Mahendra Sekaran: We have this one customer voice is, which is that in all our M365 products, when you do, submit feedback, it, it comes in and you can imagine when there's like, you know, hundreds of millions of users using the product and submitting feedback, having a human go through the data and come up with meaningful insights is, how it's pretty time consuming task. 

Tom Arbuthnot: Yeah.  

Mahendra Sekaran: And now we have agents that mine the data for us and gives us the themes that we need to go act on. 

Tom Arbuthnot: Welcome back to the Teams Insider Podcast. This week we have a great conversation with Mahendra. He is a Corporate Vice President of Microsoft Leading the core communications and collaboration platform, or IC3 you might have heard of. This is the backend infrastructure that supports Teams and calling and Teams Phone and the dynamics contact center. 

Really great insights into what's going on with IC3, how AI is impacting in terms of features, but also in terms of development and operations and where we can expect AI features in IC3 and comms and collaborations through 2026. Many thanks to Mahendra for jumping on the podcast. Always a pleasure to have him on, and also many thanks to AudioCodes to are the sponsor of this podcast. 

Really appreciate all their support. On with the show. Hi everybody. Welcome back to the podcast. Really excited to have Mahendra on the podcast. We've had him on a couple of times before and it's, 2026, a new year, so really excited to get his perspective. everything that's going on with the IC3. 

Team Mahendra, welcome back to the pod.  

Mahendra Sekaran: Yeah. Good. Good morning, Tom. great to be back. and thank you for your continued advocacy and championing of Teams in a variety of forums. it's always a pleasure to be on your podcast.  

Tom Arbuthnot: Awesome. I appreciate it. So, just to recap quickly for people that may not know, I feel like everybody in our space would know, but, like your role and what the IC3 team do, can you just give us a quick recap on that? 

Mahendra Sekaran: Yeah, so I, I'm a CVP here at Microsoft and I lead, the product management and data science functions for all of IC3 and IC3. For people who are not familiar with it, it stands for Intelligent Conversations and Communication Cloud. And you should think about this, IC3 as the core, comms and collab platform for, the entirety of Microsoft. 

so we build all the media technologies, we build all the messaging services, calling services. PST and connectivity. And, so we are the core, comms hub for the company. And, you know, we are kind of the backbone for Teams, as you may know, but like we also support a variety for the products like dynamics and, any multimodal interaction that happens in products across Microsoft, is something that, we contribute to, like, whether it's Copilot or Copilot Studio. 

yeah.  

Tom Arbuthnot: Yeah. So when we see all those big numbers from, from Teams like the 320 million monthly active or the 26 million PSTN users, now that's your team's infrastructure on the backend kind of, holding up those sessions.  

Mahendra Sekaran: A hundred percent. A hundred percent, yeah.  

Tom Arbuthnot: Awesome. So there's so much I want to get into, but I mean, it feels like we can't not talk about AI being that it's, you know, the biggest thing going on in the industry. 

But I think particularly for your space, it's so interesting where AI and real time are. Coming together and there's so much potential, but also so many challenges with the scale and the compute costs and everything that's going on. maybe you can give us some thoughts about how AI has come into the IC3 world. 

And, again, I just love to hear you riff on it.  

Mahendra Sekaran: Yeah. So, let me first start with maybe like a internal look on how AI has impacted us, at IC3 and at Microsoft. You know, one of the magic that AI unlocks is, . Information at your fingertips. And, like it's, it's, it's, it's amazing how as humans we develop habits. 

And for me, like there are several habits that are built over the past couple of years, which, if somebody takes away, I've, I will feel very crippled. so, you know, starting off my day with, you know, going to Researcher and getting a daily brief, so I'm actually, . Prepared for my meetings, where, the agent is able to go look at previous instances of the meeting, what actions I took. 

so I don't show up. I show up to the meeting. Well, there's something that, you know, like I feel very thankful for. And it's like one of those things that you don't, you know, it's hard to put a number against it in terms of quantification, you know? but like, in terms of how much it has helped my personal productivity, it is just, like, it's like, you know, it's like it saved me hours and hours and hours. 

. Of interactions. Yeah,  

Tom Arbuthnot: it's, it's hard to measure the, like relative performance improvement as well, isn't it? Because you're not Correct. It's not just minutes say, but it's like you and I have the same thing. I use Copilot voice mode quite a lot, so on the way in, in the car, I'll be like, tell me what's going on for the day. 

