Tom Arbuthnot: What are you most excited about this year and this semester? Where's your focus?

Mahendra Sekaran: You know, there's several things I'm super excited about. One is reimagining how we get a job done is both exciting and challenging. I can tell you that even for me, like it was— it took a lot of intentionality because there's so much access to information. But just getting into this habit of leveraging agents as assistants for me or getting jobs done for me has been a meaningful change. And the efficiency gain that we get like where all of us can become like 10X PMs or 10X leaders or 10X engineers over the next six months is something I'm super excited about. Not just for the sake of efficiency. It is not like I'm just gonna be going and playing golf from the efficiency gain that I get. It is I can do more for our customers

Tom Arbuthnot: Hi, and welcome back to the podcast. Really excited to have Mahendra on the pod again, getting a perspective on what's going on with him and the IC3 team. Also, how that team are using AI in development, and Mahendra's general perspective on AI and agentic and what's going on in the space. Always great to have Mahendra on the show, really appreciate his time and insights. And also many thanks to Pure IP, who are the sponsor of this show. Really appreciate all their support to the community. On with the show. Hey, everybody. Welcome back to the show. We are in the summertime as we record this but into the new Microsoft financial year and first engineering semester as well. Very excited to have Mahendra back on the show. So it's been a little while. Welcome back, Mahendra.

Mahendra Sekaran: Thank you, Tom. It's always great to be back talking to you and the community through you as well, so thanks for the time.

Tom Arbuthnot: No, I appreciate you coming on, and it's an exciting time to talk 'cause you've kind of had the, had the end of year, had hopefully a little bit of time to, to think about what's coming and I'm, I'm sure plans are underway for the, the n- the c- well, now current year already.

Mahendra Sekaran: Absolutely. You know, I think you know, summers are usually slow. But this summer although it feels a little slow, it is faster than it has ever been. So things are moving at incredible pace and velocity and you know, it's a super exciting time to be working in tech, I would say.

Tom Arbuthnot: Yeah, we were saying this in the planning, like it used to be years ago, you know, it would pretty much drop off a cliff around this time, so there'd be a lot of pressure up the end of year, and then the summer would be no news, no major changes, and then we'd start ramping up again. Yeah. But everything in AI continues to move, so therefore everything at Microsoft is continuing to move as well.

Mahendra Sekaran: That's right. That's right.

Tom Arbuthnot: Awesome. So for… I'm sure everybody in the community pretty much knows who you are, but for, for those who may be tuning in for the first time, can you just give a little bit about your role and what you look after?

Mahendra Sekaran: Yeah. So my name is Mahendra Sekaran, of course. I lead the Product Management and Science team for M365 Core focused on communications and collaboration. So pretty much any, you know, like we basically are the platform that delivers comms and collab capabilities to products across the company whether it's Microsoft Teams, whether it's Copilot, whether it's Copilot Studio or Foundry agents you know. My team is in the middle of it, kind of delivering core messaging, calling audio and video experiences. So

Tom Arbuthnot: Awesome.

Mahendra Sekaran: Yeah.

Tom Arbuthnot: And let's start with Azure, actually, 'cause we just got the end of last year financial results, and Azure has had a massive surge of growth, a l- a lot of it driven around AI, of course. I'm guessing you're seeing and feeling that internally on, on your side.

Mahendra Sekaran: Absolutely. Like, I think one of the big things we are seeing is you know, strong, strong adoption of AI. It's reflected in, like, I think Satya and Amy kinda shared that in our earnings call, like, a week or two before. - But like our cloud revenue continues to grow Copilot continues to grow, and Azure growth is at about forty-three percent this time around. And what's, what's more interesting is to kind of like validate that customers are deeply investing in AI and the cloud are, you know, our remaining performance obligation is a pretty large number, and that continues to grow. And if I recall correctly, I think Amy shared that number in their earnings announcement. It was like six hundred and seventy-eight billion dollars or something. So yeah, it is it is certainly a lot of growth, a lot of demand and tons of opportunity here.

Tom Arbuthnot: Awesome. I wanna get a little bit into how you're using— your teams are using AI internally. But first it'd be interesting to hear from you. I feel like comms and collab is like a really hot area for AI, you know, over Dynamics, over Teams, over Copilot Studio. What's your perspective on that part of the AI space?

