Skip to main content
v2026.11,986 entries · CC-BY 4.0

Editorial · CASRAI · AI and ML research outputs

Stanford AI researchers launch Noeri with $110 million

Noeri launches with $110 million to build specialized AI that learns from an organization’s work, with enterprise and scientific ambitions.

Published 16 Nov 2026· 4 minute read

Ask CASRAI · free to try

Ask about this story

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

CASRAI is the reference for research administration — bookmark it for the next question.

Stanford computer scientists Yejin Choi and Carlos Guestrin launched Noeri on November 16, 2026, announcing $110 million to develop specialized AI for enterprise work and scientific discovery. The company’s pitch is that organizations should own systems that improve through use. For university researchers, the relevant questions are how that learning is evaluated, what ownership permits and which results have been demonstrated.

What Noeri announced

The company’s announcement, distributed by Business Wire, names Menlo Ventures as lead investor, with Madrona Venture Group, Kindred Ventures, Wing Venture Capital and NEA participating. Strategic investors include Cisco Investments, NVIDIA and Snowflake. The release identifies Choi and Guestrin as co-founders and co-CEOs.

Noeri says its system examines an organization’s data and workflows and proposes changes across instructions, skills, tools, model selection and, where useful, model parameters. It says proposed changes are explained and submitted to a person for approval. The release describes compatibility with closed and open-weight models, with customer ownership subject to third-party model licenses.

That license qualification matters for research procurement: ownership of a workflow or a trained artifact does not automatically remove conditions attached to an underlying model. Teams comparing open weights with hosted access should keep the model’s access arrangements separate from the rights promised by a service provider.

A product organized around learning after deployment

On its product page, Noeri describes four activities: learning from tasks and corrections, improving or creating tools, training models around a customer’s work, and tracking quality and cost against that customer’s traffic. These are the company’s descriptions of its product, rather than independent measurements of savings or accuracy.

The same page places Choi at Stanford as a computer science professor and senior fellow at HAI, with previous research leadership at Ai2 and NVIDIA. It identifies Guestrin as a Stanford professor and director of the Stanford AI Lab, a co-creator of XGBoost and a co-founder of Turi, which Apple acquired in 2016. Those biographies help explain the research focus without establishing how the new product will perform.

The research agenda behind the launch

Noeri’s research page sets out four related problems: automating the construction of AI systems, specializing with limited data and compute, learning after deployment without forgetting, and coordinating models to investigate new questions.

The page connects those goals to named research projects. It describes TTT-Discover as continuing training on the problem being solved, SDPO as learning from structured feedback such as test failures and runtime errors, and ToolOrchestra and AgentFlow as learning how to choose tools and models for multistep tasks. These descriptions identify the technical approaches Noeri says inform its work; they are not a substitute for evaluating the underlying studies.

For an academic group, the distinction between changing instructions, changing tools and changing model parameters is consequential. CASRAI’s assessment is that a model evaluation suite should record which component changed and compare the resulting system with a defined baseline. Otherwise, an apparent gain may be difficult to attribute or reproduce.

Ownership and discovery are central to the founders’ argument

In their vision essay, Choi and Guestrin argue for many specialized intelligences shaped by the organizations using them. They propose that owners should control where systems run, what they learn and who can learn from them. They also argue that smaller systems tailored to a problem can make sustained scientific search more affordable.

The essay says Noeri’s first proof-of-concept partnerships span enterprise and scientific domains. It frames discovery as an iterative process in which models propose, critique, test and refine ideas alongside researchers. These are stated aims and early partnership descriptions. They do not establish a generally available scientific platform or a validated discovery outcome for a prospective university customer.

What independent coverage emphasizes

Securities.io’s November 16 report confirms the financing and describes Noeri’s focus on improving agents already operating in production as well as building new ones. It characterizes the early partnerships as the setting in which the company must demonstrate customer-specific gains.

i10X’s separate report highlights questions still open at launch, including named customers, revenue, product pricing and a general-availability timetable. Its analysis stresses the difference between raising capital and demonstrating that improvement continues after an initial pilot. That is a useful test for research institutions considering an evaluation.

What research teams can assess next

CASRAI’s practical interpretation is that a university pilot should connect three records: the data a system may learn from, the changes a human approves and the evidence used to accept each change. A system that evolves during use also needs a record of its successive configurations for computational reproducibility.

Noeri’s launch establishes a financing event and a published research agenda. The next evidence that matters to research buyers is a documented deployment with reviewable performance and clear contractual rights. The funding announcement provides a reason to examine that evidence when it appears, rather than assume the founders’ economic and scientific claims already hold for a particular laboratory.

Sources

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →