Reworking R&D with agentic AI: Introducing Microsoft Discovery


We’ve got architected Microsoft Discovery to be extremely extensible, enabling researchers to combine the most recent Microsoft improvements with their very own fashions, instruments, and datasets in addition to a variety of accomplice and open-source options.

We’re asserting a brand new enterprise agentic platform known as Microsoft Discovery to speed up analysis and growth (R&D) at Microsoft Construct 2025.

Our aim is to convey the facility of AI to scientists and engineers to rework the complete discovery course of—from superior data reasoning and speculation formulation to experimental simulation and iterative studying. Microsoft Discovery permits researchers to collaborate with a staff of specialised AI brokers mixed with a graph-based data engine, to drive scientific outcomes with velocity, scale, and accuracy.

We’ve got architected Microsoft Discovery to be extremely extensible, enabling researchers to combine the most recent Microsoft improvements with their very own fashions, instruments, and datasets in addition to a variety of accomplice and open-source options. Constructed on high of Microsoft Azure, belief, compliance, transparency, and governance are key design rules of this enterprise-ready platform to allow accountable innovation, preserving the researcher in management.

At Microsoft, our researchers have leveraged the superior AI fashions and high-performance computing (HPC) simulation instruments in Microsoft Discovery to find a novel coolant prototype with promising properties for immersion cooling in datacenters in about 200 hours—a course of that in any other case would have taken months, if not years. This speedy discovery lays the groundwork for future developments in safer and sustainable options throughout a number of industries and is an illustration of how Microsoft Discovery can doubtlessly rework R&D in any firm.

We’re working with a notable set of Microsoft prospects who’re involved in co-innovating in numerous industries together with chemistry and supplies, silicon design, vitality, manufacturing, and pharma. We’re additionally working with a broad accomplice base that’s constructing on high of the platform to drive this acceleration, and we couldn’t be extra excited. The probabilities are infinite as we understand the complete potential of AI in R&D and we’re simply getting began!

The agentic imaginative and prescient for science

At Microsoft, we need to amplify the ingenuity of scientists to usher in a brand new period of accelerating discovery and increase the horizons of analysis. Doing so requires empowering R&D groups with transformative applied sciences that may drive significant enterprise impression. Nevertheless, R&D has very particular challenges in comparison with different domains:

  • Scientific data is huge, nuanced, and distributed.
  • The invention course of is numerous and dynamic, involving a number of extremely specialised strategies and duties, making it very arduous to attach the dots throughout the completely different domains concerned.
  • R&D is iterative. There are not often easy, clear-cut solutions. As an alternative, scientific data evolves by way of proof, discourse, and refinement.

This complexity calls for a brand new paradigm—one which isn’t geared toward doing the identical experiments quicker, however somewhat basically altering the paradigm of how we method R&D.

Think about if each researcher may collaborate with a tireless staff of clever, synergistic AI brokers with the only goal of accelerated innovation. That is our imaginative and prescient for a brand new agentic R&D paradigm, embedding AI in each stage of the scientific methodology.

On this new world, folks and specialised AI brokers will cooperatively refine data and experimentation in actual time in a steady, iterative cycle of discovery—all whereas sustaining the management, transparency, and belief that enterprises and governmental establishments require. This requires a complete platform the place AI can seize each the scientific area and the cognitive processes concerned in managing scientific thought. To comprehend this imaginative and prescient, scientific AI brokers should be capable of:

  • Cause over a posh and contextual graph connecting all data sources.
  • Specialize throughout distinct domains and duties.
  • Study from outcomes and adapt whole analysis plans accordingly.

Introducing Microsoft Discovery

We’re taking a giant step towards realizing this imaginative and prescient with Microsoft Discovery, bringing agentic R&D to life by leveraging the most recent improvements from Microsoft and the broader scientific ecosystem.

Graph-based scientific co-reasoning ​

The appearance of huge language fashions (LLMs) hinted at this new period, providing capabilities to hurry up sure scientific duties, notably for info retrieval and speculation technology. Nevertheless, LLMs usually lack the contextual understanding required to deeply purpose over distributed, nuanced, and infrequently contradictory scientific information.

