You and AI -- Building The Future Together

Well, thank you so much. And uh you know, welcome to the Microsoft campus. Uh I've been lucky enough to to be here on the campus for uh close to 16 years. And I'm really happy to uh to have you all come here and join us. And you know, from the Visual Studio and .NET and SQL and C++ teams and Azure and whatnot, we got a bunch of speakers, we got a bunch of engineers, people actually build the products. They're all here in this building, they're in the booth, they're like hanging around. And they all really want to talk to you. This is the conference where this happens. It doesn't really happen anywhere else where we have that concentration of uh engineers, people who make the products that we all use every day. Uh and they really, really, really want to talk to you. So, please don't be shy. Go tap people on the shoulder. Uh you know, like Brian was saying, "Tell us what's wrong." And we'll try to help you out, right? Or tell us how we can improve because that's that's kind of why we're here. And uh so, please take advantage of that. We will do that, too. We might be poke, you know, uh coming up to you and ask you a few questions out of the blue. So, be prepared. So, um yeah, Visual Studio and AI, that's what we're going to talk about here today. And uh you know, it seems like every session is about that this this time around, right? It's been It's been coming. It's like everything is about AI, and everything is moving so quickly. And it wasn't that long ago where Visual Studio AI kind of felt like it was tagged on, right? And now it seems much more integrated. And things have moved pretty quickly, not just on Visual Studio, but everywhere in this space. And it could be kind of overwhelming. Like things are moving. Things you thought was the new best big thing that you learned very recently is like no longer being talked about. And it it it it's overwhelming. Um the the pace of this kind of is is staggering, and it's faster than anything else that at least in my time uh I remember in the industry happening. So, um Like an example, do you remember uh prompt engineering? This used to be the thing like you could be a prompt engineer. This is like a year ago. You could now be a prompt engineer. You'd see people hiring a prompt engineer. When was the last time you heard even the term prompt engineering? Like it came, it was the biggest thing, and it went away. Um >> [clears throat] >> Oh. Here we go. We also hear this a lot. Like if you're on social media, Twitter, at conferences, when when when was the last time you wrote code? I don't write code anymore. The agent does that for me. I have multiple agents. In fact, I have agents that run agents, so I don't have to write or look at code. It's like, "Oh, that seems that seems a little extreme, maybe, but you know, makes me also feel like a little anxious maybe that I'm not doing that. Does that mean I'm falling behind? What did I not learn that maybe I should have learned?" It's like, "Ugh." I don't know. Is it a good thing? Maybe. It's also kind of doesn't feel good. Um And the future is now. It's agentic. I think we all know this. We know that AI is kind of disrupting and taking over parts of our jobs and parts of our responsibilities as ours then shift to do other things and manage all the agents and stuff. But it's all like not necessarily clear how that's going to work out. There's a bunch of people that have very strong opinions about it, but it seems everyone is a little bit uh spread out on the spectrum of a little bit of AI and like I have a Mac mini that orders me lunch every day without me even knowing that I like chicken on Tuesdays, right? Something like that. And so, we all fall in the spectrum of somewhere in between. And this This kind of what we see on social media. This world of just, you know, Minority Report, like managing all these agents, I'm in control, and all these things are happening around me and sub agents, and all these skills, and all this thing. And and um like, "Oh, yeah. Maybe not. Maybe not the job I used to have. Like, maybe I like programming. Maybe I don't like being the Minority Report person." But, what about this? Uh this I can get behind. What if What if having uh the agentic workflow being more like something that frees you up? So, yeah, you're managing agents and stuff, but maybe you have time for a little uh just sit on the beach while you do it. So, this is kind of the ideas that you see a lot of floating around about all this, and it's I don't know. It doesn't feel like reality, which looks a lot more like this. Like, this is what my reality looks like. Probably what yours look like. We can't just sit. We have Visual Studio open. We probably have a terminal. We probably have some other tools as well. And we code, and we have a good time doing that. That's why we're in this business. We love coding. We love that technical thing, whatever it is, if it's coding, or you know, database solutions, or whatever it is that that made us interested and passionate, and come to conferences like this about this stuff, right? What? Is the microphone not on? It's on. Oh, it's it's simply just too far. Oh. Here, let's do something else. I'm not demoing, so I have a hand free here. All right. So, um this is kind of um the