7:55
Francisco with all of you at the Windows event. So those are the three main things we've been working on. And today the idea is how to bring these three things to Windows? So we take Agent 365, Microsoft IQ, and Copilot. Those are the three main pillars of what we do, right? Agent 365 covers IT operations, security, financial operations -everything you need to truly trust your agent. Microsoft IQ is essentially a system around a model. In fact, over the last, I would say, three or four years, we've realized that just having a model is not enough. You really need to manage it. You need memory outside the model. You, of course, need this layer of control, capable of combining multiple models, context, memory, and action space, to
actually direct the trajectory of the model's output. And that's exactly what Microsoft IQ does. And the last thing is form factors, right? It's everything from chat to collaboration, to writing code, and now, of course, agents and autopilots coming together. Let's start with insulation. Existing sandboxes are insufficient to protect agent workloads, where isolation requirements can change with each tool invocation and each request. We created Microsoft Execution Containers (MXC) to meet this need. With MXC, agents are in a composite " sandbox" during execution. Organizations can set policies about which files and networks an agent can access, and Windows enforces those policies while it's running. Today, I'm excited to announce that Microsoft Execution Containers
is moving from a pre-release to a general release on Windows 11. MXC is a cross-platform library. It works on Windows, Linux, and macOS. Only on Windows is MXC deeply integrated into the OS. So, organizations get more isolation options and more control over their data. The spectrum of isolation that Windows offers ranges from processes and sessions to WSL C, virtual machines, and Windows 365 for agents. Developers can start integrating their agents with MXC today. Next, let's talk about identification. Regardless of whether the agent works under your direction or independently, every action must be attributed to the agent, not to you. Only Windows provides such a distinction at the operating system level. It is this distinction that makes it possible to
observe actions. Next is controllability. With built-in isolation and monitoring capabilities, Windows takes agent manageability to a new enterprise level. With Agent 365, IT sets policies once, and Windows applies them to every device. Windows turns agents into assets that can be managed across the enterprise. So, let me summarize. Local agent models have spread rapidly, but they are fundamentally dangerous because they have broad access to the system. With MXC, Microsoft Security, and Agent 365, we are unlocking the full potential of agents on Windows. Here is the website I created. I am using GitHub Copilot to update the Python version on my site according to the instructions on this external page. I am
not aware that this page contains malicious instructions aimed at redirecting agent behavior. As with a user, Microsoft Defender detects and blocks the attempted malicious action and also marks the agent as risky. That's pretty cool. Now the detection tells us that something went wrong. Distinguishing between agent and user allows organizations to respond accurately and contain agent-related risks. In fact, Windows makes this possible through OS-enforced agent attribution, which separates the agent's actions from mine. This means that access restrictions apply to the agent, not me. Without this attribution, think for a moment, blocking this malicious agent could also block the user, reducing productivity. Here I ask the agent if he can access this
document. What would happen? The agent no longer has access to SharePoint. But my access as a user of this system remains unchanged, so I can continue working. We are moving towards a future where hybrid intelligence is always available. Works locally when appropriate, turns to the cloud when necessary, and seamlessly combines the best of both worlds. Bringing hybrid intelligence to life requires intelligent routing, models, runtimes, and processors that work together to create a quality experience. We are working with the GitHub team to bring our vision of hybrid intelligence to Windows. Earlier this year, GitHub launched Hydro Fusion, which routes each task to the
appropriate model in the cloud. Today, we are extending intelligence to Windows, allowing Hydro Fusion to use models running locally. And let's talk about these models. At Build, we introduced MAI code 1.0 in flash, a model for writing code with 130 billion parameters, built for real-world developer tasks. Now we're bringing that same intelligence to your device. This is where full-stack engineering comes into play. We have converted this model to three-bit precision, reducing its size by almost 80% while maintaining code quality. This is an incredible engineering achievement. But in reality, reducing the model for the device is only half the battle. MAI code can run an impressive 256K context window directly on your device. Our goal is simple.
Allowing AI to run locally, remaining responsive, while leaving space for everything else on your PC. Taskbar search is where intent begins. Therefore, it is quite natural that our story begins with him. The new Windows Search is built on speed and efficiency, so you can find apps, settings, and files faster. Starting this fall, you'll be able to perform thousands of quick actions right from the taskbar on your Windows 11 PC. You can switch to a dark theme. You can turn on Do Not Disturb mode to help you focus better. You can increase the microphone volume, and then use the slider to set exactly the level you need. All of this as fast as you
can type. And actions can apply to more than just settings. You can take a screenshot, arrange windows- all from the taskbar. And my personal favorite: you can send a text message to a friend or colleague to warn them that you'll be late without taking your phone out of your pocket. A fast and intuitive way to move from intention to action. Of course, you can connect to the new Copilot app right from the taskbar whenever the need arises. Ask a quick question and Copilot will provide the answer right in the search bar. If you want a deeper conversation, you can go straight from search to the full version of the Copilot app.