Microsoft Discovery is constructed on high of a strong graph-based data engine. As an alternative of merely retrieving info, this engine builds graphs of nuanced relationships between proprietary information in addition to exterior scientific analysis. This permits the platform to have a deep understanding of conflicting theories, numerous experimental outcomes, and even underlying assumptions throughout disciplines.

This contextual reasoning can also be clear. Relatively than outputting monolithic solutions, it retains the knowledgeable within the loop with detailed supply monitoring and reasoning, offering the extent of transparency in AI programs that builds belief, ensures accountability, and permits specialists to validate and perceive each step or make any changes as wanted.

Specialised discovery brokers for conducting analysis

As an alternative of siloed and static pipelines, Microsoft Discovery implements a steady and iterative R&D cycle the place researchers can information and orchestrate a staff of specialised AI brokers that be taught and adapt over time—not only for reasoning, however for conducting analysis itself. The definition of those specialised brokers captures each area data and course of logic, merely by way of pure language.

R&D groups will be capable of construct a customized AI staff aligned to their particular processes and data, simply encoding these brokers with their experience and methodologies to make sure they’ll adapt and orchestrate as analysis progresses. This method is much extra versatile than hard-coding behaviors of at present’s digital simulation instruments, which regularly are extremely specialised and lack streamlined integration with others, and it implies that analysis groups not require computational experience to drive impression. For instance, customers can entry and outline numerous brokers’ specialties, akin to ‘molecular properties simulation specialist’ or ‘literature evaluate specialist.’ They’ll even counsel which instruments or fashions the brokers ought to use or create, and the way they need to collaborate with others.

This natural, bidirectional collaboration is a game-changer for managing R&D: brokers aren’t solely able to working for the researchers, however with them in a way that may really amplify human ingenuity—seeing each the forest and the bushes directly.

On the middle of this collaboration is Microsoft Copilot, appearing as a scientific AI assistant that orchestrates these specialised brokers primarily based on the researcher’s prompts. Copilot is conscious of all of the instruments, fashions, and data bases in a buyer’s catalog on the platform, can determine which brokers to leverage, and might arrange end-to-end workflows that cowl the complete discovery course of by combining superior AI and HPC simulations by way of the joint work of those brokers. 

Extensible and enterprise-ready

Microsoft Discovery is constructed on high of Azure infrastructure and providers, leveraging by design the belief, compliance, and governance controls on the core of Microsoft’s safe cloud basis.

We consider within the energy of an open ecosystem that leverages the strengths of Microsoft’s newest developments together with different progressive options from prospects and companions. Microsoft Discovery permits R&D groups to increase the platform’s catalog by bringing their toolkit of option to cowl their particular analysis wants in a complete scientific bookshelf. This extensibility on the core of Microsoft Discovery simplifies the onboarding of their alternative of computational instruments, fashions, and data bases—whether or not they’re customized developments, open-source, or business options. As we convey to market new capabilities in dependable quantum computing and embodied AI, the platform will stay future-proofed with one of the best applied sciences obtainable at Microsoft and throughout the trade.

Actual impression: Discovering a novel, non-PFAS coolant prototype

Over the previous months, now we have made vital strides aiding computational scientists of their analysis and incorporating cutting-edge improvements from Microsoft Analysis. This has led to outstanding breakthroughs, akin to discovering a novel solid-state electrolyte candidate that makes use of 70% much less lithium in collaboration with the Division of Power’s Pacific Northwest Nationwide Laboratory (PNNL) and enabling speedy computational simulations that speed up scientific discoveries at Unilever. Microsoft Discovery is designed to convey these improvements to each scientist—not solely these with deep computational experience.

One of many extra thrilling early use instances of Microsoft Discovery is unfolding on the Pacific Northwest Nationwide Laboratory, the place scientists are utilizing Microsoft Discovery’s superior generative AI and HPC capabilities to additional develop machine studying fashions that predict and optimize advanced chemical separations—a crucial course of in nuclear science. These separations are important for successfully isolating radioactive components after the nuclear fission course of, a notoriously time-sensitive and extremely chemically advanced job.  Sooner or later, the staff goals to make use of these developments to cut back the time scientists should spend in hazardous radioactive environments, whereas bettering yields and purity, enhancing each security and effectivity.

