Article | Beyond the Layout, Authors: Ban Jun, Huahua
Google has just completed a personnel adjustment without any losers.
At least it appears that way.
On August 6th, Google bid farewell to two of the most important figures of the AI era in a single day.
One is DeepMind founder Demis Hassabis.
The other is Google Chief Scientist Jeff Dean.
The dramatic twist is that no one actually left Google. They were both promoted, receiving the most dignified arrangements, each posting perfect official smiles on their social platforms.
A giant company worth over $4 trillion conducted an organizational reshuffle without generating any shockwaves or grievances.
This event is more worthy of careful scrutiny than the AI models themselves.
I. Google AI: A Decade of Internal Warfare
To understand what happened on August 6th, we must go back further in time.
In 2014, Google acquired DeepMind for £400 million. At the time, the company had only a few dozen people, no products, and no revenue.
Google saw potential in Hassabis himself—an AGI obsessive who played chess at four, became a master at thirteen, sold his gaming company to pursue a PhD in neuroscience.
To guard against Google's commercial influence, Hassabis even signed exceptionally strict terms in the acquisition agreement: DeepMind retained its London headquarters, established an independent ethics committee, and vowed never to directly serve Google's search advertising business.
Post-acquisition, Hassabis reported directly to Google CEO Sundar Pichai, but remained far from the commercial bustle of Mountain View.
Meanwhile, at the Mountain View headquarters, there was another loyal force—Google Brain, led by Google's legendary technical guru Jeff Dean.
Two teams, two cultures, waged a decade-long internal war within Google.
Google Brain was the native, hardcore engineering faction. They fought on the front lines of search, ads, and products, focused on whether technology could be deployed and how it could make money for the company. DeepMind was an elite laboratory in an ivory tower. They didn't care about product cycles or commercialization, focusing solely on pure AGI.
Google Brain viewed DeepMind as a spoiled money-burning machine, consuming hundreds of millions in annual losses, supporting scientists who published papers in Nature/Science daily but couldn't deliver a simple app.
DeepMind viewed Google Brain as fundamentally missing the poetry of AGI. Although Google Brain participated in or led foundational technologies of the large model era like sequence-to-sequence models (Seq2Seq) and the Transformer's attention mechanism, they merely treated hard tech as screws to improve search experience.
The sudden emergence of ChatGPT delivered a resounding slap to such arrogance.
Mutual disdain, yet mutual necessity. This immense internal friction tore at Google for ten years.
During this process, some cleared the field early and left, like DeepMind co-founder Mustafa Suleyman.
Suleyman and Hassabis go way back. He was the best friend of Hassabis's younger brother; the two met in their teens, with a friendship spanning over 20 years. But this co-founder, with his strong desire for control and eagerness to push AI into social governance and commercial application, was forced to leave Google in 2019.
At the time, external rumors attributed it to overly aggressive management style causing internal backlash, but the deeper reason was a clash of visions between the two. Suleyman was eager to ground AI in practical reality, while Hassabis stubbornly guarded academic purity.
Pichai's management team at the time exploited the strategic rift and information asymmetry between the two, coupled with Suleyman's troubles with privacy allegations in the DeepMind Health project, to smoothly marginalize him.
After leaving Google, Suleyman founded Inflection AI, later recruited by Microsoft CEO Satya Nadella, and now reigns as the head of Microsoft AI.
This incident didn't cause major waves back then. But looking back today, a hidden thread was already laid. Google's greatest strength has never been desperately retaining geniuses; it's using refined tactics to make geniuses willingly hand over their positions.
Suleyman wanted to build applications, couldn't do it at Google, so he went to Microsoft. Years later, the same playbook landed on Hassabis. He chose to focus on AI drug discovery; Google's official line was focusing on long-term AGI breakthroughs.
Two DeepMind founders, one left Google, one left Google DeepMind. The logic is startlingly consistent, except the latter's exit was packaged more gloriously and dignifiedly.
II. 2023: Who Swallowed Google Brain?
ChatGPT launched at the end of 2022; GPT-4 became ubiquitous in 2023. Google's internal chaos ensued, hastily launching the Bard model, which flopped during its debut demo, wiping $100 billion off its market value overnight.
Larry Page and Sergey Brin, who had retreated behind the scenes for years, couldn't sit still, frequently storming back into Mountain View conference rooms. What ChatGPT truly shattered in Google was not just the technology gap, but organizational rigidity.
OpenAI was a single army, a single voice; Google had two top-tier teams, two sets of roadmaps, two completely overlapping computing budgets.
According to disclosures by The New York Times, founders Page and Brin gave management a final command: AI must regroup.
The decision was made by the two founders, but the executor of this grand play was Pichai.
