team are the pioneers in truly leaving a major corporation to pursue this vision. Their approach is also distinctive. The new company is registered as a Public Benefit Corporation, a structure that permits profit-making while allowing management to make decisions that "may not be solely driven by financial gain." Dean candidly stated in an interview, "We may make decisions that aren't purely profit-oriented."

The quartet has yet to secure an office or hire their first employee. When queried about the CEO, Dean paused, slightly embarrassed, and remarked, "I guess I'm the CEO. Everyone's looking at me."

Investors are unconcerned with these details. After meeting with them at his office on Sand Hill Road, Vinod Khosla remarked, "With this team, I don't need to know what they're doing to invest." Khosla Ventures and Radical Ventures have invested, as has Alphabet itself, with Google committing to supply computing power for the new company's inaugural year. The parent company investing in a competitor founded by former employees is, in itself, a microcosm of the ongoing AI talent war.

Hassabis's departure offers another clue.

He no longer oversees the day-to-day operations at Google DeepMind but retains his position as CEO of Isomorphic Labs. This company is dedicated to AI-driven pharmaceuticals, the most direct commercial application of DeepMind's technology. His new title, Chief Scientist of Alphabet, may sound more prestigious than it is. Koray Kavukcuoglu succeeded him as head of DeepMind, reporting to Pichai. Kavukcuoglu has been with DeepMind for over thirteen years, founded the deep learning team, and led research on WaveNet and DQN. His credentials are impressive, but the situation he inherited is far from stable.

Axios reported that DeepMind is experiencing a wave of departures, with employees anxious about the progress. The delay of Gemini 3.5 Pro is partly attributed to low morale. While these anonymous sources cannot be independently verified, combined with the talent exodus of the past few months, they paint a picture of instability.

An earlier departure was Tim Rocktäschel. He left Google earlier this year to found Recursive Superintelligence, which also focuses on AI self-improvement. His reason for leaving was straightforward: the startup environment is entirely different—faster, more focused, with smaller teams and less bureaucracy and politics. This doesn't come across as criticism but rather an acknowledgment of an irreversible trend: when the AI frontier has shifted from "can we build large models" to "can we make models improve themselves," those genuinely committed to answering this question would rather do so in their own garage.

These departures cannot be solely attributed to salary considerations. Dean's compensation at Google was sufficient to secure wealth for himself and several generations of his family. Hassabis's reassignment wasn't a marginalization; he remains within Alphabet's core decision-making circle. What truly motivated them was the same thing: they saw the next step, and their organization wasn't prepared for it.

That step is known as recursive self-improvement. It involves allowing AI systems to optimize themselves, propose experimental hypotheses, build tools, evaluate results, and independently enter the next cycle. This logic is repeatedly echoed in Discovery Loop's mission statement, in the name of Recursive Superintelligence, and in Dean's description to YC students. It's not the delusion of a single individual but a shared belief among the tiny elite at the forefront of AI today.

And Google, coincidentally, is the company that first encountered this threshold but finds it hardest to cross. Its search advertising empire demands stability, its cloud business needs to cater to major clients, and its Gemini needs to catch up to competitors. All these necessitate one prerequisite: no loss of control. Recursive self-improvement inherently carries the risk of going out of control; no one can predict where a model that has modified itself thousands of times will end up. For a publicly traded company responsible for hundreds of millions of users and billions in advertising revenue, this isn't an easily acceptable proposition. For a startup, it's the entire value proposition.

Pichai stated in the announcement that Google is the only company with full-stack capabilities. He wasn't incorrect. Google possesses TPUs, data centers, the largest user base globally, and DeepMind's research prowess. But the cost of full-stack is that no layer can fail. Yet recursive self-improvement requires tolerance for failure. Its entire logic is to find the optimal solution through continuous trial and error. In a large company, this trial-and-error rhythm is slowed down by layers of safety reviews, ethical reviews, and legal compliance. Dean and his team don't lack computing power or funding; what they lack is the freedom to make all their mistakes within a sufficiently short time window.

This is the true essence behind Vinod Khosla's statement, "I don't need to know what they're doing to invest." He's not investing in a business plan but in a judgment: recursive self-improvement is the next frontier of AI, and Jeff Dean is one of the few individuals on Earth who might achieve it. As for the specifics of what to do or how to do it, those details can remain unknown at the outset. What matters is the journey itself and the absence of baggage at the starting point.

In his farewell memo, Hassabis echoed Pichai's sentiment, stating that Google is the only company with full-stack capabilities. That statement is factual. But it also elucidates why this group chose to leave.

Full-stack is Google's strength. It's also its burden. Discovery Loop has only four people, no office, and no employees. Their only assets are the names of the four individuals and a shared belief: AI can improve itself. Whether this faith can be transformed into a product remains uncertain. But at least one person believes it's worth pursuing—Jeff Dean, who for the first time in 27 years, is no longer a Google employee.

Full-stack is too cumbersome. To run fast, you must travel light.