reality, right? We're still sitting in front of our screens and doing kind of this work the our environment is similar, but it's still different. Because AI allows us to shift how we're working, how we're prioritizing, and what we can get done. But, you know, there's some blockers for some people. Uh it's gotten pretty expensive. All right, so when you're out of tokens, like maybe you can't afford running agents after that, and you have to wait till your token budget resets every month and whatever, right? It's It's simply getting a little bit too expensive. Maybe you don't get the budget that you hope to get from your organization. It could also be that you're limited in how you use AI because of privacy concerns from your company. This is absolutely very real. We hear this all the time. And so maybe you can use AI for some things, but maybe you can't use it for coding purposes, but you can use it for like saying writing documents and and planning and other things. And so you use whatever you can. Uh wherever it makes sense. But at the end of the day, things are moving so fast that you might not have time to just keep up with all the things. Like not that long ago, open club came out and was this big thing. You can run your LLM. You get a Mac Mini, and it does everything. It can reply to your emails. It can organize your entire life and is your own personal assistant. And that seemed like really interesting and fascinating, but just imagine the amount of time it would take for you to set something like that up. We're professionals. We don't have time for that stuff. Right? Or that's true for a lot of people at least. So um it's hard. And and often we are seeing these workflows, and we're hearing about workflows, but are we really taking advantage of all the things that are available inside Visual Studio or whatever tool we're using? Are we getting the workflow that we kind of hope for to really up our productivity to make us all rockstars as promised? Um there's some learning, and there's a bunch of uh that goes into that for how it will work in your organization and on your team. And you haven't had time to look at that yet. And that's totally okay. Maybe you just don't want to organize um eight or orchestrate agents. Maybe you are not you don't want to be that Minority Report person. You just love coding. That's why you're here. That's why you're doing what you're doing. And um I think a lot of people, especially people that have been in the industry for many years, they're like, "Hey, please don't take my passion away." You know? So, good news is that if you feel like that, you're the 99%. This is very common, right? So, one or more of the things I've been talking about probably applies to all of you. Um this is very common. It's very few people in the real world, people who write actually the real software running on in real companies and all that type of stuff, not just people on Twitter that, you know, vibe codes uh iOS apps that no one uses. Um but the people who build the the real software, the people in this room. All right. So, with Visual Studio, we understand this. And we understand that we come from different starting points. And um whether you're just dipping your toes in the AI pond, or you're kind of more diving in and already swimming and whatever it is that you're doing, how you're using it, you're again on the spectrum, it's all very, very different. Uh but um with Visual Studio, we want to make sure that no one is falling behind or feeling like they're falling behind. I don't think anyone is actually falling behind. It's more of that feeling of anxiety and all that. But it's also about productivity. Are we actually taking full advantage of something that's in front of us? And uh we want to make sure that everyone that can take advantage of it will take advantage of it to gain that extra productivity. And it's clear that in doing so your skills now matter more than probably ever before. Because if you're now starting to work with agents you're now the senior in the room. And you have all these junior agents that can never know the same amount of stuff that you do about your company, the way you want the code to be written, the way you want architecture, the thing that you decided in that meeting with these stakeholders yesterday about the future of something something something. There's no way AI that knows that. You know that. And you have to now steer and be the senior in the room. Whether you're senior or not, you're the senior. All right? And so having skills and know fundamentals about computer science, knowing have technical depth in the things you're building is really really important to do this optimally. And um typically when we build software we do it in like three categories or there's three tasks if you will. So one is the why. Uh this is the you know, the customer goal, the business goal. What is the problem space that we're trying to solve for? Let's say we want to build a new feature. Well, what what is the problem we're trying to solve? So we might do customer interviews, we might do surveys. Uh we might have uh you know, customer feedback channels that let us know what matters to users and customers. And then the what is okay, then how do we What is the thing that we're doing to solve it? This is the solution space. Like what's the architecture? What's the design? So here you have the architecture, you have the designers. You might have uh senior engineers and such. Um this is kind of the solution space. And then finally we have the how. So this is the implementation. Actual coding. Getting stuff done. Getting the CI/CD pipeline up and running. Right? Um So usually we see these type of disciplines and we have different uh people in each of them. So the the um the problem space is typically like a product manager. Then you have an architect on the on the solution space and then you might have a engineer on the um coding and the CI/CD pipeline and that stuff that stuff. Um And of course sometimes we can hold more than one role, which is very common too. It might be that with the emergence of AI, you're going to see your matrix here and your priorities shift a little bit. It might be that where you were you know, could do a little bit of everything before, maybe you say, "Hey, I'm actually most passionate about the solution space. That's where I want to spend my time." So I might want to spin up agents to help me out with some of the other things, the why and the how. So I can spend the time that I love. Maybe it's more like coding. This is a lot of this is passion to people, right? Actually writing code. So if that's what you want to do, do it. And maybe have help with some of the other things from the agents. So I'm a product manager. So I usually live in the why. Um I also like kind of making prototypes and stuff to figure out what could be a solution. But and and anyone on the Visual Studio team will tell you that you definitely don't want me to write any code in the Visual Studio repo. So we'll keep that very very tiny. But the cool thing is that AI helps me do that. It helps me get rid of the things that I'm either not good at or not or want to spend time on. Um so I can really double down on the why in this case. So, um, it might look a little bit different uh, in the new world where AI can help us out. I think this uh, the problem space is pretty similar. Like, what does the business want to achieve and what issues do the customers have? The problem space doesn't really change just because we have a new tool in our tool belt called AI. But, what really changes is the what because we can now as architect, we can come in and we can do uh, a lot more and a lot faster. So, the way we define write specs and all this type stuff uh, can really be elevated. Can we can really take advantage of AI that can go out and synthesize all this data that otherwise would have been pretty difficult to come up with like clear evidence of why we want to solve a problem in a certain way. We can also look at uh, scope beyond what we're just normally doing. Like, how does your product fit into the larger ecosystem of your company's other products? Does product A contribute to product B and the other way around? Sometimes it's like, let's just focus on our thing, right? And then if you know, something may come up later where we can do some sort of collaboration, but the thing is that with AI, we are now elevating ourselves. We can we can move ourselves up to a higher ground. Take the helicopter up and look at a much larger picture all of a sudden because we can like use agents to help us get all the data and align all the different things and so we can think bigger. Which is absolutely phenomenal. Um, and the same thing goes for how. Um, if we don't want to write so much code because we have help doing that we're going to now stress the reviewing and the verification of everything we're doing. We're going to relatively spend more time building our solutions, running them, debugging them, verifying that it works according to the problem we wanted to solve and the spec on how to do it. Right? So, having those skills and those tools being super fine-tuned for that is now more important than ever. And Visual Studio is here for all of this. Because for all of these different tasks, we need to understand our, you know, basically our files, our projects, our solutions, how our dependencies work. We need an interface to talk to the agents. We need our chat window, right? And other types of integration that really can elevate uh our productivity. And Test Explorer becomes more important than ever because uh when you have agents writing your code, how do you know that that code doesn't break something? Uh guardrails in forms of tests, integration tests, unit tests are like crucial. And finally, code. Because you want to code, you want to review code. Like read code, write code, you still want to do those things either because it's a really good idea or because you were like really passionate about it. Either way, those tools that we know and love, IntelliSense and so on in Visual Studio, a strong debugger, is not going anywhere. Right? Visual Studio is here for the same reason today as it was here 20 years ago, 29 years ago, when Visual Studio was first uh came to the market. And one thing that we're very clear about when it comes to Visual Studio is who's it for? Because you have colleagues, they prefer VS Code, some prefer, you know, cloud code or codex or the CLI experience or whatever it is. And how do we make sense of all these different tools and what to choose? Well, from Visual Studio's point of view, we're very, very clear on who our target audience is. It's you. The professional C# and C++ developers. That's it, okay? Professional developers. And there's a reason for that. And that is because they share something that we don't really see it a lot of other cohorts uh care so much about, and that's quality gates and tenants. We care deeply about security, privacy, accessibility, governance, and so on. And so having the tools that allow full uh administration for like big companies to roll out a fleet of Visual Studio license um machines, right? And uh have the tools in Visual Studio to handle all these different things remain a super focus for Visual Studio. You don't really find tools like this anywhere else. Um and at the end of the day, Visual Studio is for developers who care about code as an asset, not an artifact. Like if you were to just to do a prototype of something, you don't really need to look at the code, right? You can just ask the agent to build you a prototype of of whatever it might be, and you can throw it away. That's fine. So for projects like that, you know, use whatever floats your boat. But for the projects where the code is an asset, like you have to maintain this thing, not just you, but your colleagues and your future you, your own worst critic, right? Uh you want to make sure that it's uh as high quality as possible for now and years to come. And the priorities from Visual Studio's perspective does not change just because AI gets in the mix. If anything, we're doubling down. Because where we see some other products are kind of trying to find their path, we know our path. Professional developers. And so, I want to show you a little demo of a thing that shows how in Visual Studio the tools that we have and that that we rely on works great for humans, like we can use these tools. But when we sprinkle AI in the right places and we really implement AI around the depths and the roots of these features as they intersect, we come out with an experience that's quite unique and is uh leaning into the power of a full-blown IDE that Visual Studio is. So, let's check this out. >> Hey folks, today I'm super excited to share some of the new features we've been working on in Visual Studio. Uh specifically a feature around helping you improve the performance of your code. And so, to do that, I want to show you this feature working on some real code that I grabbed from nuget.org. So, searching through some of the packages there, trying to see if I could find one that would be kind of cool for a demo, and I found this QR coder library. Uh it seems to be a pretty popular library. It has almost 60 million downloads, uh and it basically helps you render QR codes in your .NET applications. And so, I went ahead and I cloned it in Visual Studio, and you can see I've got my project, I've got some benchmarks, a console test app, um bunch of different things. Uh and I want to go ahead and I want to see if I can make a improvement to their code. And so, I could run this under the profiler, um kind of doing existing profiling workflows, uh but I don't really have time for that in this demo today. And so, I want to click the easy button. And so, for that, we've gone ahead and we've added some new entry points into the profiler and the test explorer. And so now, on newer versions of Visual Studio, all SKUs can profile different unit tests. So, if you have like an integration test, this is a great way to get into the profiler, get actual performance data out of it, make changes, and kind of see the impact of those changes that you're making. But what's even cooler is we now have this profile with Copilot button. And what this does is this tells Copilot to go ahead and run the test under the profiler, try and make changes to the code that it's testing, and then try and implement some optimizations and measure the impact of that. And so, you can see it says it's identifying the scope. It went ahead and it ran the profiler to establish a baseline. And then it can start digging into um that trace that I captured. And so, you can see it said it sees the key findings. "Should match approved image." This is test framework code, making sure that it actually kind of asserts and matches what it should be expected. It says, "Can't be optimized here cuz it tries to ignore any test framework code." But then, almost half the time is spent on the QR code generation, which is good. We That's something that'll be a good target for optimizations. And then it says a moderate contributor is the draw scan lines for the actual image rendering. And so, you can see it says it's the generate QR code is definitely the thing to try and implement. It's going to go ahead and implement an optimization. So, it found specifically where it should try and implement. It found some of the functions that it should be looking at. It's digging up and down the stack trace that's in the um trace that it captured. And it can see um where things are actually being called and uh what is happening. And so now it says now it has a comprehensive view of the hot path. What are the key optimization opportunities? There is this mass code module placer. Uh it calls these um things multiple times uh because of the fact that it used an instrumentation