At its Build 2025 conference, Microsoft introduced a bold new platform called Microsoft Discovery, which it claims could transform how science is done. Designed to handle complex research tasks from start to finish, the platform uses agentic AI to support every step of the scientific process—from forming hypotheses to running simulations and analyzing results.
According to Microsoft, Discovery is an “enterprise agentic platform” built to accelerate breakthroughs by giving scientists access to a collaborative team of AI agents. These agents are specialized for different tasks and operate at the intersection of scientific reasoning, data modeling, and hypothesis testing—all running on Microsoft’s advanced AI and supercomputing infrastructure.
Can AI Really Speed Up Science?
Microsoft isn’t the only tech giant betting that AI can supercharge scientific progress. Google unveiled its own “AI co-scientist” earlier this year, while startups like FutureHouse, Lila Sciences, and drug discovery platforms like Exscientia and BenevolentAI have made similar promises. The idea is to use large-scale AI models not just for data crunching, but to assist with core research decisions.
However, the results have been mixed—and often underwhelming.
Google’s materials-discovery AI, GNoME, was credited in 2023 with identifying 40 new materials. But an independent review later found none of them were actually novel. In biotech, AI drug discovery firms have faced major clinical setbacks. High-profile platforms like Exscientia and BenevolentAI failed to translate AI-generated candidates into effective treatments during trials.
The limitations are clear. While AI can help scan massive datasets or propose options from within known parameters, it struggles with out-of-the-box reasoning and creative insight—skills at the heart of true scientific breakthroughs. Most researchers remain skeptical of AI’s usefulness in guiding real discovery, citing a lack of reliability, context awareness, and scientific rigor.
Microsoft’s Agentic AI Bet
Still, Microsoft is betting that agentic AI—AI structured as autonomous, goal-driven agents—can outperform earlier generations of tools. By making the Discovery platform extensible and collaborative, the company aims to build a system that not only processes information at scale but interacts meaningfully with researchers to suggest, test, and refine ideas.
Whether this effort will succeed where others have stumbled remains to be seen. But with the backing of Microsoft’s cloud ecosystem and a growing market interest in AI-powered R&D, Discovery could become a key part of the next wave of science infrastructure—if it can deliver on its ambitious vision.
