
16 Nov 2026
Stanford's Yejin Choi and Carlos Guestrin launch Noeri with $110 million
Noeri builds software that studies a company's own data and workflows, then proposes improvements to its AI system, with a human approving each change like a code review. Menlo Ventures led the round.
Noeri is a $110 million bet against the one-model-to-rule-them-all future, and Nvidia and Snowflake are hedging right alongside it. If companies can get frontier-ish results from systems they own, the big labs' pricing power gets a lot more negotiable.
On Monday, Nov. 16, Noeri launched with $110 million. Menlo Ventures led. Madrona, Kindred Ventures, Wing Venture Capital, and NEA joined. Cisco Investments, Nvidia, and Snowflake also invested. The company did not disclose a valuation, the price the round puts on the whole company. GeekWire says Bloomberg reported the funding first. The round size and the investor names are Noeri's, in the Business Wire release. The missing valuation and the Bloomberg line are GeekWire's.
The founders and co-CEOs are Yejin Choi and Carlos Guestrin, Stanford professors who spent years at the University of Washington. Choi is a 2022 MacArthur Fellow, a no-strings grant for people the foundation picks out. She led common-sense reasoning research at the Allen Institute for AI. GeekWire says she was a distinguished scientist at Nvidia until she started Noeri in May. Guestrin helped create XGBoost, a widely used machine learning tool, co-founded Turi, which Apple acquired in 2016, and later led Apple's central machine learning and AI organization. Machine learning, here, is software that learns from examples instead of following a rule someone typed for every case. The co-CEO line, the 2022 fellowship year, and the Nvidia role are GeekWire's. The Turi sale, XGBoost, and the Apple job are in the company release.
What the software does, in the company's words. It studies a company's own data and workflows, then proposes changes to the AI system around a specific task: the instructions, the tools, which model handles each step, and sometimes the models themselves. Each change is explained and sent to a person for approval, like a code review. A code review is the pass a teammate makes over a change before it ships. The system works with closed models and with open-weight models, models whose weights a customer can run. Customers keep ownership of the workflows and models they build, subject to the licenses on any third-party model in the mix. Those lines are Noeri's.
Where the idea came from, as GeekWire reports what Choi told Bloomberg. It grew partly out of helping a Stanford microbiology and immunology team use OpenAI's models more cheaply. The small model they built learned to use older OpenAI models and, Choi said, outperformed GPT-5 on that team's research needs. That comparison is her account of one lab's work. It is not a public benchmark, a shared test anyone else can rerun.
The name means "intelligence, in the plural," the release says. In a launch essay, the founders wrote: "Few companies would build their core operations on a single consultant who also advises every competitor, can change its terms at will, and can never be brought in-house. A single, centralized AI offers exactly that arrangement." GeekWire is the one that prints that essay quotation.
Noeri is based in Palo Alto, with an office in Zurich. That line is the company's. GeekWire, citing a July report from The Information, names two others in the work of tailoring models to a customer: Thinking Machines Lab's Tinker, and Applied Compute. Those names are GeekWire's comparison, not a list Noeri published.
The picture is two side-by-side portraits, Yejin Choi on the left and Carlos Guestrin on the right. They are company photographs of the co-CEOs. The credit is Noeri, via GeekWire.
In plain terms, two Stanford professors who used to work at the University of Washington started a lab that says it will tune a customer's AI system to that customer's own work, and make a person approve every change. They raised $110 million, led by Menlo Ventures, with Nvidia, Snowflake, and Cisco in the round. They did not say what the company is worth. The line that a small model beat GPT-5 is Choi's, about one research team's needs.