profiler in this case, uh it gets actual call counts. And so it can see like if you've called things multiple times and it can optimize that by um just making the call once and hanging on to the result. Uh it found a point where we use floating point math, which is slower, uh and it thinks it can use integer division. Uh and then it found another spot where it looks like we're doing some indexing into a module uh a 2D array uh matrix and then it can now cache the row reference. And so it's reading the code, it's going ahead and it's doing the optimizations. Uh it went ahead and it actually uh it's kind of getting ahead of us a little bit. Uh it is going ahead and it's done the optimizations. You can see it made changes, it tried to do the build, and it looks like it was doing method input options, uh aggressive inlining. My guess is it was trying to inline a function call, uh and then it's seeing that the namespace isn't there, so it's going ahead and it's going to uh correct that. And then uh with any luck, the build will now pass, and then it is going to go ahead and it'll rerun that unit test, and then it'll see the impact of the change that I made. So our build is successful, which is awesome. Uh it's now running the test to verify correctness. Uh that's one of the benefits of using tests to go ahead and profile is it knows that the test will actually test for for functional correctness and so as it makes changes it can make sure that the um it's not breaking the functionality of it and then it can make sure that it is actually faster and so it said there was a failure on .NET Framework only and that's regarding um source information uh but it works on .NET 5 and so it's gone ahead and it re-profiled it uh and made the changes and you can see it then list off all the different changes it made uh it looks like the create QR code is faster uh place modules which previously was in our top functions is no longer listed that makes sense if it was probably inlining it um and then it was making some additional changes and then that pattern five where it switched from uh float arithmetic with a floor down to integer math uh it should have improved that there and so now the self percent dropped by approximately 63% reflecting you know um that it has in fact improved the code uh and so I can go ahead and I can look at the code and I can see like oh you know it went from uh obviously floating point math uh down to integer math maybe that's faster maybe I don't know why and the nice part is is when Copilot is doing these optimizations I can follow it up with my own questions and I can say why is the integer math faster than the floating point math and then not only will Copilot go and try and optimize my code it'll actually teach me along the way like hey this is why um this is faster and so talks about uh floating point unit versus ALU pipeline. Um it says that math.floor does far more work than just, you know, right shifting by one, uh and then the floor is redundant, and then the cost of this is compounded. And so it looks like it's doing it in for every cell of the QR code. Uh so it's a 29 by 29 grid. Uh so there's over 6,000 calls to it. So even a little bit of optimization where you're optimizing, you know, very small amounts of floating point to integer can actually have a pretty big impact. And with that, we can, you know, we could continue to click on buttons here and uh continue to follow up on more and more things and have Copilot optimize stuff for us. Uh but it really goes to show like, you know, couple of minutes here I've been clicking on some buttons that launched into a performance investigation. Copilot helped me find and identify things that we could change that actually have a positive impact on the code. Uh and then it made those changes, validated them for me, and then taught me something along the way. And so with that, thank you everyone, and I hope you enjoy your time at the conference. >> All right. Thank you Nick, who couldn't be here today, but Nick, he's uh he works on the Visual Studio team. He's on the diagnostics team, and and his team is the the team that built the profiler. And so what this illustrates is, I think, very uh clearly how we're looking at how Visual Studio fits into this picture both now and going forward. Like you have tasks that you do today. You probably rely on the debugger because it might be the best debugger in the industry. Is that that's what you tell us at least. Um so and and we believe you. Uh the profiler is one of those things that helps us instrument our code to learn how we can optimize it. And that's what he did here, and he did it from the test explorer. So, there's an integration here that otherwise would have been kind of maybe difficult to to do any of the the profiling um without it. But, what's really nice is that you can write any test, you can do integration test. As long as you can start the profiler from that test, you can have the AI help you out. And I don't know if you know the difference between floating point math and integer math. Yeah. Maybe. I for sure don't. Right? And that's the beauty of this. Like, I can run that profiler. I would have probably been able to squeeze some performance out of uh whatever I was profiling, but I would never in a million years have thought I