Unleash Your AI Superpowers: The Future of Development is Here!

>> Good morning. It is fantastic to be back in San Francisco build. It is always fun to be a developer conference to see great change. Developer conferences are always about understanding tech shifts and understanding the new stack. It is also about really coming to grips with the new opportunity for us as developers, for the companies we work at as well as the broader world. Today we are going to unpack this. This conference is all about that. If there is one key take away it would be this, how do you all participate fully in this frontier intelligence ecosystem. It is not about any one piece of technology you hear about or even the platform itself, it is about the value that you can build, you can compound, you can create on top of the platform. That's what the conference will be about and that's what we will

focus on. Let's take a look at this AI stack this conference is going to unpack in a great amount of detail. It starts with ubiquitous compute fabric that expands the edge and the cloud. You then have this layer which is the emerging layer where you have the models, you have context, you have the tools, the models can access, and then you have the run time where you deploy the agents and applications that you build on top of the content player and model layer. Of course you have the best tooling to do all of this and you have the security and compliance and governance. That's the simplest form. Let's start where it all starts which is infrastructure.

In fact let's start at the edge with windows. The amount of compute there is, at the edge is actually astounding. Think about every NPU, GPU, CPU even, every PC we ask ourselves one simple question. AI. In some sense we are delivering that already. When you look at something like outlook summarized it is using on board AI locally. Same thing with PowerPoint all text. Or teams super resolution. It is just not Microsoft. It is Adobe effects or premier. Using windows ML in order to happen to all of this compute power is to expand the scope of windows ML and windows AI. You now are full install base of GPUs you can get to. I am thrilled every developer out there can count on building for local on board AI and then have it run across all of the install base. Now if you are -- we are also announcing two cool models that are all going to run on windows in INVOX. Eye on instruct, it is a great reasoning model. We have the planning model eye -- Ion plan. You have a full local agentic loop you can give it tools access and build a fully agentic applications without having to run trip to the cloud. In order to really push the limits, realize this unmetered intelligence, there is a lot of hardware that we are adding too. We are thrilled to see the innovation through the ecosystem. The stuff from Intel is exciting to see. Qualcomm announced two sets of things which are great one on the high end snap dragon X on the lower end to sub 500PCs. Of course that brings us to inn NVIDIA. This is next generation SOC for PCs. It brings together the CPU the GPU as well as AI capabilities into a single SOC. It includes unified memory, architecture and also integrated

DRTM. You have the SOC coming together. One we build is the Surface Ultra it is a beautiful device. It brings the power of inn NVIDIA together with the design and Craftsman ship of Surface, 128 giga bites of unified memory. Beautiful 2,000 display, and it has an all day battery life. Excited to see this later in the fall. It is wonderful to see all of the designs from all of the OEM partners building and taking advantage of the SOC and the new platform and bringing out pretty exciting machines that are going to be all available come this fall. So, of course, we said, okay, this is fantastic. What can we do next? We said, let's try and push this. Push the architecture to its limit for developers. What if we could just Max the compute, Max the memory, build out that developer machine that is the dream machine, and that's what we are announcing today. Surface RTX, dev VOX. Let's roll the video. [Music] >> It is truly a dream machine. It has 1 Peta flop of AI compute. 20CPU cores. All of those things have 128 giga bites of unified memory access. So super excited. I am on the wait list as well, there's a wait list. We will get there.

[Laughter] So then we said why stop there? We said what if we did one more thing? In fact I think Jensen did that already. He said windows is coming to the DGX station. I describe it as the desktop data center which you can have running a 1 trillion model locally. It is close to what we perhaps had when we built GPT-5 or 3 one of the first super computers. It is crazy to think we have come this far with where you can have a data center on your desktop. We are extending the developer end points to the cloud. Windows 365 has the developer distribution that is optimized for developer productivity in the cloud that's windows 365 that I use every day. It is about making windows whether it is on the laptop, desktop, the cloud wherever the best place to build. And so to that end we have tons and tons of updates. We have starting -- we are starting with the favorite things for all of us, which is distraction 3 dev environment. We are introducing an intelligent terminal that has built in GitHub Copilot. Copilot intelligence. Of course lots and lots of Linux love for windows now. You have 70 plus utilities. Grab in full glory is now available for regular windows access. 70 plus utilities as I said. All of that is coming to windows. We are also bringing that you love. Home brew will be on native as well. Switching all of this. Having first class support for containers while will hp -- will really help us when developing locally. Showing you all of this on new surface on RTX. I wanted to invite up on stage Kayla to take a spin through all of the Dev tools. Kayla, go ahead.

[Applause] >> Our team has been working hard to make Windows a great place for development. Today I am going to show you major improvements and delighters you can try out today. What we have here is the default experience on the Surface RTX Spark DevBox. It is calm. No news feed no widgets popping up no notifications. We are in dark mode of course. I am ready to start development. There's one thing I would like to change. I like to put my task bar on the left. Let me jump into my task bar settings using run which is built with power tools xand pallet. No I want it all of the way on the left. There we go. After popular demand we are excited to announce vertical task bar is available in Windows insider build. The new surfacSurface RTX Spark bunch of tools already installed like python and node. All developer goodness all in one file. If you want the same experience on your device we are making this file available to everyone right now. We have a public repo set up with configuration file and how to apply it. It will make adjustments to windows and install all of the tools. One cool thing I have running here is power toys new utility called grab and move. Hold ality and move the window around from anywhere. You can enable end task which let's you end the process without having to open task manager. Let me jump into my Dev drive. It runs with defender running in async and authorize performance when comes to developer scenarios. Get aware. We have stuff like last name change message status of each file and the branch name is on the bottom left. So now let's get started building and open our terminal. It helps working with agents even more seamless. When you install intelligent terminal you can pick your favorite agent. I am using GitHub Copilot today. You can use which ever agent speaks to you. In intelligent terminal I have terminal pane at the top and agent working on the bottom. I, work between them while the agent helps along the way. Here is an air being generated. Agent pain is able to detect it and provide a fix which is great when I don't understand the sin tax. I am working on open claw I have worked it using WSL container. It is a native container experience on Windows plus it can leverage the GPU. It can reference your existing container files just like the one in open claw project. This is open in Microsoft edit got sin tax highlighting and latest version. You can see the container running with container command. Since we are on the topic of WSL there is a profile that is comfortable for those using tools like star, CSH and home brew. It has all of the favorite utilities and available in the repo I showed earlier and includes B top one of my favorites. The Surface RTX Spark serves large local models for coding. I have done development with 120 billion parameter model which most machines can't load. We can see how many tokens I have used locally. 3.4 million tokens leveraged on the device itself. We can kickoff agents using fleet. Just so you don't have to watch me type I am going to use Copilot voice feature which is leveraging its own local model. I will hold face for it and tell it what I want it to do. Find any console dot right line calls in the tray and node projects and convert them to the standard logger used elsewhere in the code base. There we go. The main agent will delegate sub agent tasks of appropriate complexity to the local model use liesing GPU and making it cost efficient. As developers while debugging we are looking through log files to diagnose any issues. Finding the location of the log files is a challenge. I would love to be able to type something like grep log and find them all. Sweet. On top of already adding curl, car and pseudo to windows now we are adding over 75 command line utilities like end, be head, tail and touch for those of us who love to live in the terminal. So I found all of my log files, but now having to pars them is the second challenge. I have had eye on instruct practically performing analyses on my log files this whole time. Now I can quickly diagnose anything that has gone wrong in my development, plus I don't have to worry about token usage since it is all local. We can take a look at the machinery sources. Models are loaded you can see 90 gigs of ram being utilized by the GPU using the full power of RTX Spark. We were able to use them simultaneously unmetered going through the dev flow without a hitch. That's huge. [Applause] So I know you are going to love what the team has been working on. We hope it gives you a glimpse of what's possible on windows today and where we are headed next. Thank you. Back to you Satya. >> That's the beginning of unmetered intelligence having the models and agents using the models work in parallel to what you are doing in the cloud. That's what the platform enables as first class.

Let's move to the cloud. The driving equation for us remains the same, which is tokens per dollar per watt. How do we optimize around this? When we think about the systems problem we think about electrons coming on one end and tokens on the other end and how do we think about the systems optimization end-to-end. It starts with the data center design itself. The core compute, storage, network. All of the accelerators that are going to accelerate each of the components. How do you think about the DC to DC connectivity and networking as well as the off load to something like the local compute. That's the sort of systems challenge. But before we even get into all of the systems and the technology and the innovation, perhaps the most important design criteria for us is how do we earn the permission from the communities in which we are building these data centers. That's where these principles ground us and focus us. How do we ensure that the DCs do not increase the electro city prices. Making sure we are using water use. Creating jobs for the local residents. Adding to the tax base. Making sure we are strengthening communities by investing in had local training and the nonprofits in the area. Only when we live up to the principles do the hard work is when we innovate and build. We have been doing a lot of data center build up. Today Azure spans more than 500 data centers in the region. We have the most expansive hyperscaler footprint out there. We have added more data center capacity in the last 18 months than the first decade of to put that in perspective. We built out the cloud infrastructure for hetero genius workloads that span the enterprise. When you look at what we are building when you think of all of the gigawatts that are going to come on-line we have three dominant workloads. There is training, inference and the agent run time. These are three dominant workloads. In fact when you look at fair water it was sort of our first AI super factory. It spanned two regions, Georgia and Wisconsin. The entire system was designed from the ground up for AI. We worked in fact very closely with inn NVIDIA on this. It is a two-story architecture that lets us he goes -- essentially put the number of GPU densely with network access. Means you really got fantastic higher performance networking, lower latency and more band width across the entire cluster. We are rethinking even the power delivery. Pou do -- how do we deliver 100 kilowatts her row with the conversion loss to the grid to silicon. We took a new approach to it. Also changes with the cooling system and water. The tooling loop is filled and the data center can operate effectively with zero water consumption. The daily water usage over the course of the entire one year is equivalent to what the single restaurant would use. [Applause] When it comes to the systems in the silicon, we have a lot of choice. We have first party silicon, we have partner systems. We have the first cloud to bring up inn NVIDIA. We work close with AMD. Working with them on the NX generation AMD GPU. Meyer 200 is continuing to scale it is live in Arizona will deploy internationally. It delivers 30% per tokens per dollar verses what's is the leading GPU today. We have validated it with GPD55 and we are going to use that to power Microsoft 365 Copilot. When it comes to running these agents, interesting thing now it is no longer about having AI accelerator or GPU. CPU is critical. Ratios may be coming to one is to one. That's why we are innovating with cobalt. We are announcing preview oOf Cob Cobalt00VMs. Next generation on generation CPU designed for cloud native and agent workloads today. It is exciting to see Cobalt make progress as well. [Applause] One thing we are trying to make sure being optimized for the new agent workloads. Cobalt delivers 30% better than Cobalt on cloud native. Using GitHub Copilot traces these agentic traces to see. The call patterns are so different. We are seeing 33% lower latency for the agent calls, 14% faster speed, 23% higher through put. This is about the co design of both the AI accelerator and the CPU for the agent. And of course, when you talk about AI workloads you need scale, you need reliability and that's why the network becomes super critical. We have innovated with the architecture and rebuilt how traffic across Azure moves to support workloads which is synchronous data workloads. They span 10s of thousands of GPUs. You need to have them coherent. That's what the next frontier is to make sure we are able to keep scaling the network architearchitectur architecture. It is not just about outside it is inside the data centers. It is connected through our continent spanning AI with the truly fungible compute fabric. All of this innovation is exciting, but when you think about innovation designing for AI and I would say deep understanding of systems, there's no better company than

NVIDIA and there's no other person than Jensen to talk about it. I want to invite Jensen Wang from NVIDIA. I know it is late for you in Taipei. I appreciate you staying up. I have been looking at social and people talking about your keynote. The concept of unmetered intelligence right at the edge is so hot again. You have thought about this, talked about this, now of course with RTX Spark really delivered what is a break through system for AI to be much Moreau -- more ubiquitous. Maybe you can share where you see this going? >> It all started about three years ago with a conversation between you and I. We were talking about how we could build a new class of PCs incredible for designers and creators and it would be incredible for artificial intelligence. It would be one of these systems that has the processing capability but also the software stack that is integrated into the world's design packages and creator packages and all of the things we are doing with AI. Here we are three years later we built an incredible new chip, and this system is supported by all of this new software you created for Windows. We have autonomous agent running on the PC. When you step back and think about what does that mean? For 40 years or some 30 years we have been working together we went from inventing direct X together to creating now this incredible computer that has autonomous systems running. The PC evolved from being an incredible tool to now being a tool that is used autonomously by an AI assistant. The idea I could be traveling and on the phone and text my PC and ask my PC to get coding done or an idea I have and it would fire up the tools on the PC and it would make the modifications or the changes or the design that I told it to do, and it would iterate with me while I am away from the PC. My PC became an assistant. While I am sitting there it would my great system as well. The idea that the PC evolved from a personal computer to personal AI is really exciting. To see it come to life and actually doing that, I am super excited about it. Spark you mentioned earlier has all of these capabilities Peta flop of AI performance. Has a Peta flop of MVFP4 numerical format our companies worked on together that allows it to take advantage of the 128 giga bites of memory and fit maybe a couple hundred billion parameter model. A couple hundred billion parameter model is state of the art. The days of having a really smart assistant running on the PC is here. >> It is awesome. I am excited about Windows coming to the GP300. It is a data center on your laptop. Talking about the data center side. This entire thing got started when we built the first super computer to train the GPU models. I was talking about the fair water design, this custom build for the grace black water era to Max the data center design with the system design you had. Now of course we are validating bel belly --. Maybe you can share on the clout side how you are sharing the system. >> Our journey is incredible. We built the first AI super computer together. Hopper was an incredible success. The first two generations were focused on free training. Grace Blackwell came along and all of the focus moved to post training, reinforcement learning which allowed us to have reasoning models. These reasoning models based on mixture of experts were incredibly intelligent, energy efficient. It requires giant systems. We required MV length 72. The entire rack became one computer. We had evolved from one node to now one rack. Well, Microsoft deployed the largest number of grace Blackwells in the world today. Largest number of grace Blackwells in the world. Fair water is a magnificent system to look at. It is an incredible feet. It is liquid cooled. Something you mentioned it is close looped, basically uses almost no water. It is incredibly environmentally friendly energy efficient. We are able to increase the token generation rate and reduce the cost of token generation by an order of magnitude, some 30 times over hopper. That was a huge achievement. Vera Ruben was created for a world where these AIs are now agentic. As hopper was created for pretraining grace Blackwell for training, post training and also inference. Vera Rubin is designed to run for agents. It is the same computing pattern when run on the RTX Spark. It is the same agentic system but of course larger win process enormous number of them. Many will be from different customers and different partners. The entire path, coding path from storage which is the long-term memory, working memory is encrypted. The data is encrypted in transit and data also incren crepted in encrypted in use. This entire disaggregated distributed computing system you mentioned CPU Vera is a CPU designed for agents. In the past CPUs were designed for humans. We are more parent than agents are. Agents want low latency. Vera was designed for extremely low latency. Vera Rubin I can't wait to show everybody. You stood it up. Our two teams have been working closely. Long before the chips taped out long before the systems brought up the two teams were almost completely aligned. Data centers created for Vera Rubin. It is into the networking and stack and community. The moment systems were rolling off the lines they were building stood up at Microsoft. So incredibly excited about the collaboration. >> This speed of light execution for the teams are fantastic to see. All of this is to power the ecosystem around us. You and I having grown-up with a PC the server and now with AI, have always thought about ultimately it is about creating the opportunity for every developer, every organization to build on the work that we do and the platforms we create. Speaking of that there's a lot of software that vid NVIDIA builds we will have the models in foundry, tooling in foundry. Software will help us with accelerating our workloads when it comes to the data warehouse. We will have the stuff in windows. Talk about the broader vision of what does it mean as an opportunity. Everybody talks about the one model and one piece of tech. It is about the broadest biggest opportunity maybe you want to share a little bit about that. >> We have been preparing for this moment. What happened the last several months we have been working a decade and a half together getting ready for really what happened in the last several months. All of a sudden because of agentic systems the convergence of these great models AI is now useful. If you look at GitHub, the commits into GitHub has gone completely parabolic. In the last several months the number of commits increase by a factor of three. It is clear agentic systems are useful, it is doing productive work and also it is profitable as a result. The amount of command for compute between the usage of AI the contribution necessary for agents it has gone through the roof. One thing we have been doing together is making sure all of the tools agents are using are fully accelerated. Fabric is fully accelerated. We are accelerating data processing, SQL, Spark, semantic based, graph based we are going to make sure all of the tools available on Azure are going to be fully GPU accelerated because the agents are going to be impat impatient. The faster it can iterate and generate tokens which are ultimately what the developers both customers would like to do is generate a lot of tokens that are profitable and highly intelligent. >> Thank you so much for the partnership and leadership and innovation that you bring to this entire ecosystem and really thrilled to be working closely with you and the team and bring all of this to the developers here and beyond and look forward to seeing what the next sort of few months and the next year will bring in terns of the innovation that gets built on top of the platform. Thank you again for joining this late in the night from Taipei. >> Thank you so much Satya for your partnership and friendship. [Applause] >> So so far we have talked about the edge and the cloud. The form factors, when I saw that Jensen picture from the weekend where he had all of the desktops. I felt man I am back in the 90s. It is so tool to see the lineup of all of the machines that I loved and grew up with back yet again with new functionality. The same form factor, but unbelievable new functionality, because of the on board AI capability. That is sort of what we have seen with the laptop, the desktop and of course with the

