Jeff Dean Leaves Google After 27 Years to Co-Found Discovery Loop

Google's chief scientist Jeff Dean departs after 27 years to co-found Discovery Loop, as Demis Hassabis steps back from day-to-day DeepMind leadership.

4 min
Jeff Dean Leaves Google After 27 Years to Co-Found Discovery Loop

Jeff Dean Leaves Google After 27 Years to Co-Found Discovery Loop

Google's chief scientist Jeff Dean is leaving the company after nearly 27 years to co-found Discovery Loop, a startup that aims to automate the full experimental cycle of scientific research. The departure, announced on August 5, is part of a broader restructuring of Google's AI leadership that also shifts Demis Hassabis out of day-to-day operations at Google DeepMind.

The departures

Dean is not leaving alone. Three other senior Google AI figures are joining him as co-founders: Sanjay Ghemawat, a Google Senior Fellow who collaborated with Dean on MapReduce, BigTable, Spanner, and TensorFlow; Oriol Vinyals, VP of research at Google DeepMind and a technical lead on Gemini; and Quoc Le, a co-founder of Google Brain and the researcher behind AutoML-Zero.

The group's combined track record spans some of the most consequential infrastructure in modern computing: Google File System, MapReduce, BigTable, Spanner, TensorFlow, TPUs, AlphaFold, Gemini, sequence-to-sequence models, and chain-of-thought reasoning.

Discovery Loop is incorporated as a public benefit corporation. Google is a founding investor, will serve as its cloud partner, and will supply compute for the venture's first year. Khosla Ventures and Radical Ventures are also backing the company, though the founders have not disclosed the round's size or valuation. Wired reported that Radical managing partner Jordan Jacobs will join the board.

What Discovery Loop plans to build

The company's mission is to build AI systems that can run the full experimental loop: proposing hypotheses, implementing and executing experiments, evaluating results, and iterating, at a scale sequential human effort cannot match. The plan has three stages.

First, Discovery Loop will focus on automating machine learning research itself, using frontier models and large-scale compute to propose, run, and learn from experiments. Second, it will use those automated capabilities to optimize its own technology stack, becoming its own first customer. Third, it intends to generalize to any field with measurable outcomes, initially targeting chip design, biology, drug discovery, and materials science, and eventually the National Academy of Engineering's Grand Challenges.

Quoc Le told Wired he hopes the system might discover new architectures. "I'm very excited about automating machine learning," he said. "It might be that we will discover a different transformer architecture."

Oriol Vinyals identified a key capability gap the effort must bridge: current models are not strong at generating genuinely novel hypotheses. "One of the things that we'll be obviously very focused on is how these models come up with new ideas to try," he said. "That's not something that currently they're super strong at."

The founders described early systems that will co-develop ideas with human researchers, with deeper automation as the longer-term goal.

The Google reshuffle

Alongside Dean's departure, Demis Hassabis is stepping back from day-to-day leadership of Google DeepMind to become its chair and Alphabet's chief scientist. Hassabis will continue to lead Isomorphic Labs, Alphabet's AI-driven drug discovery unit. In a message to staff, he said the "number one application of AI should be to improve human health" and that he wants to focus on demonstrating AI's value by helping to cure diseases.

Koray Kavukcuoglu, previously DeepMind's CTO and Google's chief AI architect, takes over as SVP of Google DeepMind, reporting directly to CEO Sundar Pichai. He will oversee Gemini model development, frontier AI research, and the Gemini app and developer teams.

Pichai disclosed metrics that illustrate the scale of what Kavukcuoglu inherits: the Gemini app has passed 950 million monthly users, and Gemma open models have surpassed 900 million downloads. Google will continue working with Discovery Loop as a research framework partner.

What it means

The simultaneous departure of four of Google's most senior AI researchers, including its chief scientist and two Google Brain co-founders, signals a structural shift in where top talent wants to work. Discovery Loop's public benefit corporation structure suggests the founders want freedom from quarterly earnings cycles, but Google's role as founding investor and compute provider also means the ties are not severed. The arrangement looks less like a clean break and more like a spin-out with a strategic anchor tenant.

The near-term observable is Discovery Loop's recruiting effort: the company is hiring a lean, in-person founding team and has no office yet. Its first milestone is the automated ML loop it plans to turn on its own stack. Whether it produces results the field recognizes as scientific discovery rather than optimization will determine whether the venture is remembered as a footnote or a fork in the road.

Sources

Jeff Dean Leaves Google to Automate the Scientific Method With Discovery Loop - Unite.AI: https://www.unite.ai/jeff-dean-leaves-google-to-automate-the-scientific-method-with-discovery-loop/

Google just announced a major shakeup of its top AI leadership - The Verge: https://www.theverge.com/tech/975677/google-deepmind-ai-demis-hassabis-shakeup

Our next chapter of AI momentum - Google Blog (Sundar Pichai): https://blog.google/company-news/inside-google/message-ceo/next-chapter-ai-momentum/

Jeff Dean's Next Act - Wired: https://www.wired.com/story/jeff-dean-google-discovery-loop-startup/

Written by

More to read

  • Anthropic Demonstrates Autonomous De Novo Protein Design and Chemical Analysis with Claude

    Anthropic Demonstrates Autonomous De Novo Protein Design and Chemical Analysis with Claude Anthropic has published experimental results demonstrating Claude's ability to autonomously design de novo protein binders with physical wet-lab validation and automate complex analytical chemistry workflows. The findings show frontier LLMs acting as autonomous agents across computational biology and molecular characterization pipelines. In the primary experiment, Anthropic evaluated Claude Mythos Previe

    1 min
  • Cerebras Unveils CS-4 Rack-Scale System Powered by Three WSE-3 Turbo Chips and Nexus Architecture

    Cerebras Unveils CS-4 Rack-Scale System Powered by Three WSE-3 Turbo Chips and Nexus Architecture Cerebras Systems has announced the CS-4, a rack-scale AI accelerator system designed around three of its next-generation Wafer Scale Engine 3 Turbo (WSE-3 Turbo) chips and a modular hardware architecture dubbed Nexus. Cerebras confirmed that initial customer shipments for the CS-4 are scheduled to begin in the current quarter. The new system marks a structural shift from Cerebras's single-wafer CS

    1 min
  • AI FinOps: Cutting LLM Inference Costs by 30-60% Through Model Tiering, Caching, and GPU Optimization

    AI FinOps: Cutting LLM Inference Costs by 30-60% Through Model Tiering, Caching, and GPU Optimization Inference costs have become the second-largest line item in enterprise AI budgets, trailing only talent spend according to RapidData's State of Enterprise AI 2026. This shift represents a fundamental inversion from the 2021-2023 era when training dominated AI expenditure. The compounding nature of serving costs—accumulating every hour as long as users hit the API—means that even modest producti

    1 min