IA · 5 August 2026 · 3 min read
Major Shakeup at Google AI: Demis Hassabis Steps Aside as Gemini Architects Launch Discovery Loop
In brief: Google announced a major AI leadership overhaul: Demis Hassabis steps down as head of Google DeepMind to become Alphabet's chief scientist, with Koray Kavukcuoglu taking charge of DeepMind. Concurrently, four foundational researchers — including Jeff Dean and Oriol Vinyals — are leaving Google to launch an AI startup called Discovery Loop.
by Team Mocchi's
Google has announced one of the most significant leadership reorganizations in the history of its artificial intelligence laboratories. Demis Hassabis, founder of DeepMind and leader of Google's unified AI division, is stepping down from operational management to become chair of Google DeepMind and chief scientist at Alphabet. Executive leadership transitions to Koray Kavukcuoglu, former chief technology officer of the division, who will report directly to CEO Sundar Pichai.
Concurrently, the Mountain View giant is managing a high-profile departure of veteran research talent: four of Google's most influential scientists are leaving the company to launch an independent laboratory named Discovery Loop.
Restructuring at the Top of Google DeepMind
As reported by The Verge, the executive shift redistributes primary responsibilities across Alphabet. Hassabis will focus on strategic efforts toward artificial general intelligence (AGI) and lead Isomorphic Labs, Alphabet's drug discovery subsidiary. In a memo to staff, Hassabis emphasized that the primary application of AI should be advancing human health and curing complex diseases like cancer.
Koray Kavukcuoglu steps up as Senior Vice President of Google DeepMind while retaining his role as chief AI architect. He will oversee the engineering operations driving the Gemini model family and the integration of AI across Alphabet's core product ecosystem.
The Exodus of Google's AI Brains: Discovery Loop
The leadership transition at DeepMind coincides with the departure of four key figures in modern machine learning. Jeff Dean, Google's 30th employee, co-founder of Google Brain, and chief scientist of Google DeepMind, is leaving after nearly 27 years. Joining him in departing the company are Google Fellow Sanjay Ghemawat, Oriol Vinyals (VP of research at DeepMind and technical co-lead on Gemini), and Quoc Le (Google Brain co-founder and creator of AutoML).
The four researchers announced the creation of Discovery Loop, a public benefit corporation focused on automating scientific and engineering workflows. Despite the major split, Google will serve as a founding investor in the new enterprise.
According to reporting by WIRED, the venture materialized over just a few weeks following shared discussions among the long-time collaborators. Dean had previously hinted at the concept during an appearance at Y Combinator's Startup School, outlining a vision where autonomous AI systems conduct thousands of experimental loops to accelerate scientific breakthroughs.
Automating the Scientific Method
Discovery Loop's core mission centers on developing AI systems capable of proposing hypotheses, designing experiments, running trials, and analyzing outcomes autonomously. The startup's initial application will target AI development itself, using automated research loops to invent novel machine learning architectures beyond traditional Transformer models.
Once established, Discovery Loop plans to deploy its automated platform across semiconductor design, synthetic biology, and material science. The initiative highlights a broader shift in frontier AI research: moving from pure parameter scaling toward automated logical discovery and experimental reasoning.
Mocchi's take
Google's executive restructuring and the formation of Discovery Loop signal a pivotal evolution in the AI landscape. The industry's scientific frontier is shifting away from general-purpose generative models — where architectural differentiation is narrowing — toward automated reasoning and scientific discovery engines. The departure of foundational figures like Jeff Dean underscores that small, highly focused research teams are increasingly seen as the optimal vehicle for breakthroughs beyond current Transformer paradigms. For enterprises and technology leaders evaluating AI roadmaps, this shift illustrates where long-term value will emerge: not in consumer-facing chatbots, but in autonomous, domain-specific execution loops capable of transforming complex engineering, R&D, and industrial design processes.