cloud. But it also, you know, sets up that next question. If you have that capability, which is new function and you can put it into existing form fact factors, can you even purpose build new form factors for the new function? Can you build a new platform even for the agent era? And that is the motivation behind project Solara. I wanted to invite Stevie on stage but first let's roll the video. >> First there's you. Then everything in front of you. We get it. The noise, the weight of it all. We have been building something. For an agent first world. New agent first devices. You come close, it is ready when you are. You speak, it understands you. What matters comes forward. Everything else falls away. When you want it to keep going it keeps going. Sometimes it sits with you. Sometimes it goes with you. Sometimes it sees what you want it to see. It doesn't decide for you. It lights the path. Tries to clear the way when you need it to. That's the idea. A whole ecosystem built to clear the way. Wherever you are, whenever it matters. >> I am so excited to be here. It really is great to be back on this stage. Before I talk about those awesome new devices you just saw, let me start with the why. Back at build 2023 I talked about the outside AI infrastructure. It moves within the application frame to operating globally to connect maintain context across entire workflows, devices and time scales. What if there were an ecosystem of devices specifically designed for the new type of application structure for those types of agents for that transaction. That's the impetus behind project Solara. With many possible forms which one do you pick? What is the next device. The big uh-huh for us is that it is not choosing one form factor it is about creating a system that extends your agent across a cons constellation of devices. The next computer is all of the devices working together as one system with agents cloeing up closer to where and when you need them. To realize this vision two challenges immediately show up. First many specialized form factors already exist but often rely on custom one off apps and fragmented stacks that are difficult and expensive to build, deploy and maintain. Second across every industry people and organizations are already building their own agents deeply specialized and instrumented for their work. The impact of those agents is constrained by how and where they can exist. Project Solara does both by extending agents on to new purpose built easy to manage form factors designed to reach the nooks and crannies where project creators do not exist or are not optimal. It is a turnkey solution for building unique agent first devices enabled by three pillars. First it is enterprise ready. Enabled by the AOSP based Microsoft device ecosystem platform. Second it has an agent driven interaction model with just in time UI that adapts to the form factor. Third it has extensibility so you can bring your own agents. Tying it all together is Azure unifying the system across cloud and device. That's enough of that. Now let's talk about the devices. Today we are previewing two very broad categories. The first is stationary and the second is port able. The first device is designed for your desk. It is built on media tech silicon. With hello for business walking up to the device securely signs you in giving you direct access to your agent. Just like Nathan is about to show you here. For the information work this means frictionless yet protected access to Microsoft 365 Copilot grounded in work IQ. What matters next in your workday helping you think, plan and even act by delegating task to your agent with a simple tap or just using your voice. Dedicated Ambian device for work. It acts as campaign onto your existing windows Copilot Plus -- PC. Or access windows 365 in a connected monitor. How cool is that? [Applause] The second is port able. Re-imagining a wear able millions of people use every day. The access badge. Built using Qualcomm silicon wear able this badge is light weight form factor designed for agents on the go. Adapt able across variety of verti verticals and workflows. I have an early prototype of the badge. Using my finger print I tapped to unlock the device now I have all of my agents in a secure manner. Gather content for social media post for today. Why not do it right now. I am going to hit record and the device's camera is recording. Hope you don't mind I am going to take shots. Find good shots from this clean them up and send them for me and my team to review. There you have it. Now my agent is off running through multiple tasks I clean them up and send them to me and the team. Pretty cool. I know it is a simple demo but it is all agent driven and there are so many verticals and opportunity. Imagine in healthcare from the moment you pick up the device the right agent shapes the experience around the roll and the workflow helping with check ins, patient records and critical insights. All through enterprise grade secure access. And with the built in microphones the nurse can start a hands free voice based documentation including dirizezation and and notation. The side phasing camera can help with asht vitals or scan medications or help verify workflows. This is a few examples of how the small purpose built wear able can bring intelligence directly into the flow of patient care helping nurses access and gather and even act on information while staying print with the patient. Both devices are an expression of devices the core hardware and software are designed to be highly flexible. With a few changes loading a different agent adjusting the shape, screen size, sensors or input methods the same foundation, the same software can be adapted for many verticals and workflows such as retail, industrial, hospitality, financial service, legal and so forth. That is the power of the platform, that flexibility. Whatever scenario there are places where computing has not naturally fit before. While this is an early look, we are really excited that ak ue whether, best buy, target and others are working towards exploring how devices can improve their workflows. This is the broader opportunity for the ecosystem. Agents moving outside of the app and taking shape in devices designed for specific scenario, a specific customer and a specific place. For all of you this is the moment to imagine where your agent should live, what form they should take and what new work they can unlock. Last week Satya sat down with Qualcomm, Chrissian yaw to talk about the future. >> Incredible to have you at device lab and talk about the reference designs. >> Happy to be here. I have been to the device lab a

lot recently. >> One of the things that you and I have chatted for a while is how there's a real platform shift. We are moving from building operating systems, devices for apps for agents. You want to talk about how you see this? >> This is technology this one seems to be a big one. Agents change the whole nature of the device in itself. Starting with the fact that if AI understands the world as we understand it it will be closer to our senses, closer to our eyes and mouth and ears. It will be things we wear. You need computing which is geared towards real time context. And from silicon to cloud. >> This is one of the references we have that I love. There is stuff happening 07b9 edge here. Things on the cloud. You want to talk about the core systems and silicon implications for such a new ecosystem. >> You need a power efficient CPU. The silicon is designed for you to have a cloud native experience. You have a lot of sensors for context. It is personaized. It is a thank change in the nature of devices. Wear able platform is changing itself. You see incredible factors. >> The computer platform, how do you build it such that there is an open ecosystem. It is not about one agent it is about any agent. >> Because the smartphone today is at the center of your digital life, the job of those devices today is to be extending the functionality of the smartphone. You saw many platforms became vertical. It is a natural thing to have a vertical platform from the same company. The phone was at the center. That changes dramatically when you think about agents. Agents becomes the center of your digital experience. They will be looking for an open horizontal platform that enable the agents to be interacting with the best possible device for different applications. >> We want to make it possible for anybody working with agentic system not just bound to devices but imagine they can be many devices that carry that intelligence in different context. That's the open ecosystem we want to build together. It is great to partner with you to get this started. >> We are proud of this partnership and that's just the beginning. >> Thank you very much Steven and Christiano. >> It is a new platform but it is a set of new platform rules that don't in some sense hem in what you can imagine the type of form factors where your agents live. Where the new platforms come you get to rewrite even the rules of how new platforms operate. That's what we are trying to get done with project Solara so you as developers and enterprises have the flexibility to imagine the form factors that you want and have agents be ubiquitous.

Let's go up the stack to the next layer. We are building a new intelligence layer bringing together the models, context as well as the tools. It starts with model choice. Every customer every developer is going to choose the right model for the right task and eval mix. Agency budget and foundry today has over 121 sthou models. The largest model catalog out there. From OpenAI to an thropic. Even MAI models. Last week we brought the OpenAI real time voice models along with Claude Opus 4.8 to foundry. We are continuing to bring these frontier models. The consideration about the models is becoming increasingly key. The last developer counsel to now when you are building the agentic system having this context really shaped right is becoming super important. In fact it starts right at the data tier. The data to date has been built for application that is supported user facing applications. Now you have to change build them for agents again different call patterns even to the data tier. Agents are restoring, retrieving, reasoning, acting and learning. That is what is happening in a continuous loop. They are using Cosmos DB, ChatGPT does, Azure search retrieval of ind -- indices. They started the business logic tier effectively for agents. You have fabric real time intelligence which now brings me to an exciting new service Verizon DB which is fully managed SQL server on Azure. One of the things we wanted to make sure we built ground up Proest gres managed service for higher availability scaleout. It is automated fill over 128 terra bites per cluster. The read heavy workloads. You can scale with this managed service. In our internal testing we are seeing Verizon is having 3X through put compared to PostgreSQL. It is critical thinking about the scale you need. The other data workload we are changing dramatically is the data warehouse in a world where agents are constantly querying data, data warehouse is critical. One thing to have it mission critical for users but when agents need analysis done on the fly bringing GPU acceleration to fabric is super key. We are seeing 7X performance gain. It is thrilling to see that AI acceleration. [Applause] So if you sort of have the data tier, the layer about this is the IQ layer that we are building that brings together essentially the model capabilities to -- along with the data. You kind of mix the data and model capability so you can deliver that right context to unlock intelligence. In fact when people talk about token efficiency this is perhaps the most important consideration. If you structure the context right and feed the models you are by definition going to be so much more token efficient. The first domain is the web. Web grounding is so important. You need that fresh, high quality and fast web data. And that's why we are really, really excited to announce today web IQ. [Applause] And web IQ is built on our global infrastructure that is already serving over a billion users. But fundamentally rearchitected for the LLM and agentic workflow. It is model agnostic, MCP native. Plugs into any agent run time. It has web and news, images, video, so agents can ground responses in flesh, verify able content, and web IQ leads across all of the three key criteria. It is best in class in quality, best in class in speed as well as in cost. So we are very, very thrilled about web IQ being in the developer's hands as you build out agentic systems. We are not stopping beyond the web. Every developer wants to ground their agents on what is the most valuable data across the enterprise. We are bringing together foundry, fabric and Microsoft 365 as this unified IQ layer. Continuously updated understanding of your organization to show you all of this rich IQ in action building real life agents here is Elijah. Elijah take it away. [Applause] >> Agents are only as good as the context we give them. Microsoft IQ unifies enterprise intelligence for every organization. I am here at a power utilities control center and will start by running a long running agent. This agent is going to help us assess the current grid operations incident and produce a brief for us so we can respond accordingly. Now I am going to go ahead and kick that off. While that runs let me show you how we got here. We built our agent And its also wired to a Foundry IQ knowledge base. A single grounded source that packages our documents, operational data, and people into context the agent can reason over. After building the agent in Foundry, we published it to Microsoft 365 for the whole team. Let's go see it in action. Now, here in M365, I will start with a question about current events that an ungrounded LLM wouldn't be able to answer. I'm going to ask about current electricity prices in SF. Our agents pulls in the first IQ tool, Web IQ, search built for the AI era. It delivers industry-leading quality, velocity, and efficiency. Web IQ constantly indexes fresh official sources and does a great job with semantic documents. And it give us our answer. Grounded in reality. Not only can we deploy our agent here, but we can embed it. Let's head back to the control center. Let's how we handleeds a previous incident. Can he asked for -- Deasked for details at our at risk substations. The power of Microsoft IQ comes when we combine that external knowledge with our own internal enterprise context. For this information our agent pulls in the next layer, Fabric IQ. Here is bright line's grid represented as a tab Rick on theology, an operational of the live grid, and we didn't build this from scratch. It takes models used by millions of customers and lets teams extend them into rich ontologies that helps run the business. This model -- yes! This model is coupled with live telemetry, so it reflects the real operational state of the grid minute by minute. This is what Microsoft IQ means by enterprise intelligence. [Applause] Not data scattered across disparate systems, but a single living model of the business that an agent can reason over. Now, turning back to our agent. We can see that it gave us a table of our most exposed substations. That's the agent query and the ontology you just saw. Now the agent has access to the outside world and our COMBRID. The last -- and our grid. The last piece turns the situation into a response. Our policies and our people. By asking, "What are the steps to respond to a substation trip," we activate the final layer, Work IQ. This is Brightline's response procedure in SharePoint. It's the play book the team reaches for when something goes wrong. And the important thing, the agent isn't working from a stale upload or copied snapshot. It's answering from the same source the team maintains day to day. When the procedure changes, the answer changes with it. No reuploads. No prompt rebuildings. And no stale versions. And critically, this is your knowledge. The assets you create stay with you, no matter what model or agent is reasoning over them. Now, if we return to our agent, we can see the proper way to respond to this issue. Not a generic recommendation, but our play book applied to this incident. Now, we just saw one situation, three questions, and one connected answer. Now let's go check back in on our long-running agent. And ... oh! Let's check the backup really quick. Moment of truth. And ... Boom! Our task finished! Ha-ha! [Applause] Here we can see every step the agent took. First, beginning with Web IQ, connecting it on the outside world. Second, fabric IQ through foundry, organicing it in the is -- anchoring it in the state of operations. And work IQ, grounding in people and procedures. I triggered this manually, but with Foundry routines, this can run on a schedule, turning a one off response into continuous, proactive execution. And if we look closely, we can see it even use the power of work IQ to alert me directly about the incident. And if we go ahead and check Teams, it sent me a brief notifying me of the situation. [Applause] That's the power of Microsoft IQ. When the crisis its, teams don't chance answer. They get an answer all in one place they can trust. Back to you, Satya. [Applause] >> All right. Thank you, Elijah. And now, you know, let's move up from this context model layer to deploying these agents and thinking about the run time.

