CAS Brings Newton AI Agent Into Microsoft Copilot for Lab Workflows

Google DeepMind's Gemini 4 Argon scores 96.4% on GPQA Diamond, Epoch AI's leaderboard shows, yet laboratories still cannot hand their daily work to AI agents. CAS, the scientific information provider, argues the gap sits between model capability and trusted workflow — and on Oct. 20 it announced an integration with Microsoft Copilot designed to close part of it.

The connection gives Copilot users access to CAS's curated scientific content and to CAS Newton, an AI agent that draws on CAS data and tools to handle complex scientific questions, inside the Microsoft applications researchers already use. CAS supplies the information and capabilities through Model Context Protocol (MCP), a standard for linking AI systems to external tools and data. Copilot can retrieve information directly or route harder questions to Newton.

Andrea Jacobs, senior director of AI at CAS, frames the problem in plain terms. "We're really at the point where raw model capability isn't the number-one pain point in most AI adoption," said Jacobs, who trained in chemistry and computer science and has worked at CAS since 2010. "How do you build trusted workflows that are increasingly long-running?"

What does the Copilot integration actually deliver?

Copilot's advantage is placement. "One of the advantages of Copilot is that it lives on many users' desktops already," Jacobs said. "Many users in the world are already subscribers to the Microsoft ecosystem, and much of their work already happens in Microsoft products."

The integration splits labor between two resources. Direct retrieval answers questions that map to specific facts, such as a compound's properties. Newton handles questions that draw on several kinds of CAS content at once. "Newton knows CAS, our content, and has access to everything that we have," Jacobs said. "So it can take those complex and multifaceted questions, engage with them at a deeper level within our context, and bring that back to the user's agent context."

That context matters chemically. The pain reliever acetaminophen is also known as paracetamol; the asthma drug albuterol is also called salbutamol. CAS assigns each substance a unique CAS Registry Number regardless of its names, giving an agent one identifier to check against. In chemical notation, a single character can separate two molecules. "A one-character difference in science can be a dramatically different chemical compound or a dramatically different chemical concept, and that has meaningful and sometimes dramatic safety implications for a scientist, and accuracy implications for their work," Jacobs said.

Provenance is the second mechanism. CAS Connections links claims back to CAS source material whenever that content feeds an AI-generated answer. Jacobs said this helps roles beyond bench scientists — business development, project management — who handle scientific information without the judgment to verify it themselves. Validating an answer without grounding, she said, takes "quite a lot of work and intellectual energy and prior knowledge."

How far does CAS Connections reach?

CAS Connections, launched with several initial collaborators in June 2026, reaches both laboratory software and enterprise AI platforms. In electronic laboratory notebooks and laboratory information management systems, integrations handle specific tasks such as looking up a compound's CAS Registry Number or finding reference papers and reactions. For Copilot, Claude and ChatGPT Enterprise, CAS uses MCP. "That allows us to meet a much more diverse set of use cases through that generalized framework," Jacobs said.

What stands between AI agents and the bench?

Jacobs sees two remaining problems: trusted data and physical integration. "How do you solve the physical-world integration challenges?" she said. Anthropic previewed its Model Hardware Standard on Aug. 27, a specification that lets AI agents operate equipment such as microscopes, liquid handlers and robotic arms. "So, kind of like what we have with the Model Context Protocol, we need to adopt something like that for the physical world," Jacobs said.

Hardware limits persist regardless of protocol. Instruments remain in service for years; liquid dispensing systems may struggle with solids; automated chemistry equipment supports a limited range of reagents and reaction conditions. Even a capable AI plan depends on what the machines can physically do.

As agents take on more work, Jacobs expects the scientist's role to shift from planning experiments to directing software — a management task. "How do I direct the work of a team of AI agents that might be doing things on my behalf?" she said. Progress in computationally verifiable fields, where Gemini models now score above 93% on graduate-level science tests, gives her confidence that chemistry and biology applications are achievable; the open question, she said, is how quickly the human and physical-world challenges yield.