AI rewrites the rules of the lab!

An AI system found a new coolant in under two weeks. Four months later, scientists had it in hand!

Microsoft’s new agent-based R&D platform, Discovery, is designed to embed AI throughout the scientific process from hypothesis generation to molecular simulation and literature analysis. The system is already surfacing in labs and industries that rely on long, expensive discovery cycles.

By turning AI into a research collaborator rather than just a tool, Microsoft and other companies are betting that science itself can be sped up, structured differently and scaled across fields that range from materials science to pharmaceuticals.

“This is not just AI speeding things up. It’s AI participating in the very act of science,” Kunal Sawarkar, Distinguished Engineer for Generative AI at IBM, tells IBM Think in an interview.

AI joins the research team
Discovery's use of intelligent agents could open up research once confined to specialist teams and high-end labs, Sawarkar says. Each agent is trained for a distinct task, such as running simulations, reviewing literature or adjusting hypotheses, and can be directed through natural language prompts. Researchers can customize agent teams to reflect their own process.

The system runs on a graph-based knowledge engine that draws from internal datasets and external sources, mapping relationships between competing theories, experimental results and underlying assumptions. The result isn’t just an answer but a documented reasoning trail.

The platform’s orchestration relies on Microsoft Copilot, which allocates responsibilities among agents based on prompts and context. Agents collaborate, rerun tasks based on new findings and adjust research pathways without requiring the user to rewrite code.

Microsoft’s Discovery rollout lands amid a broader shift in how companies are using AI to overhaul scientific workflows. At Lila Sciences, a startup spun out of Flagship Pioneering, autonomous labs driven by AI and robotics are already running thousands of experiments at once. The company raised $200 million to scale “Science Factories”, facilities where AI proposes, executes and interprets experiments with minimal human input.

In one case, Lila’s platform discovered novel catalysts for green hydrogen production in just four months. By comparison, breakthroughs in similar traditional methods were expected to take up to a decade. The approach reflects what Lila CEO, Geoffrey von Maltzahn called  “scientific superintelligence”— AI systems that can operate across every stage of the scientific method.

Similar sentiments are echoed by IBM’s Payel Das, who tells the IBM Think newsletter that responsible oversight and human guidance remain essential while AI is accelerating progress in fields like drug discovery and materials science. “AI agents can generate ideas and analyze data,” she  says, “but human experts define the problem space and ensure that discoveries are safe, useful and ethically grounded.”

Science roving ground
Discovery’s public debut followed an internal experiment in which AI agents discovered a non-PFAS coolant compound suitable for immersion data center cooling. The system modeled molecular interactions and suggested candidates in under 200 hours. Microsoft researchers synthesized the most promising compound in less than four months. Initial tests showed a close match between predicted and observed properties.

Sawarkar says the idea is to “bring AI agents into the lab not just for analysis but for active collaboration in scientific workflows.”

Discovery is pitched as a flexible system that allows teams to integrate their own models, select preferred tools and remain in control of decision-making.

“What we’re seeing here,” Sawarkar says, “is a shift toward democratized supercomputing, where even non-coders can partner with intelligent agents to drive innovation.”

Microsoft is positioning Discovery as an enterprise-grade research engine with built-in governance and auditability. The system is built on Azure’s infrastructure and is designed to serve regulated sectors where transparency and traceability are essential.

“Discovery represents a comprehensive scientific co-pilot,” Sawarkar says. “It’s like giving every researcher a personal team of assistants who know the literature, can run simulations and explain their reasoning as they go.”

-- Sascha Brodsky, IBM.