And, you know, when you're building a first-class agentic system, you need a first-class agent run time and a platform that we are going to ship both with Windows as well as part of Foundry in Azure. We want Windows to be a fantastic place to run and scale agents, and agents effectively are a new execution environment. Right? It's a new paradigm, even. They reason continuously. They generate and run code dynamically. They take action across files and devices, as well as across the network. Obviously there's a lot of power in it. Right? The fact that it can generate code and act on it, on a long-running agent that's autonomous. But obviously it creates new risk. And that's why today we're introducing Microsoft execution containers, or MXC, which is a new policy layer that lets Windows apply isolation and containment, using AI native -- OS native primitives. You need to bake this into the operating system, so that the containment is enforced by policy. You can have process level isolation for lightweight agent actions. You can have session-level isolation for user separation. So Windows and Linux machines, including WSL, for much stronger boundaries. And in fact, if you want full isolation and containment, Windows 365 for agents, for maximum isolation in a separate managed environment, effectively. You can pick the right containment option for the work load, and windows will enforce it. And I think this becomes pretty critical as you think about deploying agents at scale on your Windows desktop. We want to ensure containment is enforced, of course, regardless of who builds the agent. This is why you want to bake it into the operating system. On to that end we're working with many partners to ensure that taint, we're supporting real developer workloads out there and addresses the needs. And NVIDIA is bring Open Shell so Windows, and today, we are really thrilled to announce that Open Claw redundancy on Windows leveraging MXC. [Applause] We are very deeply engaged with the team to make Open Claw run super well on Windows. Let me hand it over to my colleagues scat and Samantha toish -- Scott and smoothened to show you. Scott and Samantha? [Cheers and applause.] >> Hey, friends! Open Claw came out in November of last year, and it took the world by on the stomach. And for the -- by the storm. And I've been using it to stay on top of my health. It can help me manage my blood sugar and gives me notifications. I've got line triaging email, GitHub issues, and buys movie tickets. What are you using your for? >> I turned my Open Claw agent into my triathlon coach. It developed a work back plan for me, using my data to notify me how I'm progressing and keeping me accountable. >> We're working closely with Open Claw to make them successful and more successful on Windows. We've collaborated on GitHub to bring you an Open Claw app. It will help you set up your own on or connect to existing ones. And the Windows companion, we're going to sandbox the Open Claw tool calls to keep you and your system safe. >> Yeah! You'll see the Open Claw windows app rin running in the background. Right-click on it, Scott. >> All right. >> It looks awesome! It looks like a native windows app because it is. It's got information about I gateway, other machines, my sessions, and my usage. I've also got quick access to things like chat, Canvas, the main dashed, and more. Let's jump into companion settings. Within the app, we've got full chat support with tool calling, and you'll notice in the corner, we've got lots of PERMTHSZ options, along with our sandbox configuration. >> Thepermissions options, along with our sandbox configuration. >> The sandbox is interesting because this is using MXC, and for this, we're going to be using process isolation. Now, newer version of Windows will have more containment options, so keep your eyes out for news. I've got one-click security option settings, but Samantha, talk to me about custom folders. >> Yeah. You've got full support about what files and folders you want Open Claw to have access to, and really granular security features like clipboard access or talking to the internet itself. Now, I've given it read-only desktop -- read-access to your depth folder. Open Claw already has a rich safety layer, augmented more by me or policies by IT. For the purposes of this demo, I'm going to do something scary and ask Open Claw to delete all the fails on your depth -- files on your depth. >> I like a clean desktop. >> You hid all your icons from the audience while we were on stage. I think we need to show the audience who you really are >> That's so disrespectful. I know where everything is. [Laughter] It's just -- don't touch my stuff! I know where they are. Sam, I need to make sure that you're clear that a messy desktop is an organized mind. I'm pretty sure that that's the ... that's the quote, right? So what we've done, asked Open Claw to delete those files from the Windows node, and the only thing that is going to keep it from happening is MXC because we've turned off all of the layers that Open Claw offers, but our IT, in this case, Samantha, has stet to -- you can see the attempts, and checking the directory, and deleting again. It wants the files gone, and I want them to stay. [Laughter] Ope! No! The read-only sandbox is there. 94 je pegs are still there. They're safe! >> Foiled again! >> Oh, my goodness. So bad. >> We've seen security sandboxing, and brought together all in this alpha release of a Windows companion app. We think this app is a great opportunity to showcase Open Claw on windows, and it's only going to get better in the coming months >> That is right. And, by the way, I want to note that we're doing a all this development work on this calm Windows development machine with all the tools that I love, like WSL, containers, a containment, and I've even got GitHub Copilot with multimodel support ready to go. This is literally how the team and I have been working on the Open Claw app on GitHub. Now, people wonder how this came together. Turns out that over the holidays, I got a DM from this random guy on the intent, and it takes me a couple of days to get back to him. And then when I did, we just had this kind of cool idea. That maybe he should come to -- to Microsoft Build. I think there's one more person that we want to thank for bringing the next generation of agents to the world. Everybody put your hands together for the claw father himself, Peter Steinberger! [Cheers and applause.] >> Samantha, it's just showing off my secret DMs! I'm so excited to see Open Claw native on Windows. You know, watching a claw try to delete all your desktop files and just fail makes me really happy. [Laughter] Because six months ago, that totally would've worked. [Laughter] I built Open Claw to have access to everything. My files, my machines, my chats. Always on, and fully open source. That's what makes it so powerful! And that's what also makes companies a bit nervous. You know, what I kept hearing was, "Peter, I love my claw. Can I use this at work?" And that's what we spent the last few months on. With Microsoft, git Hick, open AI, NVIDIA, just to name a few. We had observability, auto mode for permissions. We changed how access works. It's not all or nothing anymore. You can pick which folder should read only, which ones should be hidden. Here's the nutrition: -- Here's the news: . You can run it inside your company now. [Cheering] And we even made the harness itself a plugin. You can bring your own. Copilot, codex, whatever you already trust in and your rules come with it. And you put Open Claw on top of it. Persistent memory, heartbeats, and a claw right inside Slack or Teams. It's been really exciting to see Open Claw grow into something much bigger. A global movement. And a community. And I started the Open Claw Foundation, a really nonprofit. Any model, any operating system. Because we are entering a new era of building agents, more capability for the people who don't code, and more power for those who do. And we get to do this, we get to build it together in the open. So my task is simple: Come build with us. Thank you. [Cheers and applause.] >> Fantastic. Thank you very much to Peter, to Samantha, to the companion app team for their hard work. Thank you to the Open Claw community for giving everyone here a crucifix crustacean -- a crustacean of their own. And that's the first time you've even Open Claw running on a Surface ultra. [Cheers and applause.] >> All right. Thank you very much, Samantha, Scott, and Peter. It's so wonderful to see Open Claw come to Windows. And have all of that capability in terms of the security and -- and that comfort, to be able to have these long running agents and unmetered intelligence come together. So now, let's move from the Windows side on the edge to the cloud with Foundry briar briar, and we're building Foundry into a full application platform for the agent era. Right? Every era of platform shift, when you move to the cloud, we have the cloud stack in Foundry. We're excited about the Foundry hosted agent as a runtime for long running agents. You know, agents now have -- if you're building in Foundry hosted agents, you can have all the IQ layers. You have the tools. You have the durability and the memory and the state. You have your own sandbox. In fact, it's a super-fast sandbox that you can spin up. You can generate the rubrics. You can -- and the evals. You can in fact have all the safety and the guardrails around your agentic system. And in fact, even one of the coolest things in Foundry is it's a continuously improving loop, even. It's got that self-improvement loop built in. You build an agent that's continuously getting better. And I'm really also excited today to announce a partnership with Fireworks AI, bringing all of their open wave models to Foundry. That means giving you as developmenters more choices, as well as that great inference stack, to build the next generation of these agentic indications, with all the enterprise Rayls that Foundry has, as well as the governance that Foundry has. Really excited about that partnership. [Applause]