Mahendra Sekaran: Yeah. So maybe I'll start with you know, like how our teams are using AI because one of the biggest pushes we've had over the past year or six months is, you know, kinda reimagining how we work with AI. And this is a journey just not for Microsoft. I think like every customer, every enterprise, every business out there is gonna be going on that journey. If I see like how things like PMs in our team, the way they work has changed significantly. I'll give you a, a small example, Tom, because, you know, you started off the podcast kinda referring to our planning cycles. Because we are a platform team we get, you know, feature requests. Like, we have our own opinion about things we have to go build in the platform in terms of being able to look around the corner, understanding, you know, agentic usage patterns and growth patterns, and then preparing the platform for it. At the same time, we get feature requests from, like, tons of teams around Microsoft. And the way, you know, traditionally it used to work is our PMs get those feature requirements, they work with their peer PMs and the other product teams to come up with a good feature requirement spec. We give it to our engineering managers who kinda go through the costing exercise, and then we have a prioritization framework where we run pretty much all our intake requests. And then we determine, okay, here are the things that we are able to fund, here are the things we are unable to fund, and here are the things that we believe like a way model coding will work, where those teams will come in and work on our code base. That is a traditional model that we used to have, and, you know, that, that was like, as you can imagine, when you have multiple humans involved and not very clear alignment on priorities there's a lot of back and forth that

Tom Arbuthnot: Happens. Well, a-a-and I'm sure quite a lot of competing priorities. At your scale- Absolutely… There's just so much going on. There's- Sure… 100 things you could do.

Mahendra Sekaran: Yeah, yeah, yeah, yeah. Exactly. Like, particularly with the focus on fundamentals, around security, quality, and reliability, you know, we often f— have very little capacity to go invest in new capabilities. So this time around you know one of the PMs in our team built this agent. He trained the agent on pretty much the last several semesters worth of planning intakes. So we exposed our ADO as grounding to the agent. We also kinda trained it on how the previous features were implemented and gave the agent access to all our code repositories And what the agent did was it basically was able to take this intake request and do cost estimates and insert that cost estimate into ADO, and then we could— we just had, like, a clear capacity model. Is it perfect? I would say not, but it had, like, pretty good accuracy. Like, it had about, like, close to eighty percent accuracy so far. So we were able to shrink this planning cycle that took tens or if not sometimes hundreds of people across the company several, several weeks, like six to eight weeks on average to, like, days. And that is a massive, you know, efficiency gain because you can each ask— they-they figure out, you know, if we just do the multiplication you know, it, it basically gives you pretty good context about where, what the opportunity lies, right? So, you know, that's like— and there's tons of examples. Even, like, if I look at how our PMs are working with data. In the past, when we had questions about, hey, where is SMB growth happening? What is the— what is like, you know, we have a sizable number of seats that are doing in SMB, but where did the growth come from? Because M-Microsoft as a company, we don't have a very, very strong field and sales motion for SMBs but we still see growth. We still see a ton of growth through partners through organic growth. So understanding deeper a particular customer base. In the past, when the PMs had to gain insights, they had to work with data engineers or data analysts to work through data, and it'd just take weeks for them to get a response to a simple question of, "Okay, where am I seeing calling plan growth and why?" But now we pretty much have, like, MCP servers deployed pretty much across most of our data assets. So our PMs are able to interact with the data and ask questions and fine-tune the results and are able to come up with very, very deep insights about how our product is resonating with the customers, what can we do more, where are we seeing, like, some lack of adoption. So we can actually then use that as input into our prioritization framework.

Tom Arbuthnot: Yeah, it really helps you to be very data-driven. Also, I, I find the AI helps you be more objective. Like, I think everybody's got thoughts and opinions, right? You're in the space, but actually, what does the data tell me? And asking the AI to kind of challenge me. Like, am I picking the right priorities? What would you pick? What would you highlight? That's really interesting.

Mahendra Sekaran: Absolutely.

Tom Arbuthnot: Awesome. So that's on the kinda PM side. What about the actual developer side? I assume they're using a lot of, - They

Mahendra Sekaran: Are.

