Pichai Completes Quiet Takeover of Google AI Command

Demis Hassabis handed day-to-day control of Google DeepMind to Koray Kavukcuoglu on August 5 while Jeff Dean left after 27 years to launch Discovery Loop with Alphabet as founding investor. The stock dropped more than 4 percent that day.

The moves look graceful on paper. They finish a longer shift that put operational AI power under Sundar Pichai.

One day of announcements closed years of dual reporting lines, dual cultures and dual power centers. What remains is a cleaner chain from Pichai to the people who ship models and the people who still explore outside the public company walls.

The New Org Chart Settles in One Day

Hassabis becomes Chair of Google DeepMind and Chief Scientist of Alphabet. He keeps leading Isomorphic Labs and will advise on models and research from the new Platform 37 offices in London. Koray Kavukcuoglu, previously CTO and Chief AI Architect, steps up as SVP of Google DeepMind reporting directly to Pichai. He owns Gemini model development, frontier research, the Gemini app and developer teams.

Jeff Dean, Google’s longtime chief scientist, departs with Senior Fellow Sanjay Ghemawat, Oriol Vinyals and Quoc Le. Their public benefit corporation, Discovery Loop, targets automated discoveries in machine learning, science and engineering. Alphabet joins as founding investor and Cloud partner.

Pichai and Hassabis posted coordinated Sundar and Demis staff memos framing the change as acceleration toward AGI and science.

The memos arrived the same day the reporting lines changed. That timing left little room for internal ambiguity about who now owns the Gemini roadmap and who owns the longer scientific horizon.

Three Years of Quiet Centralization

Google bought DeepMind in 2014 for roughly £400 million. Hassabis kept London independence, an ethics board and a firewall from search ads. Google Brain under Dean ran the Mountain View engineering track focused on shipping and monetizing. The two cultures clashed for a decade over compute budgets and priorities.

1. 2014: DeepMind acquisition; Hassabis reports to Pichai with autonomy clauses.
2. 2019: Co-founder Mustafa Suleyman exits after internal friction and privacy issues around DeepMind Health; later joins Microsoft AI.
3. April 2023: Post-ChatGPT crisis, Brain and DeepMind merge into Google DeepMind under Hassabis as CEO; Dean becomes chief scientist. Pichai gains unified budget and headcount control for the first time.
4. 2024-2025: Hassabis public face (Nobel Chemistry for AlphaFold); Koray quietly elevated to daily R&D command.
5. August 5 2026: Hassabis to chair/chief scientist; Koray full SVP; Dean team exits with capital tie-back.

Page and Brin pushed the 2023 merger after Bard’s rocky launch. Pichai executed it. The latest step removes the last dual-power overhang.

Each step narrowed the gap between London research culture and Mountain View product culture. Autonomy clauses gave way to unified budgets. Public scientific prestige stayed with Hassabis while daily command migrated toward operators who report straight to the CEO.

The 2019 Suleyman exit showed how costly internal friction could become once a co-founder left for a rival. The 2026 package avoids that pattern. Prestige titles, a London base and a capital tether replace angry departures.

Scientists Hand the Reins to Managers

Hassabis, Dean and Suleyman built reputations on papers, citations and paradigm jumps. Koray joined DeepMind in 2012, founded its deep learning team and delivered DQN and WaveNet. His public profile stayed low. His new mandate is velocity and roadmap delivery.

Pichai does not write models. He decides who does. The control of Google AI has moved from scientists who set direction by insight to professional managers who set it by schedule and system reliability. That is the second-order change.

I’ve been working towards AGI my whole life and now, like many of you, I feel it is close at hand. It’s critical that we collectively get the next steps right to ensure this all goes well for humanity.

Hassabis wrote that in his note, adding he needs space for the big picture and Isomorphic. The packaging is elegant. The team, budget and Gemini lead all sit with Koray.

• Hassabis retains prestige titles and science track while losing operational levers.
• Dean’s infrastructure generation built for stable billion-user products now yields to faster discovery loops outside the company.
• Koray’s 13-year engineering track record becomes the internal bet on shipping.
• Alphabet keeps a capital and Cloud tether to the departing talent.

Crowd reaction on X noted the production-layer exits clustered in one window while Gemini timelines slipped. Some called it a rational scientist response when capital expenditure favors infrastructure certainty over model gambles.

The pattern is consistent. Scientists who once set direction by insight now hold chair titles, external labs or advisory roles. Managers who measure velocity hold the direct line to Pichai and the Gemini budget.