And so now that brings us to the tools. Data itself is at the heart of all this. You know, in fact, Jensen spoke to this. GitHub is not just about the code repo. It's becoming the control plane for all the agents. And nearly everything we measure on GitHub, whether it's repo creation, PR activity, API usage, actions, all of this are going faster because of the agentic workflows. The new scale is driven by humans and agents collaborating together. And they're exposing too long -- tooling crass every form factor. We've seen tremendous growth in CLI, driven, you know, the approachability of the CLI form factor, it's allusion -- always been great to go to a termal. And now when combined with models and natural language, CLI is the thing that everything goes to. But end of the day, when you have hundreds of CLI's, it becomes pretty complicated. It doesn't scale. Especially the cognitive load that I have when you have 100CLI sessions open, is such that you kind of need something new. And so that's what has led us to build -- you know, we need like this tool, essentially, that has the speed and the flexibility of a CLI. But has the capability of an IE, and the ability to scale to infinite number of agent sessions. So today we're taking that next big step, introducing our new git Hick Copilot app -- Our new GitHub cCopilot app. [Applause] And we realize it's not sufficient. Because you still have a back toned deal with. Code is easy to generate, but what about the back end? You need to content with identities, storage, database schemas, and that's why we're really super excited about Rayfin, an agent-first SDK that connects your agents to the back end as a service, and we're bringing this to everywhere you build. That's why I'm, again, super excited about Rayfin and the relationship with Replit. You can now build apps in Replit, while the -- yeah. [Applause] You can build the app in Replit while the app and data are deployed into the enterprise-managed fabric tenant, thanks to the Rayfin SDK, which is available now for anyone else to be able to use with their tools as a back end, and essentially have the back end service. And now to show you all of the Foundry and Rayfin and building agents and the long running agents, let me invite Cassidy. Take with away. [Applause] >> Hello, everyone! I'm so excited to be amongst my fellow Devs. This is a long day. Fix your shoulders. Sit back. I see a lot of you sitting up. Great. I know you're drinking from the fire hose of information today, so I want you to go into the next few minutes thinking, "What can I drywall try out on -- what can I try out later today?" First, I can't wait to show you the new GitHub Copilot app. This is your home base for development and operations on your computer. And we think you're going to love it. Let me show you around. When you open up the app, you see the home screen where you can kick off a new session. But before night the serious stuff, you can drag Mona around, and there's a game! Look! It's so fun! [Laughter] I'm not very good at it, so let's just get back to, you can kick off a new agentic coding session. So I started off one a little bit earlier here. And it gave me a review of a bunch of release blockers. Which one should I fix? Call it out. 83? The critical ones? You know what? How about we just do all of them? Let's go! This app will now kick off a separate session for every single issue here. I don't have to worry about stashing or coding complex or anything, because the app takes care of that with Git work trees, that are isolated environments for each session. Your agents can work in parallel without stepping on each other. But you still have to merge them. Right? So -- [chuckling] -- Copilot has your back there too. If I head over -- not to this one. But to this issue here, I can run agent merge. And when I enable agent merge, cofight will baby sit this PR through CI check, code review, and merge collects. Okay. Let me keep showing you around while it's running. If I head over to my work, I can see a focused view of all of my activity ask just projects loaded in the app, PRs, everything here. And under automation, I have a bunch of reusable sessions that can run locally or on the cloud. You see there's issue poetry there. That's real. And that is load-bearing. Okay! [Chuckling]. And now undersessions, like I briefly showed earlier, these are sessions. If I want to add a new repository, I can click the button, and it will pull one, and if I were to add one, I can add a session, and this is an open source -- I can start a session anywhere and it loads it. When I look at a session within this repository, let me look at the other one, I get an integrated browser, there's a terminal. And a chat. I can technology light and -- toggle light and dark mode. And there's a button, pick and polish. I can polish anything in this app, and it adds it to the chat, and I can say, hey, I want you to add reordering to this list. And it will just work, all living in there. I have access to all the most popular models via my single GitHub Copilot subscription, including from open AI, Ann throw meic, and Google. And having a model change is agreed. Not only can you pick the right one, but Copilot can request a rubber duck review. In this session, I was use use GPT5.5, but it asked one from Claude opus 4.8. All models have blind spots, and the power of this approach means I can catch them earlier. This is all very cool. But working with AI in 2026 should be more than just chat. You just saw me scrolling up. There's so many words here. [Laughter] So today, I'm very excited to show you the concept of a canvass. I'm going to own one right there. The canvas is how an agent can build a custom UI to communicate with you. What if your AI could see? Everyone say, demo gods, bless us. Let's see if it works. Here's a fun canvas where, if I get the camera going, okay. The agent shows your PRs down here, and I can toggle it with a thumbs up or thumbs down. Let's aapprove it! Hey! It's so fun! And goes deeper. This is just the beginning of what you can do. So again, I kicked off a bunch of different sessions earlier. This is a signal box app. It's 100% agent-build. It's containerized with a database back end. Would you be able to deploy this to your enterprise with no questions asked? Be honest. No. Yes. Exact. No, but you can with Rayfin, and that is very, very, exciting. Let me open up a new terminal. All I have to do is type "Rayfin up." And then, demo gods bless you! Come on! It will ... [Singing] maybe dedeploy! Blammo! It happened! Yes! And hosted on Microsoft fabric. I know that's a lot. With Rayfin, your agents get a complete enterprise back end, so you can deploy with confidence in the way that's best for you. But this is the key thing to remember: This app is not just another session manager. Yes, it is. It managing a lot of sessions. But session managers just make it easy to create work, but this helps you to finish it. Thank you very much. Happy pride. Back to you, Satya. [Applause] >> All right. Now we're back to IDEs that have UI. That sort of -- it's so cool! [Laughter] It's come full circle. Now, let's talk about how you can observe, govern, and secure

