What the Microsoft Discovery platform is and why it matters
Microsoft Discovery is an Azure-based platform for building, running, and governing agentic AI agents that coordinate complex, evidence-driven R&D workflows across scientific and engineering domains. Rather than a single chatbot, it supports teams of specialized agents that reason over institutional knowledge, external literature, simulations, and experimental data. The heart of the platform is the Microsoft Discovery Engine, which supports the core loop of hypothesis generation, experiment design, execution, analysis, and iteration, while preserving traceability and review. Discovery is designed to fit into existing research environments, connecting to domain tools and respecting governance over proprietary data. According to Microsoft, the platform is now generally available to all organizations, with built-in controls for reproducibility, reviewability, and security in enterprise AI deployment. A new Microsoft Discovery app in preview brings a desktop experience that lets researchers and students begin working with AI-powered R&D workflows without standing up a full Azure environment.
Agentic AI agents behind Microsoft’s Majorana 2 quantum breakthrough
Microsoft is using its own Discovery platform to accelerate quantum hardware development, most notably in the new Majorana 2 topological quantum chip. Discovery’s agentic AI agents were deployed as coordinated teams that managed fabrication workflows, automated measurements, optimized the materials stack, and correlated nearly two decades of heterogeneous experimental data to uncover flaws in qubit manufacturing. According to Microsoft, “Majorana 2” achieved a roughly 1,000-fold improvement in reliability over its predecessor, and the company now expects to deliver a scalable quantum computer by 2029, cutting its original timeline in half. This is a concrete example of autonomous AI workflows augmenting human experts in a highly specialized domain. It also highlights why confidence scoring, cited references, and reviewable reasoning are central to Discovery’s design: quantum physicists need to see and challenge how agentic AI agents arrive at their recommendations before changing fabrication or design decisions.
BHP’s copper challenge: AI-powered R&D for the energy transition
In mining, the Microsoft Discovery platform is driving a different kind of breakthrough: faster copper innovation at BHP. Geochemists and data scientists there used Discovery with partners at Prescience Insilico to screen more than 500,000 potential chemical reagents for copper leaching, a process critical to extracting more of the metal needed for energy infrastructure and everyday materials. Discovery coordinates specialized AI agents to run tens of thousands of quantum chemistry simulations, rank promising molecules, and feed a much smaller shortlist to laboratory scientists for validation. This turns a slow, largely sequential trial-and-error process into a parallel, autonomous AI workflow that keeps domain experts in control of final decisions. BHP’s Vice President Innovation Jessica Farrell said the goal is to “focus on the most promising copper leaching solutions, sooner,” illustrating how enterprise AI deployment can cut wasted experimentation and concentrate human effort where it matters most for industrial-scale problems.
Target selection in pharma: Causaly, Discovery and inspectable AI
Drug discovery teams face an early, expensive decision: which biological targets are worth pursuing in the first place. Causaly, a life sciences AI company, is collaborating with Microsoft to tackle that decision using connected agentic AI workflows. In this setup, the Microsoft Discovery platform produces computational signals from large-scale modeling, simulation, and internal data, while Causaly interprets those signals against curated biomedical knowledge graphs and external literature. Causaly’s “Scientific Workflows” framework moves from a target and question through three evidence gates—mechanistic plausibility, biological rationale, and prioritization—to generate a ranked shortlist with provenance preserved at every step. Co-founder and CEO Yiannis Kiachopoulos emphasizes that “everything needs to be inspectable,” echoing Discovery’s emphasis on reviewable, traceable outputs. The combination aims to turn opaque model scores into transparent, biologically grounded insights that R&D leaders can defend when deciding which targets to advance into costly experimental programs.
From enterprise AI deployment to desktop labs: what comes next
With general availability, the Microsoft Discovery platform is moving beyond flagship projects into broader enterprise AI deployment. Its Discovery Engine orchestrates multi-agent workflows across Azure high-performance computing, institutional data, and specialized R&D tools, giving organizations a way to standardize autonomous AI workflows while keeping governance and human review in place. At the same time, Microsoft is previewing a free Microsoft Discovery app that runs locally and connects via a GitHub Copilot account, opening AI-powered R&D to smaller teams, students, and individual researchers. Together, these offerings suggest a future in which agentic AI agents support everything from semiconductor design and mining chemistry to target selection in drug discovery. As partners like Causaly and industrial users like BHP join the growing ecosystem, Discovery is positioning itself as a shared foundation for AI-powered R&D, where scientific reasoning remains visible and human judgment remains central.
