Google AI, Relieving the generals of their military commands over a cup of wine

Google has just completed a personnel reshuffle with no losers.

At least it seems so.

On August 6, Google saw off two of the most important figures in the AI era on the same day.

One is Demis Hassabis, founder of DeepMind.

The other is Jeff Dean, Google's Chief Scientist.

Dramatically, neither of them actually left Google. They both got promoted, received the most decent arrangements, and everyone posted a perfect official smile on social platforms.

A $4+ trillion giant company carried out a core power reshuffle without causing any shocks or complaints.

This matter is more worth careful scrutiny than the AI model itself.

I. Google AI, a Decade of Internal Strife

To understand what happened on August 6, we have to go back to an earlier time.

In 2014, Google acquired DeepMind for 400 million pounds. At that time, the company had only dozens of employees, no products, and no revenue.

What Google valued was Hassabis himself, an AGI fanatic who started playing chess at the age of four, became a master at thirteen, sold his game company and went to pursue a doctorate in neuroscience.

Back then, to guard against Google's commercial penetration, Hassabis even signed extremely strict clauses in the acquisition agreement: DeepMind retains its London headquarters, sets up an independent ethics committee, and will never directly serve Google's search advertising business.

After the acquisition, Hassabis reported directly to Google CEO Pichai, like an independent kingdom far from the commercial noise of Mountain View.

At the same time, there was another direct line of troops in the Mountain View base, Google Brain, led by Jeff Dean, a legendary tech guru at Google.

Two teams, two sets of cultures, fought a full decade-long civil war inside Google.

Google Brain is a native hardcore engineering faction. They fought on the front line of search, advertising and products, focusing on whether the technology can be launched online and how to make money for the company; DeepMind is an elite laboratory in the ivory tower. They don't care about product rhythm, let alone commercialization, and only focus on pure AGI.

Google Brain thinks DeepMind is a spoiled money-burning machine that eats hundreds of millions of dollars in losses every year, supporting a group of scientists who publish Nature/Science papers every day but can't even deliver an APP.

DeepMind, on the other hand, thinks Google Brain doesn't understand the poetry of AGI at all. Although Google Brain participated in or led the technical foundations that laid the foundation for the large model era, such as the sequence-to-sequence model (Seq2Seq) and the Transformer attention mechanism, they only regarded hard technology as a screw to improve the search experience.

The sudden emergence of ChatGPT gave this arrogance a resounding slap in the face.

They look down on each other but cannot avoid each other. This huge internal friction has torn Google apart for ten years.

In this process, some people left the market early: Mustafa Suleyman, co-founder of DeepMind.

Suleyman has a deep connection with Hassabis. He is the best friend of Hassabis's younger brother. The two met when they were teenagers, and their friendship spans 20 years. But this co-founder, who has a strong desire for control and is bent on promoting AI to social governance and commercial implementation, was forced to leave Google in 2019.

At that time, the external widespread rumor was that his overly tough management style caused internal backlash, but the deeper reason was the route dispute between the two: Suleyman was eager to bring AI into real-world application, while Hassabis stuck to academic purity.

At that time, the management led by Pichai took advantage of the strategic rift and information gap between the two, plus the trouble Suleyman faced with privacy allegations in the medical AI (DeepMind Health) project, and pushed the boat along to marginalize him.

After leaving Google, Suleyman founded Inflection AI, and was later poached by Microsoft CEO Nadella with a high salary, and is now the head of Microsoft AI.

This incident did not cause much waves back then. But looking back today, a hidden line has long been laid: Google has never been good at retaining geniuses by clinging to them, but using sophisticated means to make geniuses hand over power willingly.

Suleyman wanted to make applications that could not be realized in Google, so he simply went to Microsoft to do it. A few years later, the same drama fell on Hassabis. He chose to engage in AI drug R&D, and Google's official statement was that he would focus on long-term AGI breakthroughs.

Of the two DeepMind founders, one left Google, and the other left Google DeepMind. The logic is surprisingly consistent, except that the latter's exit is packaged more gorgeously and decently.

II. 2023, Who Is The Person That Absorbed Google Brain?

At the end of 2022, ChatGPT was launched, and GPT-4 swept across the market in 2023. Google was in chaos internally. Bard, the large model product launched in a hasty response, flopped at its debut demo, and Google's market value evaporated by hundreds of billions of dollars overnight.

Larry Page and Sergey Brin, who had stayed behind the scenes for many years, could not sit still and frequently returned to the conference rooms in Mountain View. What ChatGPT really broke through in Google was not only the technical gap, but also the organizational rigidity.