Yeah. And again, I, I, I find that hugely valuable to just hit my day better.  

Mahendra Sekaran: That's right. That's right. And like I, you know, like, one of the things that, since I lead the product function, you know, we are seeing, you know, okay, if you were to just ask my candid opinion about, hey, have we fully reimagined, reimagined ourselves as, fully kind of AI native, . 

Engineers and product managers. I would say the answer is no. but, if I look at everybody, across the team, they're looking at every single thing they do and looking at ways for how AI can make that more efficient for them. And, so it is like one of those things where, you know, like you have, . 

You know, like these incremental wins that collectively makes us a lot more productive. So, like we are kind of like also leaning towards more, more towards model first product building, particularly when you're building AI experiences. how we understand, the model capabilities and tailor the experiences, to, to build products for our customers is something that we have gotten a lot, lot, lot better at compared to like two years ago. 

And another, operation function that you'll hear a lot of people at Microsoft talk about, that has evolved, which, we are kind of like in the middle of as well, just given. We are the core multimodal platform, for Microsoft is the, Evals and evaluation frameworks. given that these LLMs are inherently probabilistic in nature, . 

You know, like you, you need to have the right mechanisms to evaluate how well the model is performing and be able to continuously verify it and get better.  

Tom Arbuthnot: So you guys have lots of models available to you, right? Don't you? Because Correct. Microsoft are taking a play of, will give options, obviously externally with Microsoft Foundry. 

That's right. But I'm sure internally as well, you've got your own models, you're developing your house, you've got access to all the Open. AI IP. That's right. And seems to be playing a bigger role now as well.  

Mahendra Sekaran: That's right. That's right. That's right. So, because we believe, like, you know, ultimately our, our belief is that models are not gonna be, at least the frontier models are not gonna be the, differentiating moat. 

it is the experiences and the layers that you build on top of it is where the differentiation comes in. so that, you know, like, I think from an Evals perspective, like it's, it's been a lot of learnings, I would say, but, one of our leaders kind calls it as like, crawling through class, or crawling over glass. 

It's painful, but when you reach the destination, it's actually very fulfilling. And, the places where, like, IC3, we are spending a lot of time on Evals. like you, you mentioned the Copilot voice mode. so we are, you know, like we provide the core technology for Copilot voice mode. so you can expect, you know, things like voice isolation, noise suppression, all of this stuff to happen. 

So we do a bunch of Evals, kinda introducing some of these environmental noises. to see how the system is functioning. Similarly, I know we haven't, you know, the industry or the community may be curious on, Hey, what are you doing with, Phone and AI? and, there is a lot that we are working on. 

you know, like I, I, I, I, I cannot share the details right now, but you can expect that. Pretty much, every part of the Phone experience will be reimagined with AI. And, today we do support, you know, Copilot for Phones. So you can do transcriptions, but that's like a, it's almost like a table stakes thing. 

but you can think about capabilities where we are driving. customer engagement, whether it's through call queues or individual call handling, there'll be intelligence wrappers around it. But we are also not just thinking about, like, I don't think about AI as just enhancing the comms experience. 

AI is going to, you know, like, I think evolve the communications experience. So communications. It doesn't become a point in time activity, it becomes part of a workflow. So it's like activating the workflows that proceed or follow conversations or meetings or calls is where we are spending a lot of, you know, thought power on and product building capabilities on. 

So when we get it to our customers, they can clearly see, you know, the, the incremental value they get by, you know, using our products. Because ultimately, you know, if you think about it, like our customers don't buy our products because they like Microsoft or the, you know, they think Microsoft is cool. 

Of course they do. Like Microsoft. Microsoft products are cool, but they are buying our products so they can do their job more effectively. Yeah. so if I'm a researcher, Pfizer. I, I, you know, the technology is just a mechanism for me to get my job done, and we are committed to building the best technology that allows the customers to unleash, the, you know, their fullest capabilities. 

Tom Arbuthnot: Yeah, there's a few things I wanna pick up on. Like, like one of the things is I think AI is clearly massively disruptive and changing for multiple industries, but comms, it feels like is one of the potentially most positively disrupted and potentially the most, easy to adopt. Things like meeting recaps have been so incredibly popular across everybody because it's everything. 

Everybody understands it, it happens automatically. So I'm really excited about. Our space from that point of view. but then you've kind of got the, the, the, I dunno if you call it basic, but like, like a real all, we've all seen the real time voice models. That's really exciting. Yeah. So you mentioned Phone. 

