Skip to content
NewsLeadership

Hassabis steps back at Google DeepMind as Jeff Dean and three researchers depart to co-found Discovery Loop

· by Pondero Newsdesk

The short version

On August 5, Demis Hassabis handed Google DeepMind's day-to-day operations to Koray Kavukcuoglu and became Alphabet's chief scientist, while Jeff Dean and three senior colleagues left to co-found Discovery Loop, a public benefit corporation automating scientific research, with Alphabet as a founding investor and cloud partner.

Hassabis steps back at Google DeepMind as Jeff Dean and three researchers depart to co-found Discovery Loop

Four of the researchers who built much of the infrastructure underlying modern AI left Google together on August 5. Demis Hassabis, who co-founded DeepMind and has led it since its acquisition by Google in 2014, stepped back from daily operations the same day. Alphabet announced it would serve as a founding investor and cloud partner in the departing researchers' new startup, named Discovery Loop.

What happened

Hassabis became Chairman of Google DeepMind and Chief Scientist of Alphabet, handing day-to-day leadership to Koray Kavukcuoglu, per CEO Sundar Pichai's memo posted on the Google blog. Kavukcuoglu, who had served as both DeepMind's chief technology officer and Alphabet's chief AI architect, takes the title of SVP of Google DeepMind and reports directly to Pichai. Per Pichai's memo, Kavukcuoglu has worked at DeepMind for over 13 years and started its deep learning team; his technical credits include WaveNet and DQN.

In his own note to staff, also published on the Google blog, Hassabis said he believes AGI is "close at hand" and cited that belief as the reason for stepping away from operational management. He said he would spend more time at Isomorphic Labs, the Alphabet unit working on AI-powered drug discovery. Hassabis shared the 2024 Nobel Prize in chemistry for AlphaFold per Fortune's reporting.

Jeff Dean, Google's chief scientist and a 27-year Google veteran, departed alongside three colleagues: Sanjay Ghemawat (Google Senior Fellow), Oriol Vinyals (Google DeepMind research VP), and Quoc Le (Google Brain co-founder). Pichai's memo named Dean and Ghemawat explicitly, describing their work together as spanning "our early search infrastructure to the neural networks that helped create the modern AI era." The four co-founded Discovery Loop, a public benefit corporation based in Palo Alto and confirmed in Pichai's memo, which described Alphabet as a "founding investor and Cloud partner." Radical Ventures and Khosla Ventures are also listed as backers per GeekWire.

Alphabet shares fell about 4% on the combined announcement per GeekWire.

What Discovery Loop is building

Discovery Loop's stated goal is to automate the scientific experimental loop: propose an experiment, run it, evaluate results, and iterate, running thousands of such cycles in parallel. Starting with ML research, the startup plans to treat itself as its first customer before expanding to other scientific domains, per its website. Longer term, Discovery Loop has named National Academy of Engineering Grand Challenges as targets, including engineering better medicines and making solar energy economical.

Collectively, the four founders' credits span much of the scaffolding of present-day AI: MapReduce, BigTable, TensorFlow, Pathways, TPUs, AlphaFold, AlphaStar, word2vec, sequence-to-sequence models, and chain-of-thought reasoning, per the company's own listing.

Dean told Wired the idea came together only in recent weeks per GeekWire's account of that interview. He added that operating outside a public company gives the team room to make decisions not driven by immediate financial returns.

Why it matters

How the Kavukcuoglu appointment affects the Gemini roadmap is the most immediate operational question for AI platform teams. He has run DeepMind's technical operations for over 13 years and, per Pichai's memo, has been deeply involved in Gemini model development. That suggests more continuity than the title change implies.

The surrounding context is harder to set aside. Fortune reported that Gemini 3.5 Pro, the next flagship model, was months behind its original June launch target when this announcement landed. Google also lost Gemini co-lead Noam Shazeer to OpenAI and Nobel laureate John Jumper to Anthropic in recent months. Dean, Ghemawat, Vinyals, and Le now extend that departure list.

Alphabet's position here is structurally unusual. Google is funding a startup formed by four of its most senior researchers, keeping them as cloud customers and collaborators while they build AI systems that could attract top research talent that might otherwise join Alphabet units. Pichai's memo framed this as a deliberate choice, noting plans to "collaborate on a research framework for ML systems and related infrastructure advances."

For teams choosing between Gemini and rival foundation models, Kavukcuoglu's first substantive decisions on delivery cadence will be the relevant signal. The operational leadership is now in the hands of someone who built the lab's technical capabilities over 13 years, so the question is execution pace, not institutional knowledge.

Context and reactions

Hassabis's note to staff cited the Gemini App reaching over 950 million monthly users per his note on the Google blog and referenced the as-yet-unreleased Gemini 4 by name, describing DeepMind as "the AI engine room of Google" with "the full stack" from chips through applications. His note did not address the delayed flagship model or prior researcher exits.

Pichai framed the departures positively in his public letter: "After an incredible 27-year run, Jeff Dean is at a moment where he wants to try something new, and we're excited to support him in that." He added that Google would continue to work with Dean and Ghemawat as a founding investor and research collaborator.

What to watch next

Gemini 3.5 Pro's release is the most immediate product milestone. Whether Kavukcuoglu accelerates or extends the current delay will be the first clear signal of how the leadership change affects output cadence. Discovery Loop has said it plans to start with ML research and publish results as it validates its automated-experiment approach. Any public output from the startup in the next 90 days would give external researchers a first look at the methodology and confirm the team's pace coming out of the gate.

Sources