Jeff Dean and other top AI researchers are leaving Google to launch their own startup

Jeff Dean, one of the longest-serving and most influential executives at Google, is stepping down from the search giant to launch his own AI startup.
Coming with him as co-founders are several other top researchers at the company, including Sanjay Ghemawat, a top engineer and senior fellow at Google; Quoc Le, a key AI researcher and founding member of Google Brain; and Oriol Vinyals, a senior research scientist at Google DeepMind. Dean reportedly plans to serve as CEO.
Together, the group is starting Discovery Loop, a public benefit corporation that seeks to use AI to turbo-charge scientific research.
Discovery Loop says it plans to use high-octane algorithms to initiate and iterate thousands of experiments simultaneously, with the goal of partially automating the research process and expanding the scale at which experimentation can be conducted.
The startup is also interested in using AI to help create more powerful AI (a process known as recursive self-improvement ), which would cut human iteration out of the loop entirely.
“While science and engineering have tremendously advanced society over past centuries, progress has traditionally relied on slow, sequential human iterations, creating a significant bottleneck,” the company said in a press release. “Discovery Loop is developing advanced AI systems that leverage massive computational scale to fundamentally transform the speed and efficiency of innovation by automating complete experimental loops.”
Using AI to accelerate scientific discovery has been a key interest of the science and tech communities for many years, but until recently, it remained a largely experimental field with limited commercial application.
The company has received financial support from a number of different sources, including Google’s parent company Alphabet. The initial funding round is being co-led by Radical Ventures and Khosla Ventures, the company announced Wednesday. Kleiner Perkins, Lightspeed, and Doerr Capital also participated.
“The next great frontier for AI is to go beyond answering questions and to begin making discoveries,” the founding team said in a joint statement. “By fundamentally accelerating how engineering and scientific discovery are conducted, we can deliver the benefits of transformative technologies to the world far sooner.”
Dean has worked at Google since 1999 and was Google’s 30th employee. Over the years, he has contributed to Google search’s core infrastructure , including its crawling and indexing system and its query-serving system. He has also had a significant impact on Google Gemini’s multimodal models and played a leading role in the company’s early AI research.
“We think there is opportunity for AI to more fully automate what has traditionally been a very human-intensive experimental loop,” Dean told the New York Times . “You will get both a higher quantity and a higher quality of experiments, and that will lead to scientific breakthroughs and advances.”
When you purchase through links in our articles, we may earn a small commission . This doesn’t affect our editorial independence.


Scale faster. Grow your portfolio. Gain practical expertise. No matter your goal, Disrupt can empower you. Save up to $330 toda y!
Bending Spoons to buy Airtable for $1.28B
Influencers draw backlash for attending OpenAI’s first luxury trip
Sequoia’s Shaun Maguire leads $1B round for nuclear startup Valar Atomics
Malaysia is reportedly shutting down Balaji Srinivasan’s Network School
YouTuber Hank Green says his AI usage is ‘not healthy’
VC-backed startups commit more fraud, and researchers think they know why
WhatsApp is testing a new folder for messages from large businesses
Verified source · TechCrunch
Reported by TechCrunch. Open the original for full media and formatting.
More in Research
All newsThe LLM Proposes, the Executive Disposes: A Self-Verifying Agent Instrument that Dissociates Commitment Drift from Binding Drift in Long-Horizon Agents
How do you verify a long-horizon agent when its own state and self-reports are exactly what you cannot trust? We present an agent instrument built so that verification is structural rather than post-hoc. A deterministic Executive owns all belief; a language model may only file typed proposals, and a claim is admitted only when a prediction pre-registered before acting is matched against observation by code. Two properties make the instrument a verifier of its own science, not just of the agent: every run invalidates itself when per-organ write-error, render-size, or salted-canary-echo floors…
Read at arXiv cs.AIMonte Carlo Tree Search for Table-to-Multimodal Report Generation
Automatically generating professional multimodal reports comprising both textual analysis and visual charts from structured tabular data is a critical challenge in data intelligence. Existing methods suffer from fixed linear pipelines and isolated subtask processing, which hinder joint optimization of factual accuracy, visual quality, and narrative coherence. To address these issues, this paper proposes MCTS-Report, a Monte Carlo Tree Search (MCTS)-driven framework that formulates multimodal table-to-report generation as a progressive construction process over a structured search space. The c…
Read at arXiv cs.AIA Long-Run Persistence Theory for AI Systems under the Redundancy-Adjusted Artificial Age Score (AAS)
Artificial intelligence systems are increasingly expected to operate over repeated cycles of interaction, adaptation, and update rather than through isolated one-shot outputs. This raises a fundamental theoretical question: can an AI system persist indefinitely without incurring unbounded structural aging? This paper develops a long-run persistence framework for AI systems based on the redundancy-adjusted Artificial Age Score (AAS). The model extends AAS from a static evaluative measure into a cycle-level functional that generates an age sequence across repeated operation. At each cycle, stru…
Read at arXiv cs.AIFinProBench: Evaluating Financial AI Agents with Role-Grounded Rubrics Derived from Professional Deliverables
Evaluating financial AI agents requires criteria aligned with real professional work. Existing rubric methods typically derive criteria from task prompts or model outputs, overlooking tacit standards visible only in practitioner deliverables. We introduce FinProBench, a benchmark for professional financial tasks, and Role-Grounded Rubric Construction (RGRC), a reusable pipeline that derives rubrics from deliverables produced by practitioners in the same role. RGRC comprises four stages: Deliverable Collection, Competency Extraction, Rubric Synthesis, and Validation. Its rubrics capture tacit…
Read at arXiv cs.AI