Google handed day-to-day control of DeepMind to Koray Kavukcuoglu, and Jeff Dean left after 27 years

Google is reorganizing the leadership of Google DeepMind, announced on August 5, 2026 in a post signed jointly by Sundar Pichai and Demis Hassabis. Hassabis is handing over day-to-day operational responsibility and becomes Chair of GDM and Chief Scientist of Alphabet while continuing to lead Isomorphic Labs, and Koray Kavukcuoglu, previously CTO and Chief AI Architect, becomes SVP of Google DeepMind reporting to Pichai with oversight of Gemini model development, Frontier AI research, and the Gemini app and developer teams. On the same day Jeff Dean ended a 27-year run at Google to launch an independent public benefit corporation with Sanjay Ghemawat, which TechCrunch reports is called Discovery Loop. ASAP works from Google's official post and TechCrunch's reporting to lay out what changed and what it means.

What Hassabis gave up is operations, not standing

Demis Hassabis is stepping back from day-to-day operational responsibility at Google DeepMind, writing on August 5, 2026 that now is the right time to hand those duties over. His two new titles are Chair of GDM and Chief Scientist of Alphabet, and he continues to run Isomorphic Labs, the drug discovery company.

The move is a split of roles rather than an exit or a demotion. The seat he leaves is the one that runs the organization; the two he takes are the one that sets research direction and the one that represents Alphabet's science agenda. Google DeepMind itself was formed in 2023 when Google Brain and DeepMind merged, and Hassabis has led the combined organization since.

Gemini models and the Gemini app now sit under one executive

Koray Kavukcuoglu is now SVP of Google DeepMind and reports directly to Sundar Pichai. The announcement names three areas under him: Gemini model development, Frontier AI research, and the Gemini app and developer teams. He spent 13 years at DeepMind, most recently as CTO of Google DeepMind and Chief AI Architect.

Consolidating those three areas under one executive is the substance of the reorganization. When the group building the models, the group shipping them in an app, and the group selling them as APIs each report through a different line, priority conflicts have to travel upward before anyone resolves them. The scale involved is visible in the two figures the post cites: the Gemini app reached 950M+ monthly users, and Gemma models have surpassed 900M+ downloads.

What Jeff Dean and Sanjay Ghemawat are building

Jeff Dean and Sanjay Ghemawat are leaving Google to launch an independent public benefit corporation intended to accelerate discoveries in ML, science, and engineering. Google's own post does not name the company; TechCrunch reports the name as Discovery Loop and says Quoc Le and Oriol Vinyals are joining as well. Dean joined Google in 1999 as its 30th employee and built much of the company's search infrastructure and AI research foundation.

Dean told TechCrunch that raising both the quantity and the quality of experiments is what leads to scientific breakthroughs and advances. TechCrunch reports the company will work on automating experimental processes and on recursive self-improvement techniques, with an initial round co-led by Radical Ventures and Khosla Ventures and participation from Kleiner Perkins, Lightspeed, and Doerr Capital.

The reorganization shortens the path between research and product

The real change in the August 5, 2026 reorganization is the number of reporting steps between research and product. Putting model development, the consumer app, and the developer platform under one executive who reports to the CEO removes a layer from the path a decision travels when a model's capability targets collide with a product's ship date.

The nature of the seats supports that reading. Chair and Chief Scientist are direction-and-representation roles; SVP is a resource-allocation-and-schedule role. Splitting work that one person previously held on both sides suggests the company's current constraint is execution speed rather than research agenda.

That structure carries a cost. When model research and product organizations share a chain of command, quarterly product targets tend to encroach on long-horizon research schedules. Google DeepMind's distinguishing strength since the 2023 merger has been running basic research and productization in parallel, so how this structure preserves that balance is the standard by which it should be judged.

Alphabet converted a departure into an investment relationship

Alphabet is turning the departure of Jeff Dean and Sanjay Ghemawat into an investor and infrastructure relationship rather than absorbing it as a loss. The joint post describes three forms the relationship takes: founding investor, Cloud partner, and collaboration on a research framework.

The framing signals a shift in how talent departures are handled. Senior researchers leaving to found companies is not new, but a joint announcement that spells out investment and infrastructure supply redefines the exit as a portfolio position. The Cloud partner clause matters most here: when a compute-hungry startup runs on Alphabet infrastructure, spending that used to be payroll returns in part as cloud revenue.

The legal form deserves attention too. A public benefit corporation can write purposes beyond shareholder return into its charter, which gives a company chasing scientific discovery a basis for distance from short-term revenue pressure. The form does not guarantee the practice, though. What actually differs will show up in the terms under which this company publishes its results.

What teams in Korea running Gemini should watch

Teams in Korea building on Gemini APIs and Gemma weights are facing one concrete change: the product roadmap now sits with a single organization. When models, the app, and developer tooling share a reporting line, API specifications and app features have more room to ship on a coordinated cadence.

Nothing here changes contracts or pricing on its own. The post says nothing about pricing tiers, regional availability, or enterprise terms, so revising procurement plans on the basis of an org chart goes beyond the evidence.

The signal worth tracking is narrower. Gemma passing 900M+ downloads shows how much of Google's developer surface runs on open-weight models, and moving that group under the same executive as the Gemini app creates a place where the division of labor between open weights and the commercial API can be renegotiated. Teams self-hosting Gemma have a new reason to read the release terms and license text for the next generation more closely than before.

What the announcement does not answer

The August 5, 2026 post is an org chart announcement, and it leaves research budgets and headcount moves unexplained. The first gap is scale: the official post does not say how many people are leaving alongside Dean and Ghemawat, or which teams they come from.

The second gap is the substance of the research collaboration. The joint post commits only to collaborating on a research framework. Who owns what results, and on what terms joint work reaches Google products, cannot be read from that sentence.

The third gap is timing. The post does not state when the new structure takes effect or which decisions Hassabis stays involved in as Chair. Boundaries between a chair and an SVP are drawn in places an org chart does not show, and they become visible at the first real priority conflict.

Source: Google's official blog post "The next chapter of our AI momentum" (signed by Sundar Pichai and Demis Hassabis, August 5, 2026) and TechCrunch reporting (August 5, 2026), compiled by ASAP.