—Scott Godwin, Director, Heart for Cloud Computing, Pacific Northwest Nationwide Laboratory 

By leveraging superior AI fashions and HPC instruments for simulation that will likely be obtainable on Microsoft Discovery, Microsoft researchers found a novel, non-PFAS, immersion datacenter coolant prototype in about 200 hours.1 Present coolants usually take a few years to develop and might comprise dangerous PFAS-based chemical compounds that make them unviable to make use of, as there’s a world push to ban these “perpetually chemical compounds” in favor of extra environmentally pleasant choices on this trade and plenty of others.

After the digital discovery course of, we efficiently synthesized this coolant prototype in underneath 4 months, and it’s presently underneath additional evaluation and refinement. We’ve got already examined among the major properties of this materials they usually align to the AI predictions, which is a testomony to the accuracy of the predictive fashions used. Whereas this mission is just an experiment, it lays the groundwork for future developments and enhancements in coolant know-how and demonstrates how the mix of HPC and specialised AI fashions can speed up and rework R&D processes.

In accordance with Daniel Pope, founding father of Submer, an organization whose mission is to construct datacenters with a robust concentrate on sustainability, effectivity, and a wiser utilization of assets:

The velocity and depth of molecular screening achieved by Microsoft Discovery would’ve been inconceivable with conventional strategies. What as soon as took years of lab work and trial and error, Microsoft Discovery can accomplish in simply weeks, and with larger confidence.

A rising ecosystem

We’re placing this enterprise-grade platform into the arms of world innovators to display real-world impression throughout industries—from chemistry and pharma to manufacturing and silicon design.

It’s solely with a robust ecosystem that we’ll be capable of understand the complete potential of Microsoft Discovery, and it’s why we’re working with prospects, companions, and different Microsoft groups to convey first-party developments along with main trade instruments and area experience.

Prospects and inside collaborators

GSK is working to revolutionize healthcare, uniting science, know-how and expertise—together with world-class partnerships—to get forward of illness collectively. The corporate makes use of tech to advance science and speed up the event and supply of medicines and vaccines to positively impression the well being of individuals at scale. 

GSK’s depth and breadth of information and built-in use of tech throughout each a part of its enterprise—from early scientific exploration by way of to fabricate and supply of medicines and vaccines in market—present a novel providing when working with others. The corporate appears ahead to a doable partnership with Microsoft with the intent of additional advancing GSK’s generative platforms for parallel prediction and testing, creating new medicines with larger velocity and precision, and doubtlessly reworking medicinal chemistry to new unimaginable ranges. The probabilities forward are thrilling, and collectively, we are able to try for probably the most progressive options for sufferers and for well being. 

The Estée Lauder Corporations has gained a worldwide status for high-quality skincare, make-up, haircare and perfume merchandise that ship extremely efficient outcomes demonstrated by in depth analysis and product analysis. The corporate is happy to harness the facility of Microsoft Discovery to additional speed up the event of merchandise that uphold the best requirements of excellence.

Our proprietary R&D information, stemming from the minds of our good scientists and almost 80 years of analysis, growth, and experimentation, is a key aggressive benefit. The Microsoft Discovery platform will assist us to unleash the facility of our information to drive quick, agile, breakthrough innovation and high-quality, personalised merchandise that can delight our shoppers.

—Kosmas Kretsos, PhD, MBA, Vice President, R&D and Innovation Expertise, The Estée Lauder Corporations

Moreover, Microsoft is releasing a medical analysis agent that makes use of the identical graph-based data engine obtainable in Microsoft Discovery to boost info retrieval by synthesizing insights from trusted medical journals. As a part of a broader set of specialised brokers within the healthcare agent orchestrator code pattern in Azure AI Foundry, this agent permits researchers and builders to ship actionable and evidence-based steering tailor-made particularly to advanced, multi-disciplinary healthcare workflows—akin to most cancers care.