could just you know, maybe I didn't even know that was something called floating point math, to be quite honest with you. That would had a different performance characteristic than any other type of math. How can there be multiple times of types of math to begin with? Right? And that's the beauty. You don't have to be low-level, know how, you know, the bits and bytes of the compiler is running on the metal. But, the agent knows. And so, it can help you learn how to do this going forward. It can help you learn why it might not be a good idea to accept an I list in a as a parameter, but maybe you should have a more specific type that has better performance characteristics for the specific thing your method is using it for. It's those things that you probably don't even think about. And this can help you find them. And the good thing about performance profiling is that I don't think anyone has ever been fired for optimizing code without asking permission first. So, it's something anyone you can just go home and try it. Um some of the people from the profiler team, they actually when they go for lunch, they start the profiler agent to "Hey, while I'm gone for half an hour, go see if you can find parts of the code to optimize. And sometimes it doesn't really find anything useful and you just leave it alone. And sometimes you get a perfect benefit and you spend a little bit of time and you check it in. And now your stuff is faster. It's really, really significant and we've used the Visual Studio profiler to make Visual Studio itself faster. And other um products as well like Bing, like .NET. Um all over the map. Uh you can really use this for anything, whether it's UI apps or web APIs or whatever it might be, WinForms. Uh Brian's WinForms apps, she will work for those, too. >> [clears throat] >> So, but it's a capability that we've had already and we're adding value through AI to something that's important. So, we didn't talk about debugger, but you can go through the debugger agent as well through the unit tests or through test explorer, rather. So, we're talking about this thing about workflow. Well, how do you start taking advantage of all these things? Because one thing is that we have this in Visual Studio, but what could be a good workflow? Um I'm going to just do a little bit of uh uh prescription here, but if you really want to see a full workflow, come to Sebastian's talk about how we use Copilot internally on the Visual Studio team and what our workflow is. It's a really quite uh interesting. But the first thing we need to make sure we do and I alluded to this was that we have to add sufficient tests. This is kind of the pre-requisite. We can't really take full advantage of AI until we have guardrails in place. Some constraints for the uh agents to work within. Otherwise, we're just setting ourselves up for like having to run all, you know, all this stuff manually. We have to verify a lot more manual stuff than we otherwise would have. So, Hello, Brian. Also, get your backlog in order. So, this goes for both your product backlog, like your user stories, but also your any technical data or bugs that you might have. The getting a backlog in order doesn't just mean prioritize, it means that you have clear titles and descriptions. Because if you're going to let an agent loose on your backlog, you want to make sure it does kind of what you expect it to do. And so, having those clear descriptions, if you have links to uh user feedback or anything like that, make sure you get it in there. The good news is that the agent will help you with this. So, I do this all the time. And uh if you give it links to certain things, it will start creating a very, very detailed description of existing GitHub issues or ADO tickets or whatever it is. So, uh or Jira, what whatever you're using today. And so, that's a really good place to start. So, once you have the tests and you have a backlog of things that you're comfortable having the agent run, you can start assigning work to the agent. And so, you know, whatever floats your boat. If you want to use Visual Studio, if you want to use the CLI, whatever you want to use, assign the agent to the work. Maybe you want to use github.com, you can actually spin agents up directly from the website if you want to. Uh which makes sense if you have something it it can run and you, you know, maybe you don't have to verify it that um specifically like you do from within Visual Studio, where we have this very specific thing that is actually fantastic. When we have our code, you click F5, you run it, you verify things are working, and you debug it if it doesn't. That is an immensely powerful inner loop, is what we call it. You might have heard us talk about the inner loop. But that whole idea of like coding, verifying, testing, debugging, coding, you know, we're talking about a loop. We can now add profiling into that as well. Because it's gotten so simple and so fast because we pointed the profiler at the profiler. And so, this becomes a very, very powerful thing uh for any developer, professionals in particular because we really value those debugging sessions and those logs and all those things when we do run our apps. So, this is a typical kind of very high-level uh getting started workflow. Now, for the specifics, go to