Discovery Loop
Category: AI Agents
AI research startup founded by four ex-Google DeepMind leaders (Jeff Dean, Sanjay Ghemawat, Quoc Le, Oriol Vinyals) building AI systems to automate the scientific method — running thousands of experiments in parallel to accelerate discovery in science and engineering. Discovery Loop was founded in 2026. The company is led by Jeff Dean. Based in Palo Alto, California, USA. Team size: 1-10. Total funding raised: Undisclosed. Latest round: Seed. Key investors include Radical Ventures, Khosla Ventures, Lightspeed, Kleiner Perkins, Doerr Capital, Alphabet.
- Founded
- 2026
- Headquarters
- Palo Alto, California, USA
- Team size
- 1-10
- Total funding
- Undisclosed
Value proposition
Automating the scientific method so AI acts as the researcher rather than just a research tool — enabling parallel execution of thousands of experiments to dramatically accelerate scientific and engineering discovery.
Products and solutions
Building AI systems that automate complete experimental loops (propose → implement → run → evaluate) to run thousands of experiments in parallel. Initial focus on automating machine learning research (acting as its own first customer), then expanding to chip design, biology, drug discovery, materials design, and clean energy. Pursuing recursive self-improvement (AI that improves its own capabilities).
Unique value
A GOAT-tier founding team (three of the most-cited AI researchers and two of the most-cited distributed systems researchers) with full-stack depth spanning silicon (TPUs), foundation models (Gemini), and products used by billions, now applying that expertise to automate discovery itself.
Target customer
Initially itself (automating its own ML research); later small research teams and organizations seeking to out-invent large research institutions, plus domains in chip design, biology, drug discovery, materials design, and clean energy.
Industries served
Machine learning research, chip/semiconductor design, biology, drug discovery, materials design, clean energy, engineering
Technology advantage
Founding team built the modern AI stack (Google Search, TPUs, MapReduce, TensorFlow, AlphaFold, Gemini). Backed by Alphabet as founding investor with guaranteed cloud compute for the first year. Seed co-led by Radical Ventures and Khosla Ventures (amount undisclosed). Public benefit corporation structure.
How they differentiate
Unlike AI tools that assist human researchers, Discovery Loop aims to make AI itself the researcher — fully automating the experimental loop and running thousands of experiments in parallel, backed by a founding team that built the modern AI stack (TPUs, TensorFlow, Gemini, MapReduce).
Main competitors
FutureHouse, Sakana AI, AI-for-science efforts at OpenAI, Anthropic, and DeepMind (e.g. AlphaFold), other AI-driven scientific discovery startups
Key partnerships
Alphabet/Google (founding investor and Cloud partner, supplying compute for first year), Radical Ventures (co-lead investor; Jordan Jacobs joining board), Khosla Ventures (co-lead investor)
Major milestones
Aug 5, 2026: Founded and publicly launched by Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals after departing Google; announced seed round co-led by Radical Ventures and Khosla Ventures (amount undisclosed) with Alphabet as founding investor and cloud partner; Alphabet/Google stock dropped nearly ~4% on the news.
Market positioning
Positioned as a frontier AI-for-science research company with arguably the strongest founding team in the industry, targeting the automation of scientific discovery across multiple domains.
Geographic focus
US (Palo Alto); global AI-for-science landscape
About Jeff Dean
Ex-Google Chief Scientist (Google DeepMind & Google Research); Co-founder of Google Brain; Technical co-lead of Gemini; Co-creator of TPU, MapReduce, Bigtable, TensorFlow. University of Washington (PhD, 1996); University of Minnesota (BS). Widely regarded as one of the greatest software engineers of all time.
Latest news about Discovery Loop
More AI Agents companies
Official website: https://www.discoveryloop.com