You can see where that comes into Phone. But this idea of. Extracting the data from Sessions and Work IQ and Ethereal Intelligence coming outta your communications, that is the bigger game changer for me. 'cause like voice, an automated attendant that can talk in real time is really impressive. But I do a lot of calls and meetings, the system, knowing what's going on around my whole life because it's in, in those sessions, it's just so potentially powerful. 

Mahendra Sekaran: 100%. 100%. And again, like I Thank you. You touched on agents there like, and, and mean, like, I think. A Agentic AI, is like, you know, I would say that 2026 where we is the year where we are gonna scale AI across the board. Like it's not just Microsoft. I think across our competition, across the industry, you'll see a lot of, scaling of the AI capabilities happening. 

And, you know, the way we imagined Teams is, you know, we had this vision of. Teams kind of being the hub for M365. you know, five years ago there was a narrative, and it, it, it continues to be, but like, I think Teams will evolve to be the place where, not just humans gonna connect with each other, but where agents, are, interact with humans as well. 

and the other thing is like, you know, bringing these workflows and making them so Teams just starts become evolving to become the hub where work gets done and not just where conversations happen.  

Tom Arbuthnot: Interesting. how have you seen the AI capabilities and tooling affect how you guys think about. 

Development and, and product management, that kind of thing. I'd be interested to, for your perspective on how it's changed how your Teams work.  

Mahendra Sekaran: No, that's a great question. And interestingly, we had a leadership offsite yesterday, and it is, it is a big topic of discussion for us, in terms of how we can do more. 

But like, how do you basically, drive the throughput of our engineering Teams by 10 x? Right? I mean, like, you know, how do you make each of our engineers a 10 X engineer?  

Tom Arbuthnot: Yeah.  

Mahendra Sekaran: When you break it down, like particularly when dealing with kind of large scale complex systems that have been built over decades. 

compare that to, you know, a, a a a, a product that you're starting from scratch. I think they have different benefiting points. Like, because if you take one of these large scale systems, the cost of writing code is probably where is a fraction of the time. Relative to where the developer is spending majority of the time, right? 

If I, if, if I take, you know, if I take a pull request that I basically submit and then, validate and then roll it into production. I probably spend maybe 5%, 10% of my time with, in, in coding it, but the majority of the time is kind of going through all the other steps I need to do, getting the right security token, finding the right environment, doing the validations, and then, you know, flight this into a production ring. 

So, so the way we think about it is like, I think from a coding agent perspective, it's, it's, incredible how much, these coding agents have evolved, right? Whether it's, GitHub, Copilots coding agent, or if you look at Claude Code, you know, like it's, it's just like magical to see, what some of these, . 

Yeah, ouris are able to accomplish and you know, like, it, it, it gives me, like, I used to be a developer like 20 years ago. I haven't written quotes since then, but it gives me the confidence to go, you know, try out some new product ideas and start building myself. Right. It's, it's  

Tom Arbuthnot: amazing for. Prototyping. He is like, I'm not a developer either, but I, I used to, you know, write lots of PowerShell scripts and like that was as advanced as I got. 

And s similar thing like I'm back into doing things 'cause I'm like, wow, I can, I can scratch my own niche of this really obscure automation that's I need to do. That's, that's right. It's amazing. And seeing that Claude Code is just over a year old as an example and Open AI's Codex and what, what Microsoft are doing with Copilot as well. 

How far we've come in so little time is, is fascinating.  

Mahendra Sekaran: A hundred percent. A hundred percent. And then to just kind of fully answer your question, like I think there are a variety of aspects where we are seeing the AI tooling come into, play to help us, you know, not just write code and deploy it. 

But also like, operate the code. Hmm. so in the whole area of AI ops, we see tremendous benefits where, you know, we, we have agents that when a, when a, when an incident happens, how it's able to go correlate to other similar incidents, suggest remedial action, and in some cases even be able to take the remedial action with your, with the human confirmation on your behalf. 

Right? I mean, that entire workflow can be automated. So we are, you know, the team has built, . several agents. so we have this on-call engineer agent. We have, like a repair item agent, which is tracking, you know, how people are, treating incidents and the repair items that come from it. Yeah. So there's all these little pockets that are emerging, which, you know, collectively is driving tremendous efficiency. 

But, . The key point is when you're dealing with a large scale system that has been very complex distributed system, you cannot expect, you know, the same kind of throughput that, you know, somebody with a 10 person startup can go do.  