Code is the easiest way for anyone to create and share an application. No developer experience is required. Hybrid intelligence improves code in three key ways. First, you can now create your own Windows application with a single request. This means an unlimited number of apps and widgets, completely native to Windows, straight from the Copilot app. Creating software for Windows has never been easier. Next, we implemented MXC support. MXC support means that your code, even local, runs in the sandbox we mentioned earlier. This means that for the first time, not just developers, but anyone can confidently
write and run code on their local computer. And lastly, local models. Copilot can now offload some of the work to a local model while writing code. This means projects move forward faster and cost less, as we saw with GitHub Copilot. These three things turn your PC into a personal software factory Introducing Autopilot, we are excited to announce the first-ever, long-running, enterprise-grade autonomous agent. Let's see how hybrid intelligence makes Autopilot even better. Let's move on to a quick demonstration. So, over the past few weeks, perhaps a little late, I've been working with Autopilot to gather a bunch of paperwork for my tax return filing. So, I'll start with this simple query, and the
first thing Autopilot will do is go into my email, find the letter from the accountant, and extract a list of files that we need to collect. Thanks to hybrid intelligence, Autopilot can now use data from my entire PC. And the documents I need for tax reporting are scattered everywhere. They are in downloads, in documents, some are organized, some are not. It doesn't matter. Autopilot uses hybrid intelligence to process absolutely everything taking into account the local context. With my permission, he begins tidying up. It
moves everything to a new folder, groups the files together, and even renames them. This entire process of ordering and renaming happens using local models. My tax data is not transferred to the cloud. It creates subfolders for better structure, and then archives everything it finds into a single ZIP file that I can easily send to my accountant. Of course, Autopilot always goes a step further-it also creates a draft letter to my accountant for me with the archive already attached. But of course he stops here to make sure I'm ready to ship. I check everything. Looks great. We can send this letter. And so, all these files were sent to the accountant, and we were done. That's what Windows is. This is what
We built it for developers and creators who go beyond productivity and value portability and exceptional quality. I've been using the Surface Laptop Studio for the past few months, and it's a real game changer. Developers put a lot of strain on their machines. Here is an intensive developer workflow with GitHub Copilot. But a demanding task shouldn't mean everything else has to stop. While the work is in progress, the film is being edited in DaVinci Resolve . Such installation requires serious productivity. The CPU, GPU, and memory work together while the developer's workload is still ongoing. And Gears of War, which came out yesterday, has been running on this machine all this time. And it works great. You can't
talk about Windows and NVIDIA and not mention gaming. At launch, hundreds of your favorite AAA and indie games will run beautifully on RTX Spark. We are very excited to announce that Call of Duty will be released on RTX Spark in 2027. We built Surface Laptop Ultra with NVIDIA RTX Spark. With up to 128GB of combined memory and up to one petaflop of AI performance, you can run models on this laptop that simply can't fit on conventional devices. It has 2.5 times the thermal capacity of our current laptops. Therefore, it remains fast even under prolonged load. And you wo n't have to worry about losing performance when you're running on battery power. It retains over 99% of its power when
running on battery power. And that's great because it's portable. Thickness less than 18 mm and weight less than 2 kg. I also love our new PixelSense Ultra touchscreen display. It has the brightest peak HDR display of any laptop. He looks incredible. Our Surface Laptop Ultra has the most ports of any Surface Laptop. Our customers told us they love magnetic charging. And they appreciate the versatility of USB-C. That's why we invented magnetic coupling. The world's first built-in magnetic USB-C charging. Simply lift the connector and it instantly locks into place. Convenient magnetic charging without having to give up the USB-C port. Surface Laptop Ultra delivers faster first token generation for language models and faster
image generation times than M5 Pro-based laptops. After Build, many of you asked about the Surface RTX Spark Dev Box. For those looking to further expand their local AI development capabilities, it comes with Project Zenith-a ready-to -go Windows environment pre- configured out of the box so you can start developing from day one. I'm excited to announce that Surface Laptop Ultra and Dev Box are available for pre-order today. RTX Spark brings powerful AI computing to laptops. DGX Station takes this to the next level. Take a look at the ecosystem that makes this happen. Lenovo Yoga 9 2-in-1, lightweight convertible design with pen support. ASUS ProArt Studiobook 14, an incredibly thin design with a 4K OLED display. Dell XPS 16, seamless glass
touchpad with built-in speakers. MSI Prestige 16, a versatile form factor with swivel, pen and UHD + OLED display. HP OmniBook Ultra 16 with a special hinge design and gesture control on the touchpad. And, of course, the Surface Laptop Ultra. All of these RTX Spark laptops are available for pre-order today, with shipping starting next Friday. We have a supercomputer. DGX Station for Windows combines. With RTX Spark, we bring the latest technology to your laptop. With DGX Station, we bring last quarter's technology right to your desktop. Look at this. the first time, DGX Station runs on Windows in real time. With Dell Precision Pro, advanced models and agents work locally, turning months of complex work into moments, providing greater control over
sensitive data and costs. The HP ZGX Fury AI Station demonstrates how to run and scale with 32 agents simultaneously-a virtual token factory for development teams. Both devices will be available later this year, running on the Windows platform that organizations know and trust.
