Startups & Funding

Noeri Launches With $110M to Build Self-Improving Enterprise AI

Noeri launched November 16, 2026 with $110M led by Menlo Ventures. Stanford professors Yejin Choi and Carlos Guestrin build self-improving, customer-owned AI for enterprises and science.

Noeri Raises $110M to Build Self-Improving AI for Enterprises
Noeri Raises $110M to Build Self-Improving AI for EnterprisesPexels / pixabay
By Marcus Bennett5 min read

Why it matters

  • Noeri launched on November 16, 2026 with $110 million in funding led by Menlo Ventures.
  • Stanford professors Yejin Choi and Carlos Guestrin founded the Palo Alto- and Zurich-based company in 2026 and serve as co-CEOs.
  • Strategic investors include Cisco Investments, NVIDIA and Snowflake; Madrona, Kindred, Wing and NEA also participated.
  • Every proposed change in Noeri's system requires human approval through a code-review-like process.
  • Guestrin previously co-founded Turi, which Apple acquired in 2016; Choi is a MacArthur Fellow and former senior director at the Allen Institute for AI.

Noeri launched on November 16, 2026, with $110 million in funding to build self-improving AI systems for enterprises and scientific discovery, with Menlo Ventures leading the round. Madrona Venture Group, Kindred Ventures, Wing Venture Capital and NEA also participated, while Cisco Investments, NVIDIA and Snowflake joined as strategic investors.

The company was founded in 2026 by Stanford computer science professors Yejin Choi and Carlos Guestrin, who serve as co-CEOs. Noeri is based in Palo Alto and maintains an office in Zurich. It is currently hiring researchers and engineers in both locations.

The size of the round and the caliber of its backers signal how much investor interest remains in startups that challenge the dominant approach of scaling one general-purpose model. Noeri's founders argue the opposite: intelligence should be distributed across many specialized, customer-owned systems rather than concentrated in a single centralized superintelligence.

What Problem Is Noeri Attacking?

Enterprises today face a familiar trade-off, according to the company: pay more for the most capable general-purpose model, or accept weaker results from a cheaper option. Noeri's system is designed to dissolve that choice by adapting the AI stack to each organization's actual work.

The system studies an organization's data and workflows, then improves the surrounding software, skills, instructions, tools and choice of model for each step. When worthwhile, it also tunes the models' underlying parameters (the learned weights that determine a model's behavior).

Noeri's product overview describes agents (autonomous software programs that perform tasks) that learn from tasks, failures, corrections and real production traffic without requiring a large labeled dataset. These agents can:

  • Refine their skills, tools and orchestration — how they coordinate steps in a workflow
  • Create new tools when existing ones fall short
  • Train models directly on customer work

Quality and cost are measured on the customer's own traffic and metrics rather than public benchmarks, which makes any proposed gains specific to that deployment rather than to an abstract leaderboard.

How Does Human Oversight Work?

Every proposed change passes through a human gate before it ships. Noeri says each change is explained and sent to a person for approval through a process similar to code review, the standard software practice where engineers inspect and sign off on modifications before merging them.

The system works with both closed and open-weight models (models whose parameters are publicly available for download and modification), and it is designed to adapt as new models arrive. That design choice positions Noeri as a layer above the model market rather than a competitor to any single foundation-model vendor.

Ownership terms favor the customer. The customer keeps its data and owns the workflows and models developed through the system, although applicable third-party model licenses still apply. That structure addresses one of the chief concerns enterprises raise about AI adoption: losing control of proprietary knowledge and processes.

Why Do the Founders Reject Centralized Superintelligence?

Choi and Guestrin have staked out a clear position against the industry's prevailing direction. In a vision statement, the co-CEOs argue that intelligence should be distributed across many specialized, customer-owned systems instead of concentrated in one centralized superintelligence.

They ground that argument in economics and safety alike. Models sized to a specific task, they contend, are cheaper and less energy-intensive than large general-purpose systems. Purpose-built systems are also easier to specify, test and supervise — a meaningful claim at a moment when regulators and enterprises are pressing for explainable, auditable AI.

The founders' research record supports the thesis. Choi is a MacArthur Fellow whose research focuses on making smaller AI models more capable; she was previously a professor at the University of Washington and a senior director at the Allen Institute for AI. Guestrin previously co-founded Turi, the machine-learning platform Apple acquired in 2016, and later built and led Apple's central machine-learning and AI organization.

What Is the Scientific Discovery Angle?

Noeri intends to apply the same self-improvement machinery to scientific work involving molecules, proteins and complex technical systems. The system would form hypotheses, run experiments and learn from the results.

The company makes a pointed economic argument here: open-ended discovery strains the economics of large models because sustained search consumes far more compute than producing a single answer. Smaller specialized models are Noeri's proposed way to make that search affordable.

The research agenda covers four areas:

  • Automating AI development itself
  • Specializing models efficiently from limited data
  • Continual learning without forgetting (a known failure mode where a model overwrites old knowledge as it acquires new knowledge)
  • Discovery through search, reasoning and teams of models

Those priorities draw directly on the founders' prior work in test-time training, self-distillation and agentic orchestration — techniques where models learn after deployment, learn from their own mistakes, or coordinate other models and tools.

What Comes Next for the Company?

Noeri says its first proof-of-concept partnerships span both enterprise and scientific domains. The company is also working with strategic partners across industries, including some of its investors, and has deliberately chosen a varied group so the technology learns to generalize before specializing for individual customers.

That sequencing matters. Generalizing first, specializing second is the inverse of how most enterprise AI deployments proceed, and it determines whether Noeri's system can produce repeatable gains across sectors rather than bespoke one-offs.

The immediate test for the new lab, as the company itself frames it, is whether those partnerships can turn its research into repeatable improvements on customers' own measures of quality and cost — while preserving the human approval and ownership controls at the center of its product.

The stakes extend beyond one startup. If Noeri demonstrates that continuously specialized, customer-owned systems outperform rented general-purpose models on cost and quality, it will strengthen the case for a distributed architecture of AI — and pressure the foundation-model vendors whose pricing depends on enterprises renting ever-larger models.

Source: unite.ai

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

718 articles

Related articles

  1. Nous Research hits $1.5B valuation, launches enterprise AI agents
  2. Unsiloed AI Raises $7.5M Seed to Turn Enterprise Documents Into AI-Ready Data
  3. Norwest announces $4 billion fund with AI among priorities
  4. Nebius Stock Down 31% in Three Days on $27 Billion Spending Plan
  5. Pluto Security Exits Stealth With $42M to Watch What AI Agents Do

« Previous articleNext article »