The Handoff: Inside Sundar Pichai's Decision to Rebuild Google's AI Command Structure

On the morning of August 5, 2026, Sundar Pichai published a staff memo that did two things a CEO rarely does in a single document. It handed a chunk of his most consequential organization to a new operating executive. And it said goodbye to one of the engineers most closely tied to Google's technical history.

The memo announced that Demis Hassabis, the co-founder and chief executive of Google DeepMind, was giving up day-to-day control of the lab that builds Gemini. Hassabis would become DeepMind's chairman and take on a new title, chief scientist of Alphabet — a role focused on long-range research and AI safety rather than running a business unit. Taking his place as the executive who actually runs Google DeepMind day to day was Koray Kavukcuoglu, the lab's chief technology officer, promoted to senior vice president. Kavukcuoglu would now own the Gemini 4 roadmap. And he would report directly to Pichai — not to Hassabis, not through a layered chain of command, but straight to the CEO.

The same memo thanked Jeff Dean, Alphabet's chief scientist, for what Pichai called "an incredible 27-year run." Dean, who joined Google in 1999 as roughly its 30th employee, was leaving. So was Sanjay Ghemawat, a senior fellow who had worked alongside Dean for decades. The two were starting an independent company called Discovery Loop, built around the idea of using AI to run scientific research itself — proposing experiments, executing them, evaluating results, and iterating. Discovery Loop's own materials named two additional co-founders, Oriol Vinyals and Quoc Le, both senior Google AI researchers; Pichai's memo and public posts mentioned only Dean and Ghemawat by name. Google agreed to become a founding investor in Discovery Loop and to supply it with cloud computing capacity.

Alphabet's stock fell that day — reported in a range of roughly 4% to 5% depending on the outlet and the moment measured — as investors absorbed the fact that the company's AI organization was being restructured and losing senior technical leadership at the same time.

This is not simply a story about a company having a bad week. It's a story about a specific decision: to centralize authority over Google's most important AI work under one executive who answers directly to the CEO, at a moment when competitors are actively recruiting that organization's senior researchers. Why Pichai made that call, and what it does and doesn't resolve, says more about how he's running Alphabet in 2026 than any product launch does.

The situation Pichai inherited

Google DeepMind has been the center of Alphabet's AI strategy since Google folded its DeepMind and Google Brain research groups together in 2023, with Hassabis running the combined lab. By early 2026, the lab's central product, Gemini, was Alphabet's most closely watched bet, and its flagship next-generation model, Gemini 3.5 Pro, was supposed to arrive by June 2026, according to guidance Pichai gave at Google I/O in May. It didn't. Independent reporting places the model still unreleased as of mid-August, with technical explanations ranging from coding-performance shortfalls to a data-refresh process that reportedly left the model behind schedule. Google itself has said only that the model is "coming soon" and being tested with partners; it has not confirmed a release date or given a detailed public explanation for the delay.

At the same time, Google's AI research bench was thinning. Noam Shazeer, a co-author of the original Transformer research that underpins modern generative AI, was among several senior researchers reported to have left Google in the weeks before the August 5 announcement. Earlier in 2026, DeepMind lost a Nobel laureate: John Jumper, who had shared the 2024 Nobel Prize in Chemistry with Hassabis for AlphaFold, departed for Anthropic. The pattern was consistent enough that outside trackers had begun compiling running counts of senior AI departures from Google over the summer.

The competitive clock Pichai was working against

None of this was happening in a vacuum. Anthropic and OpenAI have spent 2026 aggressively recruiting the kind of senior research talent Google built its AI reputation on, in some cases hiring researchers directly out of DeepMind. Anthropos’s own revenue run rate surpassed $65 billion by the end of July 2026, according to Bloomberg, up sharply from the start of the year, and the company has struck a series of separate multibillion-dollar infrastructure deals in recent months, including a $9.1 billion compute agreement with Riot Platforms, as it races to secure capacity for its growing customer base. OpenAI's ChatGPT had already crossed 1 billion monthly active users in June 2026, weeks before Gemini reached the same mark, meaning Google was catching up on the consumer-scale metric it most wanted to claim.

For a CEO whose company's stock price is now substantially tied to confidence in its AI execution, a stalled flagship model and a visible outflow of senior researchers is not a minor operational headache. It sits in tension with the story Pichai has been telling investors: that Google's AI infrastructure spending, which Alphabet raised twice in 2026 alone, will translate into products that actually win.

