Google’s recent leadership shuffle has sent four senior engineers to launch Discovery Loop, a startup aimed at automating scientific research, while DeepMind founder Demis Hassabis moves into a broader strategy role.
Leadership changes at DeepMind
Koray Kavukcuoglu, DeepMind’s chief technology officer and Google’s chief AI architect, will assume responsibility for the Gemini model line, frontier research, the Gemini app and its developer teams. Kavukcuoglu, a DeepMind veteran of 13 years, helped launch the deep‑learning group and contributed to projects such as WaveNet and DQN.
In a recent interview on CNBC’s podcast The Tech Download, Hassabis described DeepMind as the “engine room” for Google’s AI work, noting that the lab builds core technology before it is distributed across the company’s products.
Four engineers depart for Discovery Loop
Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le are leaving Google to form Discovery Loop, a public‑benefit corporation that intends to use AI to automate the full experimental cycle of scientific and engineering research. Google will stay involved as a founding investor and cloud‑services provider.
Dean, who joined Google in 1999 and helped create Google Brain, was a technical co‑lead for Gemini and co‑designer of TensorFlow. Ghemawat’s work includes the infrastructure behind Search, MapReduce, Bigtable and Spanner. Vinyals and Le contributed to the 2014 sequence‑to‑sequence paper that introduced a key architecture for large language models.
Discovery Loop’s website states, “We are building AI solutions that can automatically solve important problems in machine learning, science, and engineering.” The startup plans to leverage the founders’ experience to create specialized infrastructure for research, separate from Google’s existing product‑focused systems.
Related: AI workflows get native payment support
From a practical standpoint, the move could give researchers a more streamlined path from hypothesis to result, reducing the time needed to set up experiments and analyze data. If the startup succeeds, it may lower the cost barrier for academic labs that lack large‑scale computing resources, potentially accelerating discovery in fields ranging from materials science to drug development.
Impact on Google’s AI roadmap
Alphabet’s stock reached its best performance since 2009 in December, buoyed by confidence in Gemini and the broader AI strategy. Gemini 2.5 debuted in March 2025, and Gemini 3 arrived in November, re‑establishing Google as a competitor to OpenAI and Anthropic.
Despite the departures, Google’s hardware plans remain tied to the Gemini roadmap. The company recently committed to a chip built for a single model, signaling a close coupling of silicon and software development.
Google’s cloud partnership with Discovery Loop will give the startup access to the computing capacity needed to run large numbers of experiments without constructing its own data centers. This arrangement also supplies Google Cloud with a new AI workload, reinforcing the firm’s position in the enterprise market.
Hassabis told employees that AGI now feels “close at hand,” emphasizing his desire to influence the next phase of development. The leadership change reflects Google’s aim to balance rapid product rollout with long‑term research goals.
The industry watches closely.