Tom Arbuthnot: They are… Azure

Mahendra Sekaran: AD or Skype on CLI. It is, it is. Like I think one of the things that, you know, if you, if you can imagine, a lot of our code bases are like seven to ten years old. You know, we started off kinda building this platform as a core backend for Skype consumer product because there was a scale product at that time, and then we supported Skype for Business Online, and then, of course, Teams happened, and we're seeing very, very strong growth and momentum c— with Teams. So a lot of these services are— we have like several, several, I would say like, you know tens or maybe a couple of hundred microservices that together deliver the experience, whether it's messaging, whether it's fan out, how do you do reactions, how do you do presence. So we have so many different microservices. One of the complexities that we want to basically get rid of to improve our agility and agility doesn't just mean delivering features at a massive pace, even agility in improving our fundamentals. It is like we've I would say like across a lot of our code bases one big priority for us over the last six to eight months is in building out full agentic loops. So you can imagine, you know, like where and of course, you have to go do some foundational work, particularly when dealing with a lot of these existing systems that are running out there to make sure your deployment workflows, your validation workflows, your monitoring workflows, as well as your You know, like how you roll out to production, all these things are kinda tied together. And now we have I would say fairly robust agentic loops implemented across several of our services, and our goal is in the next you know, few months to be pretty much done across our entire fleet where we have full agentic loops. So now you can s- imagine how that kinda improves the, the, the software iteration velocity, right? Like, I think the term that we use is we are kind of becoming the software factory that is just constantly getting input through product requirement documents, and it flows through, agent creates the code. We of course wanna have humans in the right steps along the way because the— ultimately, you know, there is an accountability that lies with product ownership where we want humans to have the final say in when to hit the button, go button, or when to say, "Hey, let's validate this some more.", so it, it, it has been like, I, I would say, like extremely rapid progress and you know, it does require intentionality. It doesn't come easy. Like as humans, you know, we our, our— Like change is not something that comes naturally to us. You have to be intentional about it, but the gains that you see from the system kinda almost makes it, like, very addictive, so you keep trying out new things. Like, in fact, I have, like, several agents that work on my behalf and that get, get me prepared for every day, every meeting. So it's almost— I feel like I have this cheat sheet that I can tap into going into any meeting to just basically synthesize all the conversations around it, right? So it's, it's been, like, I would say phenomenal maybe. It's like, I don't know. Sometimes I'm, I'm always cautious about using superlatives, but it has been super, super impactful in terms of the transition that we're going through.

Tom Arbuthnot: Yeah. I think if you look at, like, e- e- even if you're quite flat and objective about how it's impacted development, like, it, it, it clearly measurably has massively impacted, and I still feel like we're at the early stage of that for knowledge work. And obviously you and other people in IT are probably slightly ahead on using it with a knowledge work and management perspective. But I think that same thing we've seen in development will push through to more and more knowledge workers where we're using agents and we're using loops. And, and the PM role is really interesting because that's not a developer developer, but actually you see them using the same tools to do their work

Mahendra Sekaran: I think so. Maybe I'll just hit on two points since you mentioned knowledge work, because there was a framing that Satya used recently which resonated extremely strongly for me. So, you know, if you can think about it, Microsoft as a company has thrived on building products that enable knowledge work. We're transitioning to this world where we are, you know, enabling platforms and capabilities so agents can do knowledge work. So it's a little bit of a mind shift in terms of, you know, how we think about products, how we think of product capabilities. It's not just providing the tools to humans, it's about providing platforms and data and models and infrastructure, so agents can go do knowledge work on your behalf, right? It's— And then, and the other thing that I would say also is, like, you kinda like drew this distinction between PM and developer. I, you know, like, I think this, this is, like, I think a lot of, lot of folks in the industry have a similar opinion. I can actually see the lines blurring already between what a product manager does and what a developer does. So eventually, I think, you know, there will be a group of builders who are building products and capabilities to serve our customers and you know, like, I think that's, that's— Like, I think the traditional role definitions, I suspect, will continue to blur. But there'll still be this kinda center of excellence for each of these disciplines. You know, like, there's still, you know, even if you have a builder, data scientists bring certain unique skills to the table, and that specialty will— they will leverage that specialty. But, like, increasingly, the, the, the boundaries between these functions and roles and disciplines will, will start to kinda blur and blend together.

Tom Arbuthnot: Yeah, I think I see that too. We, we, we do. We have a change product in, in-house and our developer, like, is increasingly working with the PM more, and it's like actually the PM will do a even do a draft PR and be like, "Here's what I think based on Copilot CLI and using Fable or Opus," whatever the model of the week is. And, and it was stuff that a, a PM would have never done before, but now they have, with developer support, they can go a lot further. Nice. And, and talking of knowledge workers, Work IQ is, like, a big part of the conversation, and I think meetings and calling and transcripts, that— unlocking that knowledge to make it accessible to both humans and agents is, is a really big part of the puzzle. Can you talk to us a little bit about that?