The decision: what changed, and what didn't

Precision matters here, because "AI shakeup" headlines tend to blur several distinct moves into one.

What changed: Hassabis moved from CEO of Google DeepMind, a role with full operational authority over the lab, to chairman of DeepMind and chief scientist of Alphabet, a role Alphabet created for him. Kavukcuoglu, previously DeepMind's CTO and already carrying the internal title of Alphabet's chief AI architect, was elevated to senior vice president of Google DeepMind with day-to-day operational control, including ownership of the Gemini 4 program. Kavukcuoglu's reporting line runs to Pichai, not to Hassabis. That is a structural change, not a titular one: it removes a layer between Google's CEO and the executive actually shipping Gemini models. Kavukcuoglu is not an outside hire; he has been at DeepMind since its early days, leading reinforcement learning research before moving into engineering leadership, meaning the CEO now has an operator already familiar with the Gemini codebase and the team building it.

What didn't change, at least not according to Google's own account: Hassabis is not leaving the company. He continues to lead Isomorphic Labs, Alphabet's AI drug-discovery subsidiary, and Google has framed his new Alphabet-wide chief scientist role as freeing him to focus on frontier research questions, including, in his own words, work he believes is now bringing artificial general intelligence within reach, rather than the daily grind of running a several-thousand-person organization.

What is a separate but simultaneous event: Dean's departure was not, according to any primary source reviewed for this article, presented by Google as a consequence of the Hassabis-Kavukcuoglu restructuring. The two were announced in the same memo and on the same day, which invites readers to connect them, but the available evidence supports a narrower claim: Google restructured its DeepMind leadership and lost its chief scientist concurrently, not because one caused the other. Dean has said publicly, including in comments to the New York Times, that operating outside a public company would give him more freedom to make decisions not strictly bound by near-term financial considerations — a rationale about his own next chapter, not a critique of Google's restructuring.

The tone of Pichai's memo emphasized continuity. In it, he credited Dean and Ghemawat with having helped drive major technology transitions, from Google's early search infrastructure to the neural networks behind the modern AI era — language that credits their legacy directly, even as it accompanied their departure. Whether that framing was intended to soften the news or simply reflects Pichai's own assessment of their contributions is not something the available record can settle; what's confirmed is the wording itself, and the fact that it appeared in the same memo announcing the restructuring.

Why this, why now

Three forces were converging on Pichai by early August. None of Google's public statements ties the restructuring to a specific combination of causes, but taken together, these three developments offer a plausible, if unconfirmed, explanation for the timing.

Speed. A missed model deadline in a market where rivals ship monthly is a competitive liability that compounds. Centralizing Gemini development under one SVP who reports directly to the CEO, rather than through a DeepMind CEO who also carried broader research and safety responsibilities, shortens the decision-making chain going forward. Google has not said its previous leadership structure was itself responsible for Gemini 3.5 Pro's delay; the company has attributed the delay publicly only to ongoing testing.

Capital discipline. The restructuring landed three weeks after Alphabet's Q2 2026 earnings call, where the company raised its full-year capital expenditure guidance a second time in 2026, to a range of $195 billion to $205 billion, and reported negative free cash flow for the first time in its history as a public company. Finance chief Anat Ashkenazi told analysts the increase reflected accelerating demand for AI infrastructure, and pointed to Google Cloud's backlog, which reached $514 billion in the quarter and grew roughly $50 billion sequentially, as evidence the spending was being met with real customer commitments rather than speculative build-out. Search revenue also grew 17% year over year to $63.3 billion in the same quarter, and Google Cloud revenue grew 82%. A company spending and growing at that scale has less tolerance for organizational friction that could slow down the models the infrastructure is meant to run.

Talent retention, imperfectly executed. Pichai reportedly held multiple conversations attempting to keep Dean and his colleagues at Google before their departure became final, an effort that did not succeed. That detail matters for how this decision should be read: this was not a restructuring Pichai chose from a position of full control. It followed, at least in part, a departure he reportedly tried and failed to prevent, and was accompanied by a forward-looking announcement about who leads what next.