these agents. Agent 365 is the agent control plane, agent requires their own identities, access controls. Even when they're working on your behalf, right? You just want that work on behalf identity to be enforced. So we extended an intraagent need real-time defense, so we extended Defender. Agents require this always-on data protection and compliance, so we extended Purview. And these agents can be hosted everywhere, AWS, GCP, not just on Azure. Or build with any framework. And today we are announcing a number of updates, including the GA of agent 265SDK, and we're x expanding it to your local agents running on Windows and elsewhere, and the claws, you just saw earlier. And let us take a look at how all the Agent 365 composes. Over to you, Amanda. [Applause] >> Everyone is building agents, but that's not the hard part anymore. The hard part is intrigue grating them into your business and governing them at scale. Today I will show you how Foundry makes this easy. Let's start locally. I've already built a line graph agent, and now I will show you the value Foundry adds. First, tools. Agents need tools to get work done, and with Foundry tool box, I just add my tools once, and any agent can consume them through a single MCP end point. And because tools essentially, that means governance does as well. I've applied a guard rile that blocks PII from leaking, and all I have to do is apply this once, and all my agents are protected. Now let's make this agent enterprise-ready. To deploy this agent to Foundry, all I have to do is add this one block of code, and then I push my changes and GitHub actions takes it from there. Once deployed, I can actually use this agent in the same Foundry extension I showed you earlier. And let's now test it out. I'm asking it to track a few open items, from a standup I had earlier today. And what's actually happening right now, is Foundry is spinning up a dedicated microVM just for this session. And the session even gets its own persistent file system. What you're going to see in a minute here, if I go to the Files tab to track the open items, the agent's actually writing to a file. Pretty cool, right? Off the top of my head has server side traces and evals to show me exactly what happened on every run. But how do I know if my agent's doing a good job? That's what Foundry's brand-new rubric evaluators are for. One easy command, Foundry reads my agent and generates the evaluation criteria for me. It creates a rubric personalized just to this agent. Now let's check out this rubric in Foundry portal. Look at these dimensions. Governance outcome correctness, prescribed source usage. I didn't write any of these. Foundry generated them from production traces. And with this rubric, I can score my agent, and run evals, but we can do so much more than that. That's where Foundry's brand-new agent optimizer takes over. Here's how it works. It tunes four things: The model, instructions, tool descriptions, and scales. And then it generates improved candidates and scores each one using the rubric I just showed you. Here, I can view the candidates. I can see the strategy used. I can see their scores. And I can actually view exactly what changed. This candidate, for example, improved its score by updating the model and the system prompt. Foundry then makes it super easy to deploy the best candidate as a brand-new agent version. But this is not a one time thing. Every run feeds the next eval. And every eval tells the optimizer where to improve next. So your agents now get better the more they're used. But now, let's put this agent to work. I've published my agent to Teams and M365 Copilot. Right now I'm going to ask it to just catch me up on what I've missed, because I've been of offline all morning, but my agent hasn't. While it works on that, though, let me explain what makes this agent different. This is an autopilot agent, which means it has its own identity and productivity license. So it can work across M365 on its own behalf. Earlier, you saw me using this agent by myself. But shipping of features and new release takes a whole team. And that why this agent lives in our Teams group chat. Let's check out how it did. Serums as you didded -- summariesed all updates and call out features I need to track. But putting this to work in your enterprise, governance needs to come first. That's why every autopilot agent requires admin approval to she can review it and choose who has the ability to talk to it. But it doesn'tent there. An over approval, add minutes anterior and block -- monitor it and block it at any time. But governance can't apply to just one agent. Every agent in your organization needs to be managed with the same rigor as users, apps, and devices. And that's exactly what Microsoft Agent 365 provides. So let's zoom out. Today we saw Foundry accelerate development, take your agent from local to enterprise-ready, and put it to work in M365. Foundry makes it simple. You build the agent; we handle the rest. Now back to you, Satya! [Applause] >> Thank you, Amanda. You know, what you just saw was how we're building security for

AI. But there's also one other critical aspect, especially, you know, the news today is all about, you know, how do you defend yourself using AI against attacks that may in fact be using AI? Right? So last month, we announced our multimodel agentic security system in dash, an agent harness for security, essentially, that we build. We're bringing together a hundred agents across the frontier, and custom models, to really find these exploitable bugs better than any single model does. In fact, when we Dubdebuted thi harness, it was on top of cyber gym [dog barking] benchmark -- Jim benchmark. So look what you can do to defend against AI attacks. Over to you, Sarah. [Applause] >> Thank you. We know that security scans can take a while, so I'm not going to do one life. Let me show you the results of an ender scan I already ran on my code base. The system runs as a stand-alone CLI, but today, I'm using it in my GitHub Copilot app on my local dev machine. The scan is broken down but vulnerability, demands, and severity. And it's also an additional to -- finding traditional issues like coding errors and hard-coated secrets. It's identifying AI-specific vulnerabilities in the code base. What happened under the hood, over 100 specialized agents working together to discover, debate, and improve exploitable vulnerabilities end to end. When the scan finishes, it generates both a log and an HTML report that I can give to my management. The defender details command allows me to dig into these vulnerabilities, and you can see what it is, where it is in my code, and the severity to help me prioritize. And from here, of course, we are going to fix it. Now, using the Defender fix command, the system will remediate suggest the fixing directly in my local dev environment. When that's done, I can check out the diff. So I have full transparency about what the harness has done, and I still have a human in the loop check. But of course, everything I've shown you so far has run locally. But I can also create a PR to plug into my existing workflows and push up to my repo, and I can take the output from the scan and upload it to tools, like GitHub advanced security, and manage everything alongside my other indication security findings. I've shown it working on my code, but let me show you a vulnerability our security research teams identified using MDASH that you can look up yourself. So the TL;DR, because this is a lot to read of this bug, is the time reads an out of date map of an object. Runs off the end. And then crashes the host. So this is exactly the kind of bug that the harness was build for. Because the floor spread across three different parts of the code base. So no single file looks wrong on its own. They look absolutely fine. And we can even see here be probably my favorite part, a statement from the developers claiming everything is fine. This is exactly the sort of reassurance that fools normal scanners and single AI models. But MDASH wasn't fooled. One team of agents spotted the suspicious gap. Another team argue they it in part. And a third team build a working example that actually triggered the crash. And it did all of this in open source code base. And this is a kind of joined-up reasoning that previously required significant manual security research effort. This is MDASH helping developers create secure code from the start, coming soon to your CLI and the Microsoft Defender portal. Back to you, Satya! [Applause] >> Thank you, Sarah. So that was the stack. Before, though, we move to, you know, unpacking more of the opportunity, I want to do

something different. I want to introduce two people whose LinkedIn profiles were both super impressive and slightly perplexing. Under role had says they're general partners on Mantis. On the previous role, sold out Madison Square Garden. Welcome Alex and Drew from the Chainsmokers. [Music] Thank you so much. One of the things when my team came to me and said, hey, we're going to have Chainsmokers at Build, I thought maybe that's what we're calling our new GitHub Copilot app. But maybe they're your fans. But tell me about how you got into this. I mean, you're being added for what -- at it for what, 12-plus years as investors first, and you had your firm now for seven-plus years, and you even picked what I would have not thought is the natural place, which is B2B SaaS. Give us the back story. >> I'm sure you guys are wondering what time line you're on, where we're at Microsoft Build, but hey, how are you. We've been in the Chainsmokers for 14 years, and you know, when our music started to take off in the mid two thousand teens, we got to play a few events like this. And we met a lot of the founders from the consumer mobile cloud era of startups. And they taught us a lot and give us a front row seat into what early stage investing was like. We fell love with it, had a lot in common with the founders and the way they build their business was the same way that we thought about starting the Chainsmokers and breaking through the noise there. And we got to participate in a few of their deals. And we decided to institutionalized in 2020 and start our own fund >> That's cool. When you hear about all the AI stuff and what's happening, there's a lot going on, even in your portfolio we were talking back staining. Things are changing. How do you see the opportunity going forward? >> I mean, there's so many vectors you could talk about. There's the creative output side -- >> Yep. >> -- we've been experimenting with in music. But always important to have your authenticity when it comes to creativity. But on the invest side, we're moving from producing outputs to producing actions. Which I think presents a very interesting opportunity to reimagine the entire architecture of the way software enterprise is build. Instead of humus producing outputs -- humans producing outputs, it's machines, and rethinking the entire space in that context. >> It's fantastic to see that. And maybe to close out, any sort of -- as artists who have had

great success, when you look at a founder, what sort of is the advice you give them when you're creating, you know, you're bringing something new, and it's a creative process. So what is -- what are you looking for in founders? What is it that you're giving them as advice? >> Totally. There really are so many parallels. Maybe it seems like that; maybe it doesn't. But it's really difficult as artists to really find, I guess, what your sound s what's authentically you. What you can continue to do over and over, because with this much competition, you have to be moving authentically. You have to be connected to the product that you're creating. And -- 'cause you're going to have to iteratedate and keep it going for a long time to lock into your fan base and make something special and nearby. And we try -- and unique. And we try to advise the same thing. >> Are you game for playing for us this evening for our closest friends here? >> We'll be there. >> 6:00! 6 this evening. Thank you very much. [Cheers and applause.] Thank you, thank you. We will be right here at 6 p.m. tonight, and really looking forward to it. So that effectively sort of talks a little bit about the developer stacks. So now let's talk about what is the opportunity for every company. Right? At the end of the day, we are an institution, an organizational builders, whether it's AI nave company, whether it's a SaaS company, or whether enterprise, the first thing we want to do is make sure that as you build things black -- Build things like plugins or agents or AI apping, we want them to be discovered through the Microsoft ecosystem. That's job one for us. That's why we're doing things in Windows or Microsoft 365 Copilot or in Teams or GitHub. We want to structure them that your applications, agents, plugins are discovered. Customers are also building lots of line of business applications and agents using Copilot Studio, and we want to make that all discoverable again, as part of the Copilot experience. And Teams in some sense has become this destination for multiplayer, human to agent interaction. We want you to be able to find agents, interact with agents, right in Teams. And we are super-charging all of this. Copilot continues to evolve very quickly. It started first with chat, with some of the best models, with great access to Work IQ. We didn't have the name, but that's where it started. Then came Cowork, a new way of working, and generating the stunning artifacts, and solve these multistep problems. Right? So you assign a multistep task to Cowork. You saw GitHub, which has continued to evolve as well. And now, as of, you know, come summer, you will be bringing coding to all knowledge work within one Copilot super app. Right? That's going to be really exciting, to see -- yep! [Applause] You will have chat, cowork, and code all in Copilot. But today, we're introducing something completely new.

