Google used an Aug. 6 management shake-up to redraw control of its AI organization, wrapping a major power shift in promotions, new titles and a clean public presentation.
The two biggest names in the move were Demis Hassabis, founder of DeepMind, and Jeff Dean, Google’s chief scientist. Hassabis did not leave the company. He stepped down as CEO and became Alphabet president and chief scientist, with responsibility for long-term AGI strategy. Dean did leave Google, but not in a hostile split: he departed to start a new company, and Google quickly joined its seed financing.
The operational winner was Koray Kavukcuoglu, who was promoted to senior vice president and took full control of the Gemini business, reporting directly to CEO Sundar Pichai.
On the surface, Google managed the transition without visible turmoil. In the article’s telling, that is precisely the point. The reshuffle did not look like a purge or a rupture. It looked controlled, polished and voluntary. But set against Google’s internal AI history, it marked a transfer of real authority.
A decade of tension between DeepMind and Google Brain
The article starts in 2014, when Google acquired DeepMind for £400 million. At the time, DeepMind was still small, with no commercial product and no revenue. Google was buying Demis Hassabis as much as it was buying a company.
According to the piece, Hassabis secured unusual protections as part of the deal: DeepMind would keep its London headquarters, set up an independent ethics board and avoid direct service to Google’s search advertising business. After the acquisition, Hassabis reported to Pichai while operating at a distance from Mountain View’s commercial core.
Inside Google, another AI power center already existed. Google Brain, led by Jeff Dean, was the homegrown engineering heavyweight. The article describes the two groups as separate forces with separate cultures. Google Brain sat closer to products, search and advertising, with a focus on shipping and monetization. DeepMind leaned toward long-horizon research and AGI.
That split lasted for roughly a decade. The article frames it as an internal conflict that drained energy from Google even though both sides were critical to its AI progress. Google Brain had been involved in or led foundational work including sequence-to-sequence models and the Transformer attention mechanism. DeepMind held fast to its scientific mission. ChatGPT, in the article’s view, exposed the cost of that divide.
Mustafa Suleyman’s earlier exit
The piece also revisits the departure of DeepMind co-founder Mustafa Suleyman, treating it as an early signal of a pattern in how Google handled ambitious AI leaders.
Suleyman and Hassabis had known each other for years, but the article says they diverged on strategy. Suleyman wanted AI pushed into social systems and commercial applications. Hassabis was more protective of scientific purity. Suleyman left Google in 2019. At the time, public discussion focused on management style and internal resistance. The article places that exit in a broader struggle over direction inside DeepMind and Google.
After leaving, Suleyman founded Inflection AI and was later hired by Microsoft CEO Satya Nadella to lead Microsoft AI. In the article’s reading, Google’s strength was not always retaining every major talent figure, but shaping structures so that control could move even when prominent people did not stay in the same role.
The 2023 merger after ChatGPT and GPT-4
ChatGPT’s launch in late 2022 and GPT-4’s rise in 2023 pushed Google into crisis mode. Bard, Google’s rushed answer, stumbled in its first public demo, and the company lost hundreds of billions of dollars in market value overnight, according to the article.
The article says the deeper issue was organizational. OpenAI was operating with one command structure and one voice. Google, by contrast, had two elite teams, two different strategies and overlapping compute budgets. Citing reporting from The New York Times, the article says Google co-founders Larry Page and Sergey Brin ordered management to regroup the company’s AI resources.
That led to the creation of Google DeepMind in April 2023. Hassabis became CEO of the merged unit. Jeff Dean moved into the role of chief scientist. At the time, outside observers widely treated the move as a clear win for Hassabis and a defeat for Google Brain.
The article argues that this reading missed the deeper outcome. Before the merger, Google’s AI system was fragmented and difficult for Pichai to fully command. After it, compute, budgets and personnel authority were brought into one structure. Hassabis gained the top title, but Pichai gained tighter control over the whole machine.
Hassabis became the face, while Koray was positioned inside the system
From 2024 into 2025, the article says, Google’s public AI story increasingly centered on Hassabis. Gemini 2.0 caught up with GPT-4. Version 2.5 moved ahead on multiple benchmarks. Google’s sixth-generation TPU helped carry the company’s compute push. Hassabis also won the 2024 Nobel Prize in Chemistry for AlphaFold, reinforcing his standing as the public face of Google’s AI effort.
While that story played out, Pichai stayed largely outside the spotlight. The article presents that as a contrast to what was happening internally. One of the key moves, it says, was the quiet rise of Koray Kavukcuoglu.
