DuPont Bets on Agentic AI From Microsoft and NobleAI to Speed R&D

DuPont announced on November 12 a research collaboration with Microsoft and NobleAI designed to put agentic AI in front of its scientists and engineers, with the stated aim of shortening the time it takes to deliver customer solutions.

The deal divides the work cleanly among three parties. DuPont brings its expertise in advanced solutions and applications. NobleAI contributes Science-Based AI modeling through its Visualizations, Insights and Predictions platform, known as VIP. Microsoft supplies Discovery, its enterprise agentic AI platform built for research and development.

What does the partnership actually build?

The proposed workflow separates research orchestration from scientific modeling. Aseem Datar, Microsoft's corporate vice president of product innovation for Microsoft Discovery & Quantum, described Discovery as an agentic environment that connects R&D teams' data, models and know-how for scientific discovery.

NobleAI CEO Alex Wang drew the line between the two software products. Discovery handles planning and orchestration of research; VIP handles predictive and optimization modeling.

Wang described a workflow in which DuPont teams would use their existing data, expertise and processes to explore more ideas in silico — that is, through computer simulation — before concentrating physical testing on the candidates most worth pursuing. In plain terms: models narrow the field first, and lab work follows on the shortlist rather than on the full set of options.

That framing matters because it positions AI at the candidate-selection stage, not as a replacement for physical testing. Simulation screens; the lab still confirms. Whether that sequencing produces results that hold up outside the initial laboratory remains the practical test.

Why limited data is the design constraint

NobleAI's modeling approach targets a common industrial problem: scarce data. The announcement describes Science-Based AI as integrating scientific principles throughout the modeling workflow, with VIP turning limited industrial data into usable predictive models.

NobleAI is also one of Microsoft Discovery's first launch partners. That relationship is what carries the modeling approach into agentic workflows for industrial R&D, rather than leaving it as a standalone analytics tool.

The takeaway: this is a bet that physics-grounded modeling can compensate where pure machine learning would starve for data.

How DuPont plans to scale beyond one lab

DuPont is tying the technical work to a company-wide AI rollout model. Chief Information Officer Matt Abbott said: "This collaboration reflects how we are scaling AI at DuPont, with security and governance designed in from the start, so what proves out in one lab can be put to work across the company."

That is a governance claim as much as a technology claim. Abbott's framing treats a successful laboratory result as a candidate for broader deployment, with security built into the design rather than added afterward. The announcement does not specify timelines, cost terms or financial targets for the collaboration — a gap worth noting given the specificity of the technical ambitions.

Chief Technology Officer Marty DeGroot said the combined tools are expected to accelerate development, shorten time to market and expand innovation's impact across DuPont's portfolio. The scope is broad: DuPont serves healthcare, water, construction and industrial markets, so a portfolio-wide capability would touch four distinct sectors.

Abbott's governance model, if it works as described, would give DuPont a route for converting a successful local workflow into a company-wide R&D capability. The unanswered question is speed — neither DuPont nor its partners put a number on how much faster development cycles might run.