Autopilots. When you think of autopilots as enterprise-grade claws. These are autonomous, long-running agents with full enterprise compliance that run in your tenant. Autopilots can have a name, personality, custom connecters, context, and memory. And they're totally new way to reduce toil and get you back to what you love, to kick things off, the the first autopilot we're introducing is Scout. Let's take a look. [Music] >> All right. [Applause] As you can see, Scout works where you work, joining group chats and teams, handling threats in Outlook. Starting today for those of you on Copilot Frontier, you can try out Scout. And in the coming months, we will build this out to a complete digital team of autopilots, right inside of Copilot. Right? So you can go to the Copilot app. Scout is the one that comes by default. But you can build more of these autopilots. And so that's the future of what we think of as the Copilot ecosystem itself. And so far, we have talked a lot about what you can do to build your own agents and how these agents are discovered and things like Copilot and Teams and Windows. But when you think about what makes any organization, any enterprise, any company unique,

it is its Tatacit knowledge tha is continuously compounding through its operations. So the key consideration at some level in an AI age is to ask the question, "What's the future of the firm?" How do you continue to preserve and compound that tacit knowledge in the age of AI, where models can learn anything? From the data and the trajectories they see. To do that, we believe that every organization, whether it's an AI native company, a SaaS company, or an enterprise, will need to build their own hill-climbing machine. It's a system that continuously improves against your objectives, your private evals, compounding your advantage over time. Not someone else's. To deliver on this, we're taking the next step with Frontier tuning. We've been working with many customers already to help, you know, really build their own learning loop. The environment, the context, the tools, the ribosomes -- the rubrics, and even train their own models. And today, we're really thrilled and excited about super charging that frontier tuning capability with the innovation that's coming out of our super intelligence lab, to tell you all about this, let me hand it over to Mustafa. [Applause] >> Thank you.

[Music]. Thank you, thank you. Thank you, good morning, everybody. You know, we really are living in the most remarkable times. Since I started working in AI, the computer that we used to train frontier models has increased by one trillion-fold. That is 12 orders of magnitude of computation in just 15 years. It's clear that a consistent, exponential increase in computation leads to predictable advances in AI capabilities. And in the next few years, we're going to see three more orders of magnitude of compute apply to train Frontier models. Long linear hill climbing is the norm. The scaling laws are clearly holding, and it is a remarkable time in our industry. And so in this context, we at MAI are building towards what we call humanist superintelligence. State of the art AI capabilities that are explicitly designed to serve people and organizations and not replace them. Because the type of AI that we create really does matter. We need an AI that places humanity first. But always prioritizes human well-being and human progress. This is the core philosophy and modification behind our efforts at Microsoft, and it shapes everything that we do. And as a platform company, our job and our commitment is to keep you developers building the absolute frontier. So today, we're very excited to announce a family of seven new models across image, voice, transcription, and coding. [Cheers and applause.] These are all built with real attention to detail and a commitment to making very practical and efficient tools that are typed to -- tuned to just how you work in the real world. First up: MAI image 2.5 and its flash variant, two super strong models that deliver a step change in quality, not -- now at #2, surpassing the core of Nanabanana2. [Applause] They gave you precise editing with control and consistency, flashes here for super efficient production work loads, while 2.5 gives you that maximum fidelity and professional grade performance. Alive in PowerPoint today, they're rolling out to One Drive, and you can access them on Foundry at a market leading quality per dollar. Next, MAI transcribe 1.5. This is the best transcription model in the world. State of the art accuracy across 43 languages, beating out Gemini and open AI's flagship models. We've optimized it for real world use so you can produce highly accurate triceps, five times -- transcripts, five times faster than all rival models. It's integrated inside of GitHub, Copilot, dynamic 365 contact center, and available in Foundry where I'm excited to say it is the fastest, most official, and most -- efficient, and most cost-effective transcription model of any of the hyperscalers out there. [Applause] So paired with that, we've got MAI Voice 2. The latest speech generation. Beautiful prosody, natural delivery, fine grained control, in 15 languages with more to come. And voice 2 flash, providing the best value and speed for ultra lateliency voice, which is the giggest thing in 2026. Next up, text foundation model. MAI thinking one. This is the first reasoning model, and it's exceptionally strong, and our target use cases of reasoning and sweet tasks. It's a 35 billion active parameter E, meaning it competes in the medium sized weight class. And independent human raters on surge prefer it in overall quality side by side versus sonnet 4.6. [Cheering] It's achieved 97% on AMI2025, the key measure of its capabilities, but most importantly, it's now at 53% on bench pro that places it alongside opus 4-6, at least on the toughest coding benchmark out there. We're very happy with that. [Applause] There's plenty more for us as we get this into production and hill climb against real-world tasks and traffic, but what's most remarkable about this model, we think, it has climbed entirely from the bottom. And that means that it hasn't targeted any of the benchmarks specifically, with zero distillation. And to us, this is critical. It means that the model is created with an enterprise-grade clean and commercially licensed data lineage that means that you can put it into production in a very trustworthy way with complete confidence. Now, finally, I'm incredibly excited to announce MAI code 1 flash. This is our new inference sufficient coding model tuned for VS code and GitHub CLI. 5 51%, despite five billion parameters. Close to haiku in terms of size but cheapner cost, delivering strong coding performance at great sufficiencily. And it's rolling out inside of VS Code. [Applause] Now, alongside distribution on Foundry and optimization for our 1P products, we're excited to make our models available on open router as well as Fireworks and base ten. This means for the first time you will be able to tune the weights directly yourself in an ecosystem of your choice. Now, across this entire family, safety and security has been build in from the start. Voice models come with protectiontion against unauthorized cloning. Everything is water marked from scratch. Reduced the reFOOUFLS, improvedish -- refuseals, and accomplishing a very detailed technical report today to give you a full and transparent understanding of how we put all of this together. One of the things I'm particularly excited about is that we have been carefully codesigning our models with our own silicon. That means that we've optimized MAI thinking one on our very own myo200 Chip and benchmarked against the GB2300. So top of the improvement, we're seeing a further 1.4X performance per Watt gain when we run them on the MIO model end to end, and that's huge. As everybody knows, this scale, every Watt counts. And silicon and model codesign is a really key advantage that we think is going to help keep everybody here right on the Frontier with the most efficient and powerful thinking and coding agents out there. And we're excited that these faster and more efficient MAI models are coming to the N1X that Satya mentioned a few moments ago, and we think that's going to be able to deliver the very best performance on mounds in a new -- on windows in a few months time. This is what owning the full end to end stack looks like. It lets you customize the MAI models using our full hill climbing machine right where you want it. And it means that the discipline and very relentless engineering that has gone into building our models is now available to all of you. On a platform that you can trust, working on your behalf, to create custom agents that you will control. So the really big thing, of course, that's happened in the last year is these RLEs. Reinforcement learning environments. These unique training gyms for your AIs. They create company and task-specific agents, adapted only to you, built on MAI models. For example, within Microsoft we use our RLEs combined with MAI models to climb towards the best agentic ace uses on Excel. The model is on par with GPT5.4 on public and private benchmarks, and being ten times more efficient on cost. [Cheering] Thank -- [chuckling] And many other earlier adopters are seeing similar results.