Pichai faced a deadly choice: when merging two top teams who despised each other, who would lead?
Pichai chose Hassabis.
In April 2023, Google DeepMind was established, with Hassabis as CEO, and Jeff Dean given the honorary title of Chief Scientist. The media at the time overwhelmingly concluded: Hassabis's complete victory, Google Brain thoroughly swallowed, the unified era has arrived.
But everyone overlooked a fundamental power dynamic. In this merger, who eliminated whom?
Before the merger, Google AI was a fragmented entity Pichai couldn't effectively manage. Google Brain reported to him but had become a cumbersome tail; DeepMind was secluded in London.
After the merger, computing power, budgets, and personnel authority scattered across two locations were forcibly integrated into a single system.
Hassabis achieved a nominal victory, but Pichai gained actual control. This was the first time in Google's 20+ year history that the CEO truly held the reins of the entire AI system in his hands.
Everyone cheered Hassabis's ascent; no one noticed Pichai quietly completing his consolidation behind the scenes.
III. Hassabis Takes Center Stage, Pichai Begins to Maneuver
From 2024 to around 2025, Google's AI narrative was almost entirely dominated by Hassabis alone.
Gemini 2.0 caught up with GPT-4; version 2.5 surpassed it on several benchmarks; the sixth-generation TPU chips powered the computing engine. Coincidentally, Hassabis, with his AlphaFold protein prediction achievement, won the 2024 Nobel Prize in Chemistry, further solidifying his public-facing central role in Google's AI system.
Silicon Valley began wildly spreading the narrative of the king's return; Hassabis became the lone hero saving Google AI.
What was Pichai doing during this period?
He remained low-key, only reading routine scripts on earnings calls, occasionally appearing on podcasts, completely outside the hero narrative.
But in the shadows untouched by the spotlight, Pichai laid down a crucial piece: the silent promotion of Koray Kavukcuoglu.
As early as the 2023 merger, Koray was appointed CTO of the new group. This technical veteran who joined DeepMind in 2012 was a core contributor to hardcore achievements like DQN and WaveNet. Unlike the world-famous Hassabis, Koray was extremely low-profile, a pragmatic figure excelling in engineering delivery and team coordination.
Pichai quietly handed the daily command of the entire R&D army to Koray just as Hassabis's prestige peaked.
He buried this hidden line, quietly waiting for the day of organizational transition.
IV. When Gemini Begins to Lag
Under the spotlight, underlying issues finally erupted.
Entering 2026, Google's AI rhythm began to falter. The Gemini 3.5 Pro, originally slated for a June release, was delayed repeatedly, with no news throughout July. At the August earnings call, Pichai had to announce further increases in AI capital expenditure, with the burn rate over the past four quarters hitting a record high.
Yet, facing analysts' repeated inquiries about the new model's release timeline, management hemmed and hawed, unable to provide any concrete schedule.
The capital market voted with its feet; Google's stock price plunged 11% within two weeks.
Simultaneously, internal friction between research ideals and engineering delivery quietly evolved into a nightmare once again.
According to disclosures by the emerging US tech media Semafor, sources familiar with Hassabis revealed that his idea of stepping down as CEO had been brewing for a full year. During this year, what ignited his passion most wasn't Gemini's parameter tuning or rush releases, but breakthroughs at his AI drug discovery company, Isomorphic Labs.
Forcing someone whose ultimate dream is to change the paradigms of human science to constantly monitor adjustments to AI customer service and search ad interfaces is torture in itself.
He didn't lose focus; he simply grew weary of the endless engineering arms race before him.
On August 6th, the shoe dropped.
Hassabis stepped down as CEO, transitioning to Alphabet Chairperson and Chief Scientist, nominally responsible for long-term AGI strategy. Jeff Dean left entirely, taking his team to start a new venture, with Google swiftly following up with seed-round investment.
Meanwhile, the ever-stealthy Koray was formally promoted to Senior Vice President, taking full charge of all Gemini business lines, reporting directly to Pichai.
Some lamented Google losing the soul of its AI; others praised it as a textbook-level handover.
But if you place these three individuals' positions on a chessboard, you'll see that each of these three moves bears the signature of Pichai's intensely cold and precise style.
V. Pichai Begins to Reclaim Google AI
Hassabis's resignation was packaged as a rather glorious promotion.
Alphabet Chief Scientist—sounds lofty; moving into DeepMind's brand-new Platform 37 office in London's King's Cross, a beautifully appointed space; Pichai personally tweeted thanks; Hassabis posted warm, heartfelt remarks on X.
But strip away this exquisite packaging, and the truth is brutally stark. Teams, budgets, development leadership of Gemini—all went to Koray. Hassabis was sidelined into a dimension called "the future."