Discovery Loop Becomes the External Hedge

Dean told the New York Times the new setup lets them make decisions that are not necessarily in the purest financial interests of a public company. MapReduce, BigTable, Spanner, TensorFlow and TPU all carry his and Ghemawat’s fingerprints. He said Google’s castle was built for durability; AI discovery needs rapid overthrow of architectures and hundreds of parallel experiments.

• MapReduce
• BigTable
• Spanner
• TensorFlow
• TPU

Those systems underpinned durable, billion-user products. Discovery Loop is built for the opposite tempo: fast architectural overthrow and many parallel bets that a public company roadmap may not tolerate.

Pichai’s response after failed retention attempts was a check and a partnership. If Koray’s engineering path hits a ceiling, Alphabet already holds a seat at the table of the highest-risk external lab. Eggs stay in more than one basket.

The public benefit corporation form gives Dean’s group room to chase non-consensus paths. The founding investor and Cloud partner roles give Alphabet visibility and a commercial link without forcing those paths back inside quarterly reporting pressure.

Koray’s Delivery Machine Takes Gemini

Gemini app users hit 950 million monthly. Gemma downloads passed 900 million. Flash demand is high and a Cyber model is live. Hassabis’s memo name-checked progress on Gemini 4. Yet Gemini 3.5 Pro slipped past its June target with no firm date by the August earnings call. Capital spending hit records. The stock had already wobbled before the leadership news added another 4 percent drop.

Koray’s Koray Kavukcuoglu DeepMind leadership page frames his job as clear path and absolute velocity on the Gemini roadmap. He reports straight to Pichai. That line used to run through Hassabis.

Products that sit on top of the stack, including the Google AI Pro subscription bundle details, now rest on this tighter operational chain. Local options such as Gemma models running locally on Mac still depend on the same research engine room.

High user counts and download totals show distribution strength. The slipped 3.5 Pro date and the same-day stock drop show that markets still price delivery risk harder than installed base. Koray’s mandate is to close that gap.

Isomorphic and the Gilded Track

Hassabis has long said the number-one application of AI should be human health. Isomorphic Labs, the 2021 DeepMind spin-out, is the vehicle. It raised $2.1 billion earlier in 2026. Clinical timelines have shifted before, but the mission remains curing disease with frontier models built on AlphaFold.

The new titles give him time and a magnificent London base. They also place him far enough from the daily Gemini arms race that he no longer sets its tempo. Suleyman left angrily in 2019 and joined a rival. Hassabis leaves gratefully with aura intact. The outcome for Google’s core AI power is similar.

Pichai spent three years turning a loose federation he could not fully steer into a single system he can. Zero public enemies, zero angry exits to competitors, zero messy non-competes. The bet is that the AI race has left the workshop stage where one genius changes fate and entered the industrial stage of system stability, delivery cadence and organizational efficiency.

Whether that bet holds when the next non-consensus leap appears is the open question the memos do not answer.

The Market Prices Delivery Uncertainty

The more than 4 percent drop landed on the same day the org chart changed. It did not arrive in isolation. Gemini 3.5 Pro had already missed its June target. Capital spending had already hit records. The stock had already wobbled.

Investors saw production-layer exits clustered with a slipped model date and a clearer handoff to an operator whose public brief is velocity. The user and download numbers show scale. The price action shows that scale alone does not quiet timeline risk.

Koray now owns the direct line that must convert installed base into on-time frontier releases. Hassabis owns the longer scientific frame. Dean’s group owns an external path Alphabet still funds. The market reaction treated the internal delivery chain as the near-term variable that matters most.

Alphabet Keeps Capital Ties to Both Paths

The internal path runs through Koray to Pichai with Gemini ownership concentrated and reporting simplified. The external path runs through Discovery Loop with Alphabet as founding investor and Cloud partner. Both paths remain funded. Neither path forces the other to adopt its tempo.

That structure answers the retention problem without forcing every high-risk experiment back inside a public company roadmap. It also answers the control problem Pichai faced when Brain and DeepMind still operated as parallel cultures.

• Internal Gemini work now has one SVP owner and one CEO reporting line.
• External discovery work keeps independence while Alphabet holds capital and Cloud links.
• Isomorphic keeps Hassabis on the health mission with fresh 2026 funding.
• No messy non-competes and no public rival exits mark the 2026 package.

Pichai’s three-year centralization ends with operational levers inside and optionality outside. The industrial stage bet is now fully on the books. The next non-consensus leap will test whether that design still leaves room for the kind of jump that once came from the workshop.

<|eos|>