Sebastian's session uh later in this conference. So, Visual Studio going forward, we're going to double down on some things. We're going to add some investments around certain key areas that are important with or without AI, but certainly with AI. Okay? One is about testing. Those guardrails are important, and so the experience you have adding tests has to be as smooth and delightful as possible. Like we have an agent the test agent that makes it really easy to add unit tests. So, if you haven't uh played around with that, and maybe you don't feel that you have enough unit tests, give that a go and you'll see uh some true magic happen to your code coverage. Uh but it has to be a very, very pleasant uh experience, and so we're going to invest more on testing. Same goes with diagnostics. Debugging, profiling, all that stuff. Um we are maybe the best. We want to it to stay that way. And we want to give you all the tools both for humans and for agents so that you can decide when and where you want to use agents and when and where you want to be driving the whole thing or anywhere in between. Uh build and run. We're going to do this is now relatively to coding than we have ever before. Because we're now being We're going to be a little bit of a QA person sometime. We have agents doing work on our behalf. We need to verify things. So, how do we speed up builds? How do we make that experience better? Uh running of your app and all that type stuff. Uh integrations. If you're building something on Azure, you're using Azure services, could something light up in Visual Studio that makes that even easier? Um we already have some We're shipping some in preview right now that will help with that. What if maybe Brian's WinForms um applications here? What if he could have expert help with writing WinForms? Maybe Wouldn't it be cool if you could have a uh like you could basically have a a member of the WinForms team sitting next to you while you're building WinForms apps to kind of help you along the way. Seems like that would be pretty amazing, wouldn't it? Um I'm not saying we're going to ship members to your companies. >> [clears throat] >> And then review. Like you're going to review a lot more code than usual. And so, how do we make sure pull requests and both like reading a pull request, but also if you wanted to send a pull request somewhere? Uh again, it doesn't have to be an agent. It's humans. Like we we want to review code. We want to have a great experience uh as much as we possibly can. So, um maybe I want to get a code review of my code before I send it into my colleagues. Especially like And we know this because um a lot of people tell us this that there's certain anxiety associated with sending pull requests because you're now inviting other people that might be your seniors to look at your code that you may not feel is as good as they would have written it. So, it would be nice to have like an impartial person or agent or whatever to review your code to catch common things first, right? So, that would be nice. And then it comes time to commit. And uh the mouse hovers over the commit button and what you're feeling? Please don't result in a merge conflict, right? Anyone here loves merge conflicts? No. Uh So, we want to make that a thing of the past. Um So, again, whether you're human, whether you're an agent. And um all of this is how professional developers use the product, the thing they rely on. And then we want to make sure we add the meaningful AI to where it makes sense. It doesn't mean we're going to bolt on AI for the sake of AI. It has to mean that it changes something that you're doing for the better. You're going to do this anyway or you wish you could do it, but you don't know how. So, we want to make sure we get that in there. The AI word makes sense and where AI is the best solution. Okay? The problem space is the same, but we have another tool in the toolbox, just like you do. And so, from Visual Studio's perspective, we can now use AI to solve problems. All right. So, what got we What have we got here coming for you? Let's take a look. So, we heard from uh Dalia this morning. Visual Studio has used its own implementation of Copilot for quite a while, and that's now changing. So, it's changing to use the Copilot CLI SDK. So, all the GitHub Copilots, whether it's in VS Code, the CLI, Visual Studio, or elsewhere, it's going to use the same foundational component, the same SDK. So, you're going to have the same experience across all these and they We're looking at how we can share sessions, so you might be able to start in the CLI, move that into Visual Studio, and so on. But, right now in preview, we do have the new SDK. You can choose the agent in the agent drop-down, you can choose the preview agent, which will run um Copilot on the CLI SDK. So, no longer will you see all those uh developer command prompts show up and steal focus in the IDE. It'll all run uh the way you may be expected to. So, that's really, really nice. I've been using it for several months and it's fantastic. Look forward to that. Bring your own key. So, we talked about privacy and how that stops some people from deploying it in their enterprises, their companies because of, you know, the privacy concerns. But, what if you could run any model of your choosing