Mahendra Sekaran: Yeah, that's, you know a very good framing, I would say. Like, you know, like if you think about Teams so we are evolving Teams to be that system of work where you know, where pretty much all your collaboration, your calls, meetings, chats are all happening in the Teams canvas. We're evolving Teams to be this place where you can work with agents as well And the way I think about it is, you know, Teams is where a, a lot of the work happens with all the integrations and platform integrations we have with all the other three P bot plugins and agent plugins. But Work IQ is what basically takes all this data and makes it intelligible to Copilot and agents. So, you know, like it's, it's, it's basically a, a, like a platform layer on top of all the enterprise knowledge, whether it's documents, whether it's emails, whether it's chats, whether it's meetings whether it's call transcripts, and making that knowledge accessible to agents is what Work IQ does. So you can think about, you know, like, you know, of course we have… If you think about the stack that we have at Microsoft we call it like the frontier transformation stack. You have kind of like the apps at the top. You know, you have Copilot, Coworker, Scout, and increasingly we will start kind of bringing these app experiences together so the user is not burdened by decide— to make the choice of do I use Researcher, do I use Cowork or do I use Scout? And the app will make an intelligent choice for you. And then you have all the data that makes you know, that's available through Work IQ. Even your Dynamics data from Dataverse will be available through Work IQ. So you have Work IQ, you have Fabric IQ where all the enterprise data kind of feeds into Work IQ, which is your API layer. And then you have your governance layer, which is Agent 365. And again, as IT admins you know, you have to make sure that you have the right policies and data access models when you deploy agents. The ability to give agents an identity, be able to manage the agents, be able to decide what is the scope of tasks they can perform. Again, the mental model is you treat agents just like you treat humans, right? And then you have the models. One of the things that we feel really good about is the choice that we provide to our customers in terms of what the preferred models are. Whether it's, you know whether it's the OpenAI models, whether it's Anthropic models, whether that's our in-house models, we'll give customers the choice. You know, we'll also provide an automatic capability, so based on cost, efficiency, performance we will make the right model choice on the user's behalf. And then you have the kind of infrastructure in the bottommost layer. So that is kind of how we think about the overall frontier transformation stack So Work IQ plays a very pivotal role because every customer has a need for specific agents to do a particular job. Like, I mean, like if I'm a UBS, I have a wealth manager agent. The roles and functions of that agent will be very different from an agent that is doing risk assessment for, you know, like a loan or any financial transaction. So having— Setting up the guardrails and having different agents with different skills, which are grounded in different parts of the knowledge and have given them the n- access through Work IQ is, like a huge unlock. And we are seeing like very, very great results so far. You know, when you use Work IQ, we are seeing like a massive efficiency in terms of token consumption. We are seeing like a massive efficiency in terms of performance, the latency in terms of response.

Tom Arbuthnot: And that's 'cause it's go- it's already done some of the kind of- Right processing and semantic layer stuff. So rather than going getting raw stuff from APIs and Graph and different services, it's, it's actually ready to go to some extent for the AI.

Mahendra Sekaran: Exactly. And exactly, you hit on the, hit on one of the most important points, because y- you could basically take all this data, because this is our customer's data. They could run it through their own models and do their grounding and indexing. But then, yes, you have to go through all those processes. But the additional concern that additional thing that customers have to keep in mind is when they extract the data out, then the, the, one of the things they trust Microsoft for is the data governance, compliance, retention, all that stuff. Now, you have to kinda build that entire layer wherever you're extracting the data to, which is a pretty, pretty onerous task, and it's fraught with risk. So we think about Work IQ as that unifying layer which has like, you know, retrieve APIs, the right grounding APIs, the right context. So it is— Since the data is already semantically indexed and grounded, it is able to serve the most relevant information to the agents

Tom Arbuthnot: Yeah, I think people miss sometimes that there's Work IQ is obviously used in the end user-facing product, so you're in Copilot, it can get Work IQ, but it does have an API that is accessible to customers as well, so they can build line of business agents that tap into Work IQ, respecting the right RBAC and all that good stuff.

Mahendra Sekaran: Absolutely.

Tom Arbuthnot: Nice. And you mentioned Harness is a multi-model. I feel like this is a big part of Microsoft's story at the moment, 'cause obviously the other frontier houses are, they have their model, and then they have their Harness. Microsoft have access to multiple models, obviously the OpenAI relationship, recently Anthropic, but pretty much every model is in Foundry, including the open weights models. How are you thinking about kind of models versus Harness?