The trade-offs

The case for Pichai's move is straightforward. A flatter reporting line for the executive building Gemini models removes a layer between that work and the CEO, which could speed up decision-making going forward, regardless of what specifically caused Gemini 3.5 Pro's delay. Hassabis, freed from day-to-day management, can concentrate on the frontier research and safety questions he has said explicitly he wants more time for, while retaining influence as chairman and Alphabet's chief scientist rather than exiting the company entirely, which limits the reputational and knowledge loss a full departure would have caused. And Google secured a continued relationship with Dean's new venture, as an investor and cloud partner, rather than losing him to a direct competitor.

The case against it is equally real. Removing a technical founder-CEO from operational control of a research lab, even amicably, risks unsettling researchers who joined DeepMind in part because of Hassabis's direct leadership, at a moment when competitors are actively recruiting those same researchers. Dean's exit removes 27 years of institutional knowledge and, by multiple accounts, significant internal credibility at a time when Google needed to project stability in its AI organization. And the restructuring does not, on its own, resolve the technical problems reportedly behind the Gemini 3.5 Pro delay. Centralizing authority under Kavukcuoglu may speed future decisions, but it does not retroactively fix a model that outside reporting says needed rebuilding work.

What's confirmed, and what's still just reported

The coverage of this story has not always separated documented fact from attributed explanation cleanly, so it's useful to do that here.

Confirmed, from Google's own materials and Pichai's own words: the reporting-line change, Hassabis's new titles, Kavukcuoglu's promotion and direct line to Pichai, Dean and Ghemawat's departure to found Discovery Loop, and Google's role as a founding investor and cloud partner in that venture.

Reported, but resting on unnamed sources rather than company confirmation: the claim, made by some outlets citing internal sourcing, that low morale inside DeepMind contributed to the Gemini 3.5 Pro delay. Google has not confirmed this explanation, attributing the delay publicly only to ongoing testing. Readers should treat the morale explanation as one contested account among several, not as an established fact.

The market's answer, so far

Alphabet shares fell on the day of the announcement, with reporting citing Bloomberg placing the stock at roughly $362 in New York trading, a decline in the range of 4% to 5% depending on the exact measurement window. That is a meaningful, if not catastrophic, one-day reaction for a trillion-dollar company. Some coverage described the overhaul as "seismic," a characterization that points to one plausible reading: that investors were reacting not just to a leadership change, but to its coincidence with the departure of a 27-year chief scientist from operational responsibilities. That remains an interpretation rather than a confirmed explanation, since Alphabet has not attributed the share-price move to any single cause.

Two weeks on, the more important test is still unresolved. Gemini 3.5 Pro had not shipped as of mid-August, with Google continuing to describe it only as being tested with partners. Kavukcuoglu has publicly taken ownership of the Gemini 4 roadmap. Whether the new, shorter reporting line produces a faster, more reliable release cadence, or whether the departure of Dean and his colleagues turn out to be the first of further senior exits, is the open question the restructuring was designed to answer. It hasn't been answered yet.

What executives should take from this

A CEO can restructure leadership without a full public break, but only if the departing executive stays inside the tent. Pichai kept Hassabis at Alphabet, in a role with real scope (chief scientist, Isomorphic Labs), rather than losing him entirely. That preserved institutional knowledge and avoided the reputational cost of a founder's outright exit, while still giving operational control to an executive already embedded in the Gemini engineering team.

Centralizing a reporting line is a bet on speed, not a guarantee of it. Putting Kavukcuoglu directly under Pichai shortens the chain of command, but the technical problems reportedly behind Gemini 3.5 Pro's delay don't disappear because an org chart changed. The restructuring buys the potential for faster decisions; it doesn't buy the fix itself.

Retention efforts that fail still shape the response. Pichai reportedly tried to keep Dean and did not succeed. The lesson for other leaders isn't that retention attempts are pointless. It's that a CEO's next move after a failed retention effort matters as much as the effort itself — in this case, the departure was addressed publicly with a structured, forward-looking leadership announcement rather than left unaddressed.

Pairing bad news with a structural change affects how both are read. Dean's departure and the leadership restructuring were announced together, in the same memo. Whether by design or coincidence of timing, the effect was that the exit read as part of a broader reorganization rather than standing alone as an isolated departure. Whether that impression holds up depends entirely on what Google DeepMind ships next.