Koray joined DeepMind in 2012 and was a core contributor to work including DQN and WaveNet. When Google’s AI groups were merged in 2023, he became CTO of the new organization. The article portrays him as a low-profile operator focused on engineering execution and team coordination rather than grand public narratives.
As Hassabis’ status climbed, day-to-day command over the research and development apparatus increasingly moved toward Koray. The article treats that as a deliberate setup for the eventual handoff.
Gemini delays set the stage for the Aug. 6 reset
According to the article, Google’s AI cadence began to slip in 2026. Gemini 3.5 Pro, originally expected in June, was repeatedly delayed, and July passed without an update. On Google’s August earnings call, Pichai said the company would keep increasing AI capital spending, with the pace over the previous four quarters reaching a record high.
Management did not give analysts a firm release timetable for the next model. Google’s stock then fell 11% over two weeks, the article says.
At the same time, Semafor reported that people familiar with Hassabis said he had been considering stepping down as CEO for a full year. The article says what most energized him in that period was not tuning Gemini or accelerating releases, but progress at Isomorphic Labs.
The formal outcome arrived on Aug. 6:
- Hassabis stepped down as CEO and became Alphabet president and chief scientist, focused on long-term AGI strategy.
- Jeff Dean left Google and launched a startup with his team, with Google moving quickly to join the seed round.
- Koray Kavukcuoglu was promoted to senior vice president, took charge of the full Gemini line and began reporting directly to Pichai.
Read together, the article says, the three moves reflect Pichai’s management style. High-profile figures received prestigious and respectful placements. Operational control, budget authority and product leadership shifted toward an engineering-and-delivery command structure.
A promotion for Hassabis, but not the same kind of power
Publicly, Hassabis’ transition was packaged as an elevation. Alphabet president and chief scientist sounds more senior than a divisional CEO title. He also moved into DeepMind’s new Platform 37 office in King’s Cross, London. Pichai thanked him publicly on X, and Hassabis responded in kind.
The article’s argument is that the substance lies elsewhere. Teams, budgets and control over Gemini development moved to Koray. Hassabis was relocated to a broader, longer-horizon domain: AGI strategy and the future.
It compares that outcome with Suleyman’s earlier path. One left Google. The other left the operational core of Google DeepMind. In both cases, the article says, they no longer held real power over Google’s AI business. The difference is that this time Google built a far more elegant landing.
Jeff Dean’s exit and what it says about Google’s infrastructure
Jeff Dean’s departure is presented as the closing of another chapter.
Dean spent 27 years at Google and, alongside Sanjay Ghemawat, helped build core systems including MapReduce, BigTable and Spanner. In the AI era, he also helped create TensorFlow and TPU. The article says that Oriol Vinyals, co-lead for Gemini technology, and Quoc Le, one of the original Google Brain figures, also left.
In an exit interview cited by the article, Dean said Google’s infrastructure was built for products such as search and advertising that serve massive user bases and prize stability and durability. AI discovery works differently, he said: it needs systems that can be rapidly rebuilt, run frequent experiments and support hundreds of parallel lines of exploration.
The article treats that statement as unusually direct. One of the architects of Google’s legendary technical stack was effectively acknowledging that the system he helped build no longer fits the speed of the current AI cycle.
Pichai’s response, as described in the article, was not to lean on legal pressure or non-compete restrictions. After trying repeatedly to keep Dean, Google instead entered his new venture as an early investor and cloud partner. The article sees this as a strategic hedge: let frontier experimentation happen outside the corporate structure, but keep capital and compute ties in place.
Control shifts from scientists to managers
The article ends with a broader conclusion about who now runs Google AI. In the earlier era, the people at the top were star scientists such as Hassabis, Dean and Suleyman, figures whose authority came from research reputation and technical prestige.
Today, it argues, operational control rests with a different combination. One is Koray, an engineering veteran who knows the development pipeline and focuses on shipping. The other is Pichai, a CEO whose central role is deciding who gets to decide.
That amounts, in the article’s view, to a transfer of power from scientists to professional managers. It calls this the deepest organizational change in nearly 30 years of Google AI history.
The piece leaves one unresolved question. If another paradigm shift on the scale of the Transformer appears and requires a non-consensus bet, who makes that call? Koray’s engineering system, or Pichai’s corporate command structure?
What is clear from the Aug. 6 reshuffle is where Google is placing its wager now: on stability, execution and organizational efficiency in an AI industry that looks less like a research workshop and more like an industrial system.