Tom Arbuthnot: Yeah. Your your, that's an interesting insight that I hadn't really thought about. 

Like you say, if the, if the coding. Is the 5% and the operational integration and go live is the 95%. You can optimize the 5% all day. But conversely, there's a lot of values to be had by optimizing the, the, the pipeline and the deployment and the ongoing monitoring operations. But it's a different. More complex problem space in the sense  

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

And you know, like I just inherently as humans, there is also a basic resistance to change in terms of how you get your work done. So people fall back to their, their muscle memory and their usual habits. So it takes a little bit of intentionality as well as, you know, advocacy and, the kind of knowledge sharing and championing within the org. 

To kind of drive that, shift for everybody to be embracing AI as their, you know, assistant in getting their job done.  

Tom Arbuthnot: Yeah, and it's interesting to hear that because Microsoft obviously are a technology company, so some of the most advanced technical people working there, but you still have that cultural friction that is just human. 

That to, I've always done it this way. I have to learn a new way and, and, and AI is. So evolving that even when you don't learn AI and then you're done, you are on now kind of on a, either a roadmap or treadmill, depending which, so you look at it of like, the tools I'm using are gonna be different in six months and different in 12 months. 

Mahendra Sekaran: Exactly. Exactly. And you know, like, I think this is where Satya keeps retraining that like, you know, we've, we've kind of embraced growth mindset as an essential part of our culture for the past decade. But, you know, if you were to, there's no better time to embrace the growth mindset, as it is today. 

Where you need to kinda unlearn some of our habits and learn new ways of, building products. Like even for product management. Like, give you an example, like in the past, you know, my PMs when they had like, data needs, like say that we are seeing this little uptick in, you know, Phone growth, month over month. 

And then if they have, you know, if they better go understand, okay, let's go dig deeper to see which region, what segment are we seeing the growth. Then what's driving the growth? Are they using calling plans or is operator connect hunting? They had to go rely on a set of data engineers and analysts to go do that analysis and can come back to them. 

Now with pretty much all our data being part of a centralized data lake, we have all these self-serve analytics engines where I can just have a conversation with these data agents and derive like. Incredible insights. Like, you know, like oftentimes when I'm in interview with the team, I'll ask them, Hey, what about this? 

They'll say, okay, we'll come back to you in two weeks.  

Tom Arbuthnot: Yeah. Yeah.  

Mahendra Sekaran: And it takes two weeks because they had to go. Yeah. And  

Tom Arbuthnot: during that time you decided you wanna ask a different question anyway, so you're like, oh, okay. Yeah. Great. Now another one.  

Mahendra Sekaran: No, no. I just, you know, go ask the question myself. Or if ask a question, people will just ask the agent and gimme the answer in near real time. 

Yeah. Which is fascinating. Like if you think about just that compression in terms of how PMs function, because. one of the biggest jobs that we as, you know, PMs need to do is, have a deep understanding of how our customers are using a product. And a very, concrete way to understand that is by looking at the data and the telemetry. 

Now you have this extremely powerful tool that makes that information very easily accessible. so for the Curious pm you know, they, it's the, the, the bones are unlimited in terms of what they can accomplish.  

Tom Arbuthnot: Yeah. And that reduction in, our question to answer time is really interesting because it changes, I think, the way you think about, like, I'll fire off half a dozen random questions 'cause who knows what insight will come out and, and actually, like you are, you are in the flow of. 

Getting all kinds different insights. You don't know it, it's very hard. Again, back to the top of our conversation, hard to articulate how much value is there, but that could be immense value if that PM picks out the right feature for the right customers. And it's a successful feature.  

Mahendra Sekaran: Exactly. And, and yeah, actually you touched touch on one very interesting, topic here in terms of picking the right feature for the right customer. 

Another theme that I, I suspect will evolve in the world of AI is this whole personalization aspect. Because you know, like as these, as you build deeper insights about the user. And the product can actually learn from that. And you can think about even the user experience being kind of tailored to that individual user's needs. 

Like, you know, if, if I take Teams as an example, if I'm a heavy calling user, you can think of the UI automatically navigating to the right points and be able to, you know, pre-assess your intent and, get everything kind of done for you. So it's like you are like, it, it reduces your number of clicks by, you know, multiple factors. 

Yeah.  

Tom Arbuthnot: yeah, that's really interesting granular personalization. And it's, it's interesting to me that in this first wave of AI adoption, the popular or the primary interface has actually been text and chat, either at command line and coding or in, you know, in a chat in web based chats. But as you say, there might be a UI paradigm change there where actually the UI is slightly more. 