Mahendra Sekaran: I think, I think the, the way, you know, at least I think about it is Microsoft kinda provides that intelligence and governance layer on top of the models. The models, you know, like I think in Azure, I don't know what the latest number is, it's like thousands of models are accessible- Yeah to customers if they wanna build anything. I don't know the, I think it was like 11,000 if I, if my memory serves me right. So it's, it's a pretty mind-boggling number in terms of the number of models that customers have access to. We want to make sure that customers basically are able to use the data in a very model agno- agnostic way and give them the choice of models. So, you know, for some tasks, like because, like, I think one of the things that comes up often in my customer conversations is there's concerns about consumption and, you know, token consumption and the cost of the token consumption. And so being able to provide customers the choice so they have, they use the right model for the right task at the right price and performance point is something that we feel good about. There's, you know, of course our Microsoft models will continue to evolve. There'll be a lot of first party places where we will leverage it because it

Tom Arbuthnot: Yeah, they, they're coming into, into the first party product now, aren't they? So things like I heard Teams Transcription is using some of them, other, and- Yeah, yeah some bits in PowerPoint. So that's interesting, 'cause you're mixing in where it makes sense your own models and the, the other models that you use.

Mahendra Sekaran: Correct, correct. And I, I think, I think even if you think about the geopolitical climate of today's world, I mean, there is concerns in terms of, hey you know, like am I, am I gonna be stuck with mo- one model? What if that model gets regulated or something which denies me access to it? So again, that's another place where I believe that having this multi-model approach, not multi-modal, because we both work in communications, but multi-model approach gives our customers more flexibility and choice.

Tom Arbuthnot: Awesome. Thanks for sharing that. It's good to get your perspective on it. And I guess, Mahendra, lastly, it'd be good to understand as much as you can share, what are you most excited about for kind of this year and this semester? Where's your kind of focus?

Mahendra Sekaran: I'll you know, there's several things I'm interested or super excited about. One is, like, I think we talked about it briefly earlier, Tom, like, about reimagining how we get a job done is both exciting and challenging. It is like, I can tell you that even for me, like, it was— it took a lot of intentionality because there's, there's so much access to information, but just getting into this habit of, you know leveraging agents as assistants for me or getting jobs done for me has been a meaningful change. And the efficiency gain that we get like, where all of us can become, like, 10X PMs or 10X leaders or 10X engineers over the next six months is something I'm super excited about. Not just for the sake of efficiency. It is not like, you know I'm just gonna be going and playing golf from the efficiency gain that I get. It is I can do more for our customers. Like, because we always— If I look at our backlog, there's always so many things that our customers ask us to do, but we just don't have the capacity to do. So being able to kinda churn out very meaningful experiences to our customers with very strong fundamentals at a much higher velocity gives me a ton of excitement. The second thing I would say is you know, I-if I think about our priorities, Fundamentals remains one of our top priority, like protecting our core. What can we do? Like even Teams it continues to grow. We still continue to get a trickle of feature requests about, hey… But how do we kinda transform these products to be true agent canvases and, and

Tom Arbuthnot: The fundamentals is really interesting, isn't it? 'Cause it's one of those areas where AI is really good, because engineers typically want to be working on the new thing, and actually AI will be diligent on whatever you put it on. So if you're like, "Just look over this code base again and again and again," trying to find issues, trying to like, like… That's a real strong spot for the models to do a lot of you know, foundations work.

Mahendra Sekaran: Absolutely. Absolutely. Like, and like even security. Like I know, like if you look at the number of you know, just of course, like with the technology, just like good people benefit from it, the threat actors also benefit from technology, so the amount of threat vectors that we have continues to climb. So there, like our security teams, like our, like the IC3 security team, for example, is able to go change a bunch of code bases. In the past where they used to just fan out the work to all the service owners, they're able to leverage agents to both do it on behalf of other teams. So absolutely. Absolutely. So, so, you know, like I think protecting the core is, remains a priority, whether it's, whether it's Teams, whether it's Dynamics 365, because my team partners very closely with the Dynamics team across Contact Center, across Copilot Studio. That, that is also something I'm super excited about. And then part of that is, you know, since I mentioned Dynamics we recently announced Teams' phone agent in Frontier Public. The opportunity that voice agents can play in unlocking kinda agentic workflows for companies is something, you, I, you know, like I'm, I'm super excited to take to GA and learn from our customers there. Because I think that- I think that's, I think

Tom Arbuthnot: That's one of our most exciting spaces in the Teams voice AI space at the moment now, having those, having the native capabilities in Teams. But it's also extensible with the Unify model, so you can go a long way with Copilot Studio or with- Correct… I know some of the ISVs are working through certification as well.