on any hardware, on any URL, whether it's on prem, it's local, it's on your Mac Mini, or it's any cloud provider out there. It's Azure Foundry, wherever it might be. We're going to open up for that so that you can now host your own LLM and get that to light up within the Copilot experience without doing anything else. Okay? So, that's a big change coming here. You don't have to sign in with GitHub for this to work. Git worktree? So, that Ooh. Yes, let's hear it. So, Git worktree is a fantastic tool. Uh a lot of agents use it. Um and it's a really good way of parallelizing your work and have multiple agents work on different worktrees of your uh of your repo and and uh and make that all kind of more seamless. So, that's fantastic. Another Git feature is Git submodules. So, full support for that. Um it's now in preview. It's going to go into stable here pretty soon. I talked about the merge conflicts of which there were no fans among you. Um we're going to have a button. In case there is a merge conflict, there's one, we're going to add a button that says fix it. You know, it's not it's not being called fix it. It's What's it called? Andy, what's it called? What do we call the button? Uh auto merge, maybe? Um no, but what it does is that it will as best as it possibly can, just like you or I, it will look at the diff, the deltas and all that, and try to do it the way you would do it to uh solve the conflict. And so, this is absolutely a game changer because it removes that whole anxiety, it removes all that time wasted in figuring out and resolving merge conflicts. So, we're very excited about that one. It's coming in preview here pretty soon. And then we talked about this notion of having, you know, the WinForms team sitting next to you. Or the Azure App Service team sitting next to you when you're working on some Azure service that you're using, right? We do it through agent skills. So, the thing about agent skills is that it's a great way to add knowledge to the agent. Uh knowledge that it wouldn't otherwise know, like specific things. Very specific things, either to your company or to specific technologies like a specific Azure service, for instance. The problem is, how do you discover them? How do you find these skills? They're just markdown files. And they exist, people have written them, they're on repos around the world, you know. How do you find them? And which ones are good? So, we're taking that on and saying, well, what if the WinForms team wrote a bunch of skills and maintained that in a repo? What if the Azure App Service team and the Azure Fabric team and all sorts of different teams, they write the skills and they maintain them themselves? These are people who build the product. And so now when you're using the agent to write some of your code, they gain that knowledge from the team. Which is absolutely fantastic and we can apply it automatically. So you no longer have to go look for it. Visual Studio will simply say, "Oh, you're using some Azure stuff. Let me get you the uh let me apply the Azure skills that's relevant right now." Um and same with WinForms. And so on and so forth. So that is starting to roll out. The WinForms was the first, we just did some Azure. Um and uh and this is absolutely a game-changer for the AI because we take it removes that discoverability issue. So things are moving fast and so is Visual Studio. So we've already shipped quite a few updates. Every one of those uh orange dots or golden dots are the monthly updates we have done to Visual Studio 2026. So we used to ship quarterly, but we now ship monthly because the world is moving fast and we have to move fast with it. So make sure you update Visual Studio. Uh there's always new features available, there's always new fixes, there's always new performance gains as well. And we have um a good history. I'm just showing you a few kind of highlights, some of the things we've been talking about that we've been shipping and that we are shipping. Um and that moves us into November where we're going to see Visual Studio 2027. And what we talked about now applies not just for now, but also through next year and as long as we know, okay? We are here for the professional developers and the AI is not to take over, it's to be meaningful where it matters. Um and so Visual Studio will be here. It's with you, no matter where you are on your AI journey. If you say, "Hey, I'm just going to I'm okay waiting and just whenever Visual Studio is adopting certain features, that's when I'm going to join in." We got your back. If you want to say, "I want to be on the cutting edge and I want to use cloud and, you know, CLI tools and all this type of stuff, but I also want to use the Visual Studio debugger and so on." We got your back, okay? Integration and all that type of stuff. So, Visual Studio is here to stay. Visual Studio is here for you, the professional developer. Uh and that won't change. So, as I mentioned earlier, we are here. Everyone who People who build Visual Studio and all the underlying technologies, .NET, C++, Azure, and so on. SQL. We are here throughout the week. Please come talk to us. We really want to hear your story, want to hear what you have to say, learn how we can improve. And with that, thank you very much and enjoy the rest of your conference.