When we've tuned our models on McKensie's tasks, it delivered the highest win rate, outperforming GPT5.5, and delivering 10X grater efficiency on cost. To us, this is the advantage of very carefully calibrated frontier tuning. And importantly, unlike with some of the other companies, with shared model only you keep benefits from the hard earned workflows, know how, knowledge and your own institutional data. Only you get to control the resulting model. So with us, the RLEs and the models that you build inside of them they become your mote. It marks a new era in AI that we are all really very excited about. Okay. So now just one final announcement that I am very excited about. We are taking customization and co creation of our models to the highest level possible on what I think offers the most important application of AI, healthcare. Today we are very proud to be announcing we are partnering with Mayo Clinic for a new frontier for health and deploy it in their hospitals and beyond. Help me in welcoming to the stage physician ground breaking researcher, president and CEO of the Mayo Clinic Dr. Gianrico Farrugia. [Applause] Thank you for being here. Everyone recognizes Mayo as the leading hospital in the world with incredible track record of research, innovation and clinical practice. Tell us more about what hope to get collaboration. >> Mayo Clinic is known for being able to live up to primary value the needs of the patient comes first. We deliver outstanding healthcare. We are ranked number one healthcare organization in the world, yet we know most people in the world will not have access to Mayo Clinic. So seven years ago we decided to create a platform, Mayo Clinic platform, moving all of healthcare from a pipeline to a platform. With our partners that platform is in four continents and reaches about 100 million people. It has created the largest, to our knowledge, longitudal healthcare center in the world, multimodal genomics. We have the opportunity to do what we do best together which is create a frontier model for healthcare. What it means if you are a patient, if you are somebody interested in healthcare, you can get clinical and logistical answers to your healthcare. If your healthcare provider, your physician it can give you insight, it can act as your real-time team member it can tell you what is likely to happen next, but it can also prevent harm and therefore increase patient safety and give valuable insights that make the team better at giving you what you need most which is better healthcare. >> One of the things we are most excited about is the models are already pretty incredible about textbook knowledge. Read all of the journals and papers. What they are looking is clinical practice and clinical expertise of your team and clinicians you developed over the last many decades. How do you think you might go about the clinical practice for the model and production? >> The exciting part here is we each do what we do best. We can tackle something that Hazel lewded healthcare for a long time trusted scalable solutions. To do that you need to have the right data, certainly need to have the right people, but you need to have a very patient focused lens. Between the two of us we have all of this together so we can build this frontier model and offer safe, secure, trustworthy and effective healthcare solutions for all. >> Number one objective is put the patient first, deliver high quality that we can as trusted as we can and share the world over. Excited to share this partnership. Can't wait to share more in the future. >> Us too. Thank you. [Applause] >> So today marks some very, very exciting steps that we are taking on our journey to create humanist super intelligence at Microsoft. We have an incredible roster of seven new world class models to keep everybody working at the absolute frontier. We are really looking forward to everybody being able to co create your own unique agents adapted to you at your control. I feel this is a new era in AI. An era of AI that you control on your terms. Let's build it together. Thank you very much everyone. Now over to Tanaya to show us how frontier learning works in practice. [Applause] >> Thanks Mustafa. With frontier tuning we are making it possible for you to create your own enterprise AI, with the models and the harness tuned on your data and your workflows. Let me show you how you can build your own health planning machine. MAI thinking one is available in private preview in the foundry model catalog. You can go ahead and deploy the mo model as is or if you want to go on your health journey click the fine tune button. The first thing I will set. Next I am going to add a greater and that's it. I go ahead andee ate a -- and create a job. It has through logs, scores them and we start the hill climb. If you want full control over the training loop let me give you sneak peak into low level training API. You can see I have the MEI sinking model. I can also configure my role out strategy as well as my hyper parameters to define exactly how I want my training algorithm to work. In fact I can also incorporate my own RLM by defining the tools this model interacts with. You just saw all of the code, but as M365 customer you never start from scratch. Let's hop on to Copilot. As a part of frontier tuning we are introducing a new way to build environments based on your data and your workflows. One of our customers land O lakes one of the largest businesses in the United States is using this to perfect that butter on your moerning -- morning toast. Let me walk you through their environment. The environment has soft skills, knowledge and tools. But in the back end we create an entire origin for your agents to continuously learn the way that you work. Let's look at the butter report generation here. These tasks are very complex. They require many manual steps and a high degree of precision. In these tasks almost 80% accuracy isn't good enough. To hill climb to higher accuracy, we are extending the industry definition of skills to include Rubriks on what good looks like. This is one task. How do I scale this to qualify all o all of the tasks in enterprise. You spend a lot of time in M365 in teams, outlook, word, excel and PowerPoint. We use those signals to suggest skills and rubrics that define the way you work. Next you can add your organizational knowledge for branding like from OneDrive and SharePoint. The environment comes built in with Microsoft tools and you can add custom tools to it. Because these tools tap into real workflows we simulate execution so the model can learn without actually impacting the live state of your business. Now my most favorite part, the science. By generalizing all of these learnings into the main model as well as in the embedding model, we are able to hill climb for tasks that require high accuracy. And not only this, we are able to hill climb to more than 90% accuracy for Land-O-Lakes class for using the model. We estimate the model to be 10X efficient than the baseline models. Let's run the task. This time using this environment as an entrancing harness. Usually this task takes a couple of minutes to run. In the meantime I will show you a carshed response. They have research with multiple models including a fine tune model. What you see here is a summary that doesn't feel generic, it feels undoubtedly Land-O-Lakes. That's not it. The task holds itself to high standards. It continuously retrospect and evaluate success. With frontier tuning your agent continuously improves with environment on your data encoding the way you work. That is what I call frontier tuning as smooth as butter. We can't wait to see the environment you build. Back to you Satya. [Applause] >> Thank you very much. Thank you for the partnership here. What you saw is a pretty significant shift. Times come from just consuming frontier model to fully participating at the frontier and frontier ecosystem. That is the transition. You can have private evals and outcomes private RLEs and traces, your enterprise knowledge create scaffolding from models to hill climb. That will allow you to differentiate on-line what you control. There is a new operating point at the frontier. Where you can use a very effic efficient reasoning model and coding model and achieve frontier because you have done the hard work to create the RLE the hill climbing machine that create the model your hill traces can make to the frontier. We think the combination of the two is a game-changer and how people think about what does it mean to operate in the frontier. What does the token look like how are you in control what is the future of the firm those are the questions and really the ecosystem that gets built around as opposed to a few models hungry for all data.

To close out I want to talk about the frontiers beyond to push the frontier what's the next best thing. Building the scientific discovery loop can have perhaps the biggest societal impact. Science today is a little too leni lenient. You form a hypothesis, you run an experiment you wait for lab results and then you begin again. What if the scientific method itself can become more continuous, more program able. That's what we are doing with Microsoft discovery. Discovery brings together the models the ACP compute, the knowledge graphs of all of the scientific knowledge automated lab and simulation into one agentic discovery loop. I am thrilled to GA and Microsoft Discovery, to show you this in action let me invite David up on the stage. [Applause] >> Hello. I work in the Microsoft Discovery and quantum team. Today to recycle this plastic like this bottle you have to sled it and melt it. You cannot use it to make this bottle again. It is down cycling. We are using Microsoft discovery to advance that research. Let me show it to you. This is the Microsoft discovery app. It is VS code. Agentic Discovery has many with software engineers. I want to do three things as a scientist. I want to write a scientific paper about this topic. Instead of melting the plastic I can use proteins so you can recycle it again and again. Second I want to perform the actual discovery to define this. Third I want to create a protocol to test the results in a real lab. If you are a developer the steps will be very familiar to you. Let's launch it. This will is discovery engine. These are specialized agents always running in the background following the scientific methods. You can see the agents you can add more. Microsoft discovery includes a community of agents, models and tools across many domains. You can have officers fair parties or create your own. This is still running. It will be running for a while. Even hours or days. It is running simulations dynamically. You can see the files created here. This one is the research paper that I was asking for. Discovery using no less graph internally. It is internal knowledge. This is a critical asset because it provide compete visibility. Scientists can be in full control. How do you come up with the plastic. You need a way to generate the proteins and identify the most promising ones. There is no out of the box agent for that. That's okay. With Microsoft discovery there is not an agent for the task. You can create one on the fly which is pretty cool. Here is all of the files that it created. This one is the YAMMER for the agent. Now we need to generate candidates for the proteins. That task requires a lot of compute. It is integrated with HPC. You can use it. You can see the process. You start with C protein and created variations by replacing small segments. You can see the segments there. You apply to see if it helps or not. This is done millions of times exploring a huge tree of proteins. The result is 80 proteins sent to the lab for testing. It is like the software. Creating a protein is a little more complicated. The bacteria create the protein for me. So discovery created another file. This one here. This contains the DNA sequences. I can send them to the lab if I have a lab I can integrate it directly with the agent. Go to proteins and submit to lab. This will use a custom agent that sends instructions for the lab appointment. This is very real. That's an maut ma automated lab I have it right here. This is application controlling the lab. It has a Copilot interface so the scientist can interact with the lab to the science experiment. In this case Discovery has submitted the job. You can see it right here. This will take some time. Let me open a previous run. There you go. These are all of the steps that the agents are doing. Most of them completely automated with human supervision. How cool is that? [Applause] This is bringing the together the physical and agents in unified Discovery loop. This is one example of what Microsoft Discovery can do. Industries are using it today to embrace a new era of scientific discovery. Thank you. [Applause] Back to you Satya. >> Thank you so much David. Talking about scientific

discovery, we are also continuing to make rapid progress on our long-term goal of building a scalable computer. We announced last year our first QPU. We created a new state of matter that was only theorized 100 years ago. Proved it out that it exists. Our vision was to take a radically different approach to addressing the fundamental barriers to building a scalable quantum machine which is all about reliability, speed as well as size. Since then, we have continued to make progress across the full quantum stack with both our academic and industry partners. In Q north we will have a quantum computer powered by atom computers natural atom computers with our stack in there. We are working with algorithmic and Columbia and Zurich. We ourselves use this discovery agentic loop to accelerate the work in quantum compressing years of research into this last year. Today I am really thrilled to announce Majorana2. The resulting cube bits are reliable and maintaining their state much longer while other common approaches deliver a lifetime of micro seconds. Majorana 2 provide cubic lifetime of 20 seconds or up to even a minute. Essentially a thousand times higher than what we were able to achieve with high -- my or -- Majorana 1. They are enabling complex quantum computation in that lifetime. All of this in the same cubic form factor of Majorana 1 in one 100th of a millimeter control of digital. Which again is a super important aspect. Making it all possible to fit a million of these cube bits in a chip smaller than a credit card. It is this combination of the reliability, the speed, the size that makes the topological approach so unique. With Majorana 1 we had proven out the foundational physics and with Majorana 2 we begin the engineering scale. [Applause] Yup. Ultimately it is never about tech for tech sake.

It is about tackling those pressing challenges of people and planet. It is also the fundamental point of this conference. The question is not about whether you can build the next great model or platform even this quantum machine. It is the question of how do we build the frontier ecosystem together. Because there are really two stories people can tell about this moment. One is that technology concentrates power, reduces human agency and leaves the society to absorb the consequences. The other is that we use this next wave to unlock opportunity for developers, scientists, enterprises in every community. Our job is to make the second story true. That's our North Star for the frontier ecosystem. Let's all go build together. Thank you all very, very much. >> Thank you. [Music] >> Within reach. Quantum is our most fundamental physical theorem. For quantum to deliver real world impact it has to be both reliable and scalable. >> Reaching quantum at scale will allow us to make a more direct approach to solving problems in chemistry. >> Discovery in life sciences. >> We bring drugs to market it opens up millions of possibilities. Quantum will change the world. This change is just beginning to happen.