OpenAI is a single corps with one unified voice; while Google has two top teams, two sets of routes, and two completely overlapping computing power budgets.

According to the disclosure of The New York Times, the two founders Page and Brin gave the management a dead order: AI must be regrouped.

This decision was made by the two founders, but the executioner who carried out this power centralization drama was Pichai.

Pichai faced a death multiple-choice question: Two top teams that refuse to accept each other merge, who will be the leader?

Pichai chose Hassabis.

In April 2023, Google DeepMind was established, Hassabis served as CEO, and Jeff Dean was given the nominal post of Chief Scientist. At that time, the media overwhelmingly concluded that: Hassabis won a big victory, Google Brain was completely swallowed, and the era of unification came.

But at that time, everyone ignored the underlying power logic. After this merger, who on earth eliminated whom?

Before the merger, Google AI was a loose federation that Pichai could not control. Google Brain reported directly to him but was too big to manage, while DeepMind was far away in London holding its own power.

After the merger, the computing power, budget, and personnel rights scattered in the two places were forcibly integrated into one system.

Hassabis got the nominal victory, but Pichai got the actual control. This is the first time in Google's more than 20 years of history that the CEO has truly held the reins of the entire AI system in his own hands.

Everyone was cheering for Hassabis's ascent to the top, and no one noticed that Pichai quietly completed the power centralization behind the scenes.

III. Hassabis Steps to the Front Stage, Pichai Starts Layout

Around 2024 to 2025, Google's AI narrative was almost monopolized by Hassabis alone.

Gemini 2.0 caught up with GPT-4, and the 2.5 version overtook GPT-4 on multiple benchmarks, while the 6th generation TPU chip supported the computing power. At this juncture, Hassabis won the 2024 Nobel Prize in Chemistry for his AlphaFold protein prediction achievements, further consolidating his core public position in Google's AI system.

Silicon Valley began to wildly spread the narrative of the return of the king, and Hassabis became the lone hero who saved Google AI.

What was Pichai doing during this period?

He was still unhurried, only reading routine manuscripts at earnings calls, occasionally appearing on podcasts, completely staying out of the hero narrative.

But in the shadows out of the spotlight, Pichai laid a key move, which was the silent promotion of Koray Kavukcuoglu.

As early as when the two teams merged in 2023, Koray was appointed CTO of the new group. This technical veteran who joined DeepMind in 2012 is the core contributor to hardcore achievements such as DQN and WaveNet. Unlike the world-famous Hassabis, Koray is an extremely low-key pragmatist who is good at engineering delivery and team scheduling.

When Hassabis's reputation reached its peak, Pichai quietly handed over the daily command of the entire R&D corps to Koray.

He laid this hidden line, waiting quietly for the day when the organization would transition between the old and the new.

IV. When Gemini Starts to Fall Behind

Under the highlight, hidden dangers eventually broke out.

Entering 2026, Google's AI rhythm began to get out of control. Gemini 3.5 Pro, which was originally scheduled to be released in June, was delayed again and again, with no news at all throughout July. At the August earnings meeting, Pichai had to announce to continue to increase AI capital expenditure, and the money-burning speed in the past four quarters hit a record high.

However, in the face of analysts' repeated inquiries about the release timeline of the new model, the management hesitated and could not give any exact schedule.

The capital market immediately voted with their feet, and Google's stock price plummeted 11% in two weeks.

At the same time, a new round of internal friction between scientific research ideals and engineering delivery quietly turned into a nightmare again.

According to the disclosure of Semafor, a cutting-edge American technology media, people familiar with Hassabis revealed that his idea of stepping down as CEO has been brewing for a whole year. In this year, what can most ignite his passion is not the parameter fine-tuning and rush-to-release of Gemini, but the breakthrough of Isomorphic Labs, its AI pharmaceutical subsidiary.

Asking a person whose ultimate dream is to change the paradigm of human science to stare at the interface adjustment of AI customer service and search advertising every day is a kind of torture in itself.

He is not out of focus, he is just tired of this endless engineering arms race.

By August 6, the dust settled.

Hassabis stepped down as CEO, transferred to Chairman and Chief Scientist of Alphabet, nominally in charge of long-term AGI strategy; Jeff Dean left completely to start a business with his team, and Google quickly followed up with a seed round investment.

And Koray, who had been hiding behind the scenes, was officially promoted to Senior Vice President, taking full charge of all Gemini business and reporting directly to Pichai.

Some people sigh that Google has lost the soul of AI, while others praise this as a textbook-level handover.

But if you put the positions of these three people on a chessboard, you will find that every step of these three moves has the extremely cold and precise Pichai style.