This is a near-perfect organizational adjustment. Extracting the spiritual icon of Google AI from the operational hub, bestowing supreme honor, pushing them onto a track grand enough and distant enough—so distant it no longer impacts the current core business race.
Remember Suleyman, who left back then?
Suleyman wanted to build applications, had a falling out, and left—that was a story of departure. With Hassabis, Google gave him an independent Isomorphic Labs and a high-profile title—it became a story of a hero retreating behind the scenes.
Their essence is no different at all: they no longer hold the substantive power over Google AI.
The only difference is that Pichai learned from the Suleyman incident. He no longer gives geniuses the chance to leave in a huff, bearing resentment, to join rival camps. He prepares the most dignified exit for geniuses, letting them retire gracefully with gratitude and glory.
No bloody conflict, only through restructuring—this is Pichai-style management at its most authentic.
VI. Jeff Dean Voices Google's Biggest Problem
As for Jeff Dean's departure, it symbolizes the end of another era.
Jeff Dean, rooted at Google for 27 years, along with his lifelong partner Sanjay Ghemawat, is a Silicon Valley legend. The two are celebrated for writing MapReduce on a single computer, laying the foundation for the modern internet with BigTable, Spanner, and other infrastructures.
In the AI era, they personally built TensorFlow and TPU.
Leaving with him this time are Gemini co-technical lead Oriol Vinyals and Google Brain founding core member Quoc Le. The combined resumes of these four Google Fellows constitute almost half the history of Silicon Valley AI technology.
Jeff Dean was incredibly candid in his departure interview:
Google's infrastructure was born for billion-user products like search and ads, with stability and longevity as core tenets; but AI exploration (Discovery Loop) is entirely different, requiring rapid overturning of foundations, frequent experimentation, and running hundreds of parallel exploration lines simultaneously.
The architect of this legendary system admitted that the system he built could no longer adapt to the fast-paced rhythm of the new era.
Facing the departure of this veteran, Pichai's tactics were once again fully displayed.
He didn't invoke non-compete agreements, nor did he deploy legal teams to fight. After multiple failed attempts to retain him, Pichai simply wrote a check, directly entering Jeff Dean's new company as a founding investor and cloud services partner.
This wasn't generosity; it was an extremely shrewd calculation.
Let Jeff Dean conduct the most cutting-edge, high-risk experiments outside the system, with Google binding him through capital and computing power. If Koray's engineering roadmap ever hits a ceiling, Pichai has a top-tier backup outside the system, ready to be reabsorbed at any time.
He doesn't put all his eggs in one basket; he doesn't even keep the baskets on the same vehicle.
VII. Who Manages AGI?
Reading this far, a clear historical picture emerges.
Those who once controlled Google AI were uniformly top scientists. Hassabis, Jeff Dean, Suleyman—their paper citation counts and academic prestige were enough to command reverence from any tech company.
And who truly controls Google AI today?
One is Koray, a 13-year engineering veteran at DeepMind, familiar with every pitfall in the R&D chain, eschewing grand narratives for delivery timelines.
The other is Pichai—a CEO long underestimated, even mocked in Silicon Valley as merely maintaining the status quo. He doesn't write code, doesn't publish papers; he does one thing: decides who gets to decide.
Control over AI has formally passed from scientists' hands into the hands of professional managers. This is the most profound organizational change in Google AI's nearly 30-year history.
When Pichai issued that seemingly ordinary announcement, he was actually declaring the end of an old era.
But, if tomorrow, another paradigm-shifting revolution like the Transformer emerges in AI, requiring someone to bet on a non-consensus direction based on intuition, who will make that bet?
Rely on Koray's engineering assembly line? Or Pichai's financial statements?
This is the ultimate gamble Pichai has placed. He's betting that the AI race has moved beyond the artisanal stage where individual genius could rewrite fate, fully entering an industrialized era competing on systemic stability, engineering delivery, and organizational efficiency.
Zero enemies, zero grievances, zero negative public sentiment.
Pichai spent three full years quietly pulling an AI organization once beyond his control back into the palm of his hand.
Words from [Beyond the Layout]:
Many believe Google's biggest change today is Hassabis's exit.
I think it's something else.
For the past two decades, the tech industry adhered to a near-worshipful creed: the smartest geniuses should manage the most important technologies.
Today, Google, through an exceptionally dignified transition, tells everyone: the times have changed.
Scientists are responsible for outward exploration; engineers are responsible for grounding ideas; CEOs are responsible for deciding who sits where. AI R&D is becoming like a vast machine.
Not because the models have become simpler.
Precisely because it has become so complex that no single genius can驾驭 (control) it alone.
This might be the true sign of Google entering the second half of the AI era.
It's no longer about models taking the lead, but about the organization finally completing its evolution.