Area-specific choices

Combining Microsoft’s and NVIDIA’s strengths in generative Al and scientific computing, we plan to combine Microsoft Discovery with NVIDIA ALCHEMI and NVIDIA BioNeMo NIM microservices to speed up breakthroughs in supplies and life sciences. Supplies researchers will now have entry to state-of-the-art inference capabilities for candidate identification, property mapping, and artificial information technology. Biomolecular R&D groups will be capable of speed up Al mannequin growth for drug discovery, leveraging pre-trained BioNeMo Al workflows, all in Microsoft Discovery’s unified, enterprise-grade platform.

Researchers also can deploy their AI brokers on high-performance NVIDIA-accelerated Azure AI Foundry infrastructure, enabling them to effectively course of and synthesize giant volumes of scientific information with distinctive velocity and responsiveness for accelerated discovery and enhanced analysis outcomes.

AI is dramatically accelerating the tempo of scientific discovery. By integrating NVIDIA ALCHEMI and BioNeMo NIM microservices into Azure Discovery, we’re giving scientists the flexibility to maneuver from information to discovery with unprecedented velocity, scale, and effectivity.

—Dion Harris, Senior Director of Accelerated Information Heart Options, NVIDIA

Moreover, we plan to combine Synopsys’ trade options in Microsoft Discovery to speed up semiconductor engineering, serving to each {hardware} designers and software program builders ship superior merchandise.

Semiconductor engineering is among the many most advanced, consequential, and high-stakes scientific endeavors of our time, which makes it an especially compelling use case for synthetic intelligence. By integrating Synopsys’ pioneering AI-powered design options with Microsoft Discovery, we are able to understand the potential of agentic AI, re-engineer chip design workflows, supercharge engineering productiveness, and speed up the tempo of know-how innovation.

—Raja Tabet, Senior Vice President, Engineering Excellence Group, Synopsys

Microsoft can also be working with PhysicsX, planning to combine the corporate’s physics AI basis fashions into Microsoft Discovery so prospects can unlock new ranges of automation, optimization, and efficiency throughout engineering and manufacturing.

The Microsoft Discovery platform represents a seismic shift in how AI can speed up scientific discovery and engineering. That is about reworking how advanced bodily programs are designed, constructed, and operated throughout superior industries—in aerospace and protection, semiconductors, minerals and supplies, vitality, and automotive. Collectively, PhysicsX and Microsoft are constructing the software program infrastructure that can outline the following period of engineering.

—Jacomo Corbo, Chief Govt Officer and Cofounder, PhysicsX

Integration help

Lastly, we’re excited to accomplice with a rising record of software program integrators, akin to Accenture and Capgemini, to assist prospects and collaborators scale customized platform deployments.

Along with Microsoft, we’re shaping a daring AI imaginative and prescient for organizations who use deep science to convey progressive merchandise to sufferers and shoppers. Our laboratory transformation methods and Microsoft’s Microsoft Discovery platform create a dynamic ecosystem for scientific development. This collaboration will assist us understand the laboratory of the long run, enabling scientists to push the boundaries of discovery, experimentation, and testing with larger velocity and precision.

—Adam Borenstein, Managing Director, World Laboratory Reinvention Lead, Accenture

We’re excited to be bringing the Microsoft Discovery platform and AI brokers to R&D-intensive sectors. We consider these applied sciences have the potential to allow knowledgeable scientists to unlock step modifications within the tempo of innovation, bringing transformative advantages to enterprise and society. This partnership will drive productiveness in laboratory-driven R&D by drawing on Capgemini’s trade expertise, specialist bodily and organic AI capabilities, and science-led ‘lab-in-the-loop’ mental property, together with that of Cambridge Consultants, the deep tech powerhouse of Capgemini. For our shoppers this might imply accelerated discovery and predictive modelling or different aggressive benefits by way of utilizing information and AI at scale. 

—Roshan Gya, Chief Govt Officer, Capgemini Invent

Able to take the following steps?

Study extra about how Microsoft Discovery can assist scientists and engineers rework analysis and growth:


¹Based on the definitions of PFAS supplied by the Organisation for Financial Co-operation and Growth (OECD) (2021), the U.S. Environmental Safety Company and Buck et. al. (2011)



Related Articles

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Latest Articles