Noeri, a new artificial intelligence startup founded by Stanford computer science professors Yejin Choi and Carlos Guestrin, has launched with $110 million in funding to build specialised AI systems that improve through real-world work. The funding was led by Menlo Ventures, with participation from Madrona Venture Group, Kindred Ventures, Wing Venture Capital and NEA. Strategic investors include NVIDIA, Cisco Investments and Snowflake. Instead of building another general-purpose AI model designed to answer almost every question, Noeri wants businesses to create AI systems that understand their own data, learn their working methods and become better at solving specific problems over time. The company calls this vision plural intelligence: a future powered by many specialised AIs rather than a small number of models controlling the entire market.

Why companies need more than general-purpose AI
Businesses are rapidly adopting AI, but many still struggle to turn powerful models into reliable results inside their operations. General-purpose models can write content, analyse documents and generate code, but they do not automatically understand a company’s internal processes, specialised knowledge or unique requirements. Organisations often face an expensive choice: pay for the most capable models or use cheaper alternatives that may not perform as well on complex tasks. Noeri believes this trade-off can be reduced by building AI around the actual work a company needs to complete. Rather than expecting one model to understand everything, its approach is to combine different models, tools and instructions into a system designed for a particular business problem.
AI that learns from the work it performs
Noeri’s central technology is designed to improve the entire system surrounding an AI task. It studies a company’s data and workflows, then identifies ways to improve instructions, tools, specialised capabilities and the models assigned to different steps. Where useful, it can also work on improving the underlying models themselves. For example, an enterprise might use AI to analyse complex technical documents, process customer requests or support internal research. Noeri aims to make the system better at those specific jobs by learning from its actual performance rather than relying only on the knowledge available when a model was originally trained. This is a significant shift from traditional AI deployment, where a company selects a model, connects it to its data and largely depends on the model’s existing capabilities.
Many AI models working together instead of one model doing everything
The idea behind Noeri’s plural intelligence approach is that different problems require different kinds of intelligence. A smaller, specialised model may be more effective and less expensive for a particular task than sending every request to a large frontier model. Noeri’s systems are designed to learn how to use other AI models as tools, create additional tools when needed and choose the right model for each stage of a workflow. The company says its technology can work with both closed models and open-weight models, giving businesses flexibility rather than locking them into one AI provider. This matters as companies increasingly want control over the cost, performance and ownership of the technology running their operations. The long-term ambition is to make specialised AI systems more capable without requiring every business to build a large foundation model from scratch.

Human approval remains part of the system
Allowing AI to improve itself raises an important question: how can businesses trust changes made by a system that is constantly learning? Noeri says every proposed change is explained and submitted to a human for approval, using a process similar to reviewing changes in software code. This means the system can suggest improvements to tools, instructions, workflows or model configurations without automatically changing everything behind the scenes. The company also says customers retain ownership of the data, workflows and models they develop, subject to the licences of third-party models they use. For enterprises, this approach could be important because AI systems often work with sensitive information, internal processes and business-critical decisions. Control, transparency and the ability to inspect changes will be essential if self-improving AI is to move into important business operations.
From enterprise automation to scientific discovery
Noeri’s ambitions extend beyond enterprise software. The company also wants to apply its technology to scientific research, where AI systems could help researchers develop hypotheses, run experiments and learn from the results. Potential areas include discovering new molecules and proteins, as well as solving complex engineering and technical problems. The founders’ research provides part of the foundation for this direction. Yejin Choi is known for her work in natural language processing, reasoning and AI alignment, while Carlos Guestrin has a long history in machine learning and tools that help developers build intelligent systems. Guestrin co-founded Turi, which Apple acquired in 2016, and co-created XGBoost, one of the widely used open-source machine learning technologies. Choi is a MacArthur Fellow and has held research and academic positions at the Allen Institute for AI, the University of Washington, NVIDIA and Stanford.
A strong investor group, but a difficult technical challenge
The $110 million launch funding gives Noeri substantial resources to develop its research and work with companies across different industries. Menlo Ventures led the investment, while Madrona Venture Group, Kindred Ventures, Wing Venture Capital and NEA also participated. NVIDIA, Cisco Investments and Snowflake joined as strategic investors, bringing connections to major parts of the AI and enterprise technology ecosystem. However, building AI that genuinely improves through experience is a difficult technical challenge. The system must learn from changing business conditions without introducing errors, weakening security or making unreliable decisions. It must also demonstrate that specialised models can deliver better results at lower cost than simply using the latest frontier models. Noeri will need to prove its approach across real enterprise deployments, not just research demonstrations.
The future of enterprise AI may belong to many specialised systems
Noeri’s launch reflects a broader change in how the AI industry is thinking about the next stage of development. The first major wave was driven by large general-purpose models that could perform a wide range of tasks. The next phase may involve systems that adapt to individual organisations, learn from their work and coordinate multiple models to solve problems more efficiently. If Noeri succeeds, companies could eventually own AI systems that become increasingly valuable as they accumulate specialised knowledge and experience. That would challenge the idea that businesses must depend entirely on a handful of external AI providers for their most important capabilities. But the company’s vision remains a substantial technical and commercial bet. The real measure of success will be whether its systems can deliver reliable improvements, maintain human control and create lasting value for customers. Noeri has the funding and research leadership to pursue that ambition; now it must demonstrate that plural intelligence can work at enterprise scale.