Microsoft Discovery, a corporate Agentic platform, will accelerate R&D at Microsoft Build 2025.

What's Microsoft Discovery?

Microsoft Discovery is a new corporate agentic platform designed to accelerate R&D. The highly expandable architecture, revealed at Microsoft Build 2025, lets researchers add partner and open-source solutions, Microsoft discoveries, and their own models, tools, and datasets.

Microsoft Discovery aims to revolutionise discovery by giving scientists and engineers AI capabilities. From advanced knowledge reasoning and hypothesis formation to experimental simulation and iterative learning, this transition covers it all. With a graph-based knowledge engine and specialised AI agents, academics may collaborate on accurate, quick, and large-scale scientific outcomes.

Microsoft Discovery uses agentic R&D. This new paradigm aspires to transform R&D, not just speed up tests. It envisions a future where researchers collaborate with smart, cooperative AI agents to accelerate discovery. This requires incorporating AI into all scientific methods.

This platform addresses certain R&D issues:

Scientific knowledge is vast, complex, and scattered.

Connections between disciplines are difficult since the discovery process is dynamic, diversified, and requires various specialised techniques and jobs.

Science evolves through evidence, discussion, and improvement; research and development rarely yields simple solutions.

Scientific AI agents in Microsoft Discovery must reason across a complex and contextual graph that connects all information sources to achieve this agentic goal.

Focus on various jobs and areas.

Learn from findings and adjust study method.

Key features of Microsoft Discovery include:

Graph-based scientific co-reasoning: Large Language Models (LLMs) can speed up information retrieval and hypothesis development, but they often lack the contextual understanding needed for deep reasoning over dispersed, complex, or contradicting scientific data. Microsoft Discovery uses a powerful graph-based knowledge engine to create intricate graphs of relationships between external scientific research and proprietary data, helping users understand competing theories, experimental findings, and underlying presumptions across disciplines. This transparent reasoning lets the expert evaluate, understand, or change each step with comprehensive source tracking and reasoning.

Specialised discovery agents: Instead of compartmentalised pipelines, the platform uses an iterative R&D cycle in which researchers lead and coordinate specialised AI agents that can learn and adapt. Natural language-defined agents capture domain knowledge and process logic. CUSTOM AI teams can be created by R&D teams using their methodology and expertise. This strategy is more flexible than hard-coding behaviours in digital simulation tools. ‘Molecular properties simulation specialist’ and ‘literature review specialist’ are examples of models or tools users can suggest for agents to utilise or develop. These agents boost creativity by cooperating.

As orchestrator, Microsoft Copilot is a scientific AI assistant that drives cooperation. Copilot coordinates specialised agents based on researcher cues. End-to-end workflows that incorporate cutting-edge AI and HPC simulations allow it to choose agents and know a customer's portfolio of tools, models, and knowledge bases.

Microsoft Discovery uses Azure's governance, compliance, and trust controls to be adaptable and enterprise-ready. It integrates partner and client solutions with Microsoft technologies to create an open ecosystem. Proprietary, open-source, and commercial R&D teams can expand the platform by contributing tools, models, and knowledge bases. Embodied AI and quantum computing make the platform future-proof.

Multiple Microsoft Discovery's practical impact highlights:

Microsoft researchers found a promising immersion cooling fluid prototype in 200 hours using the platform instead of months or years. This non-PFAS prototype addresses the global ban on “forever chemicals”. The initial attributes matched AI projections after synthesising the digital finding in less than four months.

A solid-state electrolyte contender using 70% less lithium was identified with the Pacific Northwest National Laboratory (PNNL) of the Department of Energy. PNNL is also using Microsoft Discovery to improve machine learning models that forecast and optimise challenging chemical separations, especially in nuclear science, to reduce radioactive exposure and increase yields and purity.

Unilever uses it for fast computer simulations to promote science.

Microsoft, customers, partners, other Microsoft businesses, and worldwide entrepreneurs are creating a platform ecosystem. Customers collaborate on manufacturing, medical, silicon design, energy, chemistry, and materials. The following clients are mentioned:

GSK: Seeking a collaboration to improve their generative platforms for testing and prediction to speed up drug development and transform medicinal chemistry.

Estée Lauder Companies: They want to accelerate product development with their 80-year-old R&D data.

These partners offer domain-specific services:

The NVIDIA ALCHEMI and NVIDIA BioNeMo NIM microservices will be integrated to accelerate life science and materials science advances by providing cutting-edge inference capabilities and AI model development. Their discoveries will enable massive scientific data processing.

Synopsys wants to combine its industry solutions to expedite semiconductor engineering, re-engineer chip design workflows, and enhance engineering efficiency and creativity.

PhysicsX will use its physics AI foundation models to automate, optimise, and conduct engineering and production in specialised sectors.

Accenture and Capgemini help expand custom platform deployments. They want to transform labs and boost R&D productivity with their industrial experience and AI skills.

Microsoft is introducing a graph-based medical research agent to improve information retrieval and synthesis from credible medical sources. The Azure AI Foundry healthcare agent orchestrator code sample includes this agent, which provides practical, evidence-based guidance for complex, interdisciplinary healthcare workflows, including cancer treatment.