Dynamic to my use case potentially.  

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

Tom Arbuthnot: That's interesting.  

Mahendra Sekaran: And you know, again, another, another thing just maybe to noodle on Tom, like going back to this question about the discussion we had about how when you ask a question, the turnaround time gets reduced when it comes to data. 

And as leaders, when we ask these questions, we have certain intuition about, you know, what, what's happening, which is what, which is what prompts these questions. Or we have, some instincts about, okay, maybe we should go do these steps and we wanna validate that with data. But as you ask questions now, even the PMs, we'll start asking those questions themselves. 

And when they start asking those questions themselves, they will come up with the one. Once they get the answer to one question, they raise three of the questions, which they will go,  

Tom Arbuthnot: yeah,  

Mahendra Sekaran: big, deeper. So when they come present a view, it'll be a lot more complete and comprehensive view where it's fully informed has. 

Kind of thought through all the different data dimensions and you know, like, I just feel like from that perspective you're shortening a bunch of lifecycle, in a variety of ways, right?  

Tom Arbuthnot: Yeah. Yeah. And that, that takes me to the, back to the real time media and the work IQ in the organizational insights, I think is fascinating. 

Like, if you are a, a department leader or a sales leader or a product leader in an organization, like, you know, take contact center as a. Example that's, you know, we both have a, a, a hand in the conversation there. Like we've always been able to record the data, but we've never been able to really meaningfully process all that data. 

And you, you take a sample and you'd look at the samples, but now our product owner can go to contact center team and say, what are the common themes of questions around our product or service? Right. What were the questions three months ago and what like, like that insight? We've never had the ability, we've had the ability to collect the data. 

We've never had the ability to mine it in such a useful way.  

Mahendra Sekaran: Yeah. Yeah. And, and actually, look, it's, it's great you mentioned that because, you know, within Microsoft, there's several places where we do that today, or we've been doing this for the past couple of years. when, when you have a bunch of customer reported incidents, that, that hit our engineering Teams. 

Our goal is to reduce the number of customer reported incidents and, you know, like, and make sure that, because as feature requests there may be like service, incidents, as part of it. So, one function we invested in from a discipline perspective, maybe three years ago was problem management. 

Where we are able to go through, you know, the, the reams of, customer support requests that come in, identify the themes. Then, you know, figure, okay, this is a documentation issue or this is a licensing issue. This is a, you know, the default configuration seems to be, has to change. So we've identified some of these patterns and have been able to proactively go address them, and we've seen a pretty significant, a very noticeable drop in our, customer support ticket volume when we did that kind of intentionally. 

Right. Similarly, today, like we, we have this one customer voice is, which is that in all our M365 products, when you do, submit feedback, it, it comes in and you can imagine when there's like, you know, hundreds of millions of users using the product and submitting feedback, having a human go through the data and come up with meaningful insights is, how it's pretty time consuming task. 

Tom Arbuthnot: Yeah.  

Mahendra Sekaran: And now we have agents that mine the data for us and gives us the themes that we need to go act on. Right.  

Tom Arbuthnot: So yeah. That, that's such an exciting time to be in product, isn't it? And it's,  

Mahendra Sekaran: yeah.  

Tom Arbuthnot: The, the fa the fa, again, the fast feedback cycle of, theoretically as this stuff comes along, developing faster pipelines, faster feedback loop is faster, it's all accelerating. 

And, that's right. But e equally makes it very competitive, as you said, at the top of the conversation, like everybody is, is moving faster now as well.  

Mahendra Sekaran: Exactly. Exactly. Yeah.  

Tom Arbuthnot: Awesome. Well, Mahendra, thanks so much for taking this time to jump on the pod. Really exciting to get your perspective. I know there's a lot coming this year from, Teams and Phone and IC3 that we can't talk about quite yet. 

but maybe when some of that stuff's dropped, we can come back on the pod and then you can give us some, some insights into some of those developments.  

Mahendra Sekaran: Yeah, for sure. Tom, it's been a pleasure talking to you and, . Look forward to seeing you and you know, all the events and, I'll continue following you on LinkedIn. 

I'm, I always get a lot of good insights from your posts, so keep up the great work.  

Tom Arbuthnot: Awesome. Really appreciate it talking to you, Mahendra. Thanks a lot.  

Mahendra Sekaran: Yeah, take care. Tom.