V. Pichai, Starts to Take Back Google AI

Hassabis's resignation was packaged as a rather brilliant promotion.

Chief Scientist of Alphabet sounds like a transcendent position; he moved into DeepMind's brand-new Platform 37 office building in King's Cross, London, with a magnificent office; Pichai personally posted on X to thank him, and Hassabis also posted warm remarks on X.

But stripping off these fancy packages, the truth is extremely cruel. The team, budget, and the leading right of Gemini development all go to Koray. Hassabis was sidelined to a dimension called "the future".

This is a perfect level of de-administration: stripping the spiritual totem of Google AI from the operational hub of the organization, giving him extremely high reputation, pushing him to a sufficiently distant and grand track, far enough that he will no longer affect the current main business battlefield.

Remember Suleyman who left back then?

Suleyman wanted to make applications and left after falling out, which was a story of departure and betrayal. For Hassabis, Google gave him the independent Isomorphic Labs and high-profile titles, turning it into a story of a hero stepping back behind the scenes.

The essence of the two is no different: they no longer hold the real power of Google AI.

The only difference is that Pichai learned lessons from the Suleyman incident. He no longer gave geniuses the chance to leave angrily and switch to the opposing camp. He prepared the most decent way for geniuses to let them exit gracefully with gratitude and aura.

Relying not on bloody conflicts, but only on organizational restructuring, this is the most authentic Pichai-style management.

VI. Jeff Dean, Speaks Out Google's Biggest Problem

As for Jeff Dean's departure, it symbolizes the end of another era.

Jeff Dean, who has taken root in Google for 27 years, and his lifelong partner Sanjay Ghemawat are known as Silicon Valley myths. The two were recorded in history for writing MapReduce with one computer shared by two people, and the infrastructure such as BigTable and Spanner laid the cornerstone of the entire modern Internet.

In the AI era, they built TensorFlow and TPU by their own hands.

Along with him, Oriol Vinyals, co-technical leader of Gemini, and Quoc Le, founding core of Google Brain, also left. The resumes of these four Google Fellow-level figures combined are almost half of the history of Silicon Valley AI technology.

Jeff Dean spoke extremely frankly in his exit interview:

Google's infrastructure was built for billion-user products such as search and advertising, with the core of stability and durability; but AI Discovery Loop is completely different, which requires quickly overthrowing the underlying architecture, running experiments frequently, and launching hundreds of parallel exploration lines at the same time.

The founder of this legendary architecture personally admitted that the castle he built can no longer adapt to the fast pace of the new era.

Faced with the departure of this veteran, Pichai's tactics were once again demonstrated incisively and vividly.

He did not use non-compete agreements, nor did he use legal teams to tear things up. After many unsuccessful attempts to retain him, Pichai directly wrote a check and entered Jeff Dean's new company directly as a founding investor and cloud service partner.

This is not generosity, but an extremely clever calculation.

Let Jeff Dean do the most cutting-edge and high-risk trial and error outside the system, and Google binds him with capital and computing power. If Koray's engineering route hits the ceiling one day, Pichai has a top-level backup that can be incorporated at any time outside the system.

He doesn't put all his eggs in one basket, not even put the basket on the same car.

VII. Who Is Managing AGI

Seeing this, a clear historical picture emerges.

The people who used to be in charge of Google AI are all top scientists. The number of paper citations and academic reputation of Hassabis, Jeff Dean, and Suleyman are enough to make any technology company worship them.

But who is the person who actually holds the power of Google AI today?

One is Koray, an engineering veteran who has worked in DeepMind for 13 years, who knows every pit on the R&D chain, does not talk about grand narratives, only talks about delivery deadlines.

The other is Pichai. A CEO who has long been underestimated by Silicon Valley and even ridiculed as a conservative bureaucrat. He does not write code or publish papers, he only does one thing: decide who makes the decisions.

The control of AI has officially passed from the hands of scientists to the hands of professional managers. This is the most profound organizational change in the nearly 30-year history of Google AI.

When Pichai released that seemingly ordinary announcement, he actually announced the end of an old era.

However, if tomorrow there is a paradigm revolution in the AI field like Transformer, which requires someone to bet on a non-consensus direction by intuition, who will place this bet?

Rely on Koray's engineering assembly line? Or on Pichai's financial statements?

This is the ultimate bet Pichai placed. He bets that the AI competition has crossed the workshop stage where geniuses alone can change fate, and has fully entered the industrial era of competing for system stability, engineering delivery and organizational efficiency.

Zero enemies, zero complaints, zero negative public opinion.

Pichai spent three full years quietly taking back the AI empire that was once out of his control into his own palm.