Mahendra Sekaran: Correct, correct. And this is like, you know, like if you think about it, there's so many industries, whether it's retail, whether it's banking, whether it's healthcare, where the phone, you know, we— where you and I have been in the phone industry for a long time. You know, phone remains the lifeline through which their customers connect with them. So being able to have an agent, you know, front that workflow and not just answer the call and satisfy the customer, but the more exciting thing is typically once the call is done, there's a bunch of activities that happen, whether it's upgrading a CRM. Being able to build agentic workflows that cut across the business is the opportunity unlock here. And this does require, you know, agents that are pretty flexible, that can learn. But like the— you can, you can, you know, d— you, you can imagine the efficiency gain you get by actually kind of building a full agentic workflow for like, say, a UBS or Morgan Stanley or Kaiser or Mayo Clinic. Pretty much any of these customers you know, will benefit tremendously. And we are seeing a ton of excitement. Like, I think this is also an area where we see a massive opportunity for partners to play a role in terms of guiding customers through that agent building process, agent validation process. How do you create the right eval framework so that you know the agents are doing the right job? How are they performing? You know, that is some— that's something that we wanna give as much templates and other kinda blueprints from Microsoft, but I think there's still a massive opportunity for the ecosystem around Microsoft to play a role in driving that frontier transformation within our customers.

Tom Arbuthnot: Yeah, I th-I think there's a huge opportunity there 'cause th-you see the motion of the kind of the forward deploy engineers and customers needing support. And as you say, it's not a one-time thing with AI and voice. There's a continuous improvement. Eval models change, business needs change, so there's definitely a great opportunity there.

Mahendra Sekaran: Absolutely. Absolutely. The last thing I would say is like, you know doing, doing fundamentals right and then kinda driving growth it kinda gives us that permission to continue to accelerate on our core you know, AI products, whether it's you know, the work we're doing to make Copilot better, the work we are doing to make Teams better, and there's a ton of work happening in Teams with Channel Agent and other agentic workloads like Facilitator and Teams Phone Agent. We're also working very closely with the Dynamics team so that as you build kind of this contact center cape— Like the agents, like I think o-one thing that I feel you know, the success rate that we're seeing is they are— they're providing almost like a, you know, we expect to provide like a seventy to eighty percent deflection. It means that calls traditionally needed humans, an agent is able to fulfill that in a very meaningful way and, and, and, you know, like at a, at a very, with a very high success rate So how this kind of bridges with customer builds a agent to front a Teams phone number the same agent can be fine-tuned to be f- put in front of a contact center as well. So how these worlds come together is gonna be pretty interesting because even for contact centers, there's a workflow behind it. So our Dynamics team is thinking about that. So, you know, there's, there's… I, I can keep going on, but, like, it's gonna be

Tom Arbuthnot: Yeah. It's gonna be a busy year, basically, I think is the… Yeah. I, I, I completely agree with some of these areas, particularly contact center. I think that area is still ripe for disruption, and we've seen the voice models get so much better in the last six months, and we know more is coming. So yeah, the, the deflection rate and also the insights into what's happening in the… Like you say, the, the, the, the eval frameworks around that stuff, that's got me really excited as well.

Mahendra Sekaran: You know, like, I, I think for me, just in closing, what I would say is I feel that Microsoft's advantage is AI is not sitting just kinda like as a outside thing. It is being in the flow of work whether that's Teams, whether that's Outlook, whether that's all your Office products, whether it's Dynamics. And being exposed a lot of the knowledge and intelligence being exposed through Work IQ and governed through Agent 365 with access to, you know, a model that anybody customers choose running on kinda like global scale Azure infrastructure is the opportunity for us. So we are super excited about the role that we can play in driving transformation within our customers and counting on, you know, advocacy from you and support from you and the partner ecosystem in being part of this transformation journey.

Tom Arbuthnot: Awesome. Well, Mahendra, thanks so much for taking the time to catch up. Always exciting to catch up at this time of year, and I'm sure we'll hopefully catch up when later in the year when we're halfway through and see what's landed and what's coming next.

Mahendra Sekaran: Absolutely. Absolutely. Thanks for the time, Tom. Appreciate it.

Tom Arbuthnot: Thank you.