Microsoft AI Futurist Marco Casalaina on the New Innovation Factory

As software and data take center stage in the modern enterprise, the combination of purposeful artificial intelligence adoption, skilled teams and disciplined processes is becoming a catalyst for transformation. In this episode, WWT CTO Mike Taylor and Microsoft AI Futurist Marco Casalaina provide real-world examples of how organizations are rethinking data foundations and embedding AI into development and operations to unlock new levels of speed, agility and innovation.

Editor's Note: This special episode of the AI Proving Ground Podcast was recorded during WWT's Business Innovation Summit, which took place at the PGA TOUR's World Wide Technology Championship in November 2025. 

The AI Proving Ground Podcast leverages the deep AI technical and business expertise from within World Wide Technology's one-of-a-kind AI Proving Ground, which provides unrivaled access to the world's leading AI technologies. This unique lab environment accelerates your ability to learn about, test, train and implement AI solutions.

Learn more about WWT's AI Proving Ground.

The AI Proving Ground is a composable lab environment that features the latest high-performance infrastructure and reference architectures from the world's leading AI companies, such as NVIDIA, Cisco, Dell, F5, AMD, Intel and others.

Developed within our Advanced Technology Center (ATC), this one-of-a-kind lab environment empowers IT teams to evaluate and test AI infrastructure, software and solutions for efficacy, scalability and flexibility — all under one roof. The AI Proving Ground provides visibility into data flows across the entire development pipeline, enabling more informed decision-making while safeguarding production environments.

Chapters

• 0:00 AI Shifts To Agentic Teams
• 1:36 Meet Microsoft’s AI Futurist
• 2:35 Natural Language Workflows For Everyone
• 4:01 Organizing Work For Multi‑Agent Systems
• 6:00 Scaling And Evaluating Agents At Microsoft
• 8:45 From Digital Agents To Robotics And Edge

Transcript

SPEAKER_00: 0:00

From Worldwide Technology, this is the AI Proving Ground Podcast. AI is moving beyond prompts and co-pilots, and it's becoming orchestrated, agentic, a system of intelligent teammates that can act, coordinate, and accelerate work across every corner of your business. And the breakthroughs coming out of companies like Microsoft are pushing that shift from the edges to the center of modern business. We're watching AI evolve from a developer's tool into an everyday capability. Natural language workflows, teams of agents collaborating in parallel, digital systems that don't just respond, they anticipate, decide, and execute. And as you'll hear from Microsoft AI futurist Marco Casalena and WWT CTO Mike Taylor, these tools move from single assistants to teams of AI agents. That's forcing enterprises to rethink the very architecture of work, how tasks are organized, how decisions get made, and how humans manage a digital workforce that operates at a speed we've never seen before. This is the future, and it's arriving faster than any of us expected. So let's jump in.

SPEAKER_02: 1:14

Thanks for joining us. Thank you very much for joining us. Let's have a seat. All right. Let's have a seat. So my first question, and this was probably one I asked you the first time we met. It's like the coolest. Can we put it back up real quick? It's like the coolest freaking job title ever. Yeah, my job title. I do have an unusual job.

SPEAKER_01: 1:31

Yeah, well, tell tell me, what does what does that mean? What does that mean? Futurist. Well, I'm I'm I am VP products of Core AI. Uh, but yes, my other title, my Silicon Valley title, and I come from Silicon Valley, is AI Futurist. And concretely, what that means, actually, is that I get to be the first person to play with anything new all the time. And it's true. I mean, it's crazy. You would not believe what I was doing this weekend. It was nuts. Tell me about it. So, you know, one of the things that is going to happen, and I think, you know, Nate actually referred to it also, is that, you know, we think about these coding assistants and these these advancements, and we'll talk about those things in a minute, they have a tendency to expose themselves first in the world of code and the developer's world, because that's where we are. You know, we we build for ourselves. But some of the stuff I was playing with this weekend that I didn't even tell you about earlier is um is really aimed at the business user. And so, you know, one of the things that I can do now that I just did this weekend is in M365 Copilot, I can go in there now and just like even the janitor can do what I'm about to say. I could say, uh, if Mike Taylor sends me an email and he's requesting a meeting at a given time, then go look at my calendar, check it out, make sure it's open. If it is, send him an invite. If it's not, then you know, suggest another time.

SPEAKER_02: 2:44

What if you don't want to meet with Mike Taylor?

SPEAKER_01: 2:46

Well, what is then I can make a workflow that says, if Mike Taylor sends me an email, I can tell him to go pound sand.

SPEAKER_02: 2:51

Yeah, just ask just asking for 200 or so friends out in the audience.

SPEAKER_01: 2:55

Yeah, there. Yeah. But this is not a coding assistant anymore. This is actually becoming more and more accessible to everybody. Everybody can use this thing. It's a natural language. You can see it grow the little workflow, a little tree that comes out where it makes a little tree. It's like, okay, you know, email comes from Mike Taylor. What do I do? I go check the calendar and that kind of thing. So it's it's moving quite rapidly into the realm of the everyday person.

SPEAKER_02: 3:21

So do you do you uh, you know, kind of from what what we just covered and what Nate covered, this idea that, you know, the developer is one of the best personas served today in terms of true agentic assistance. What patterns do you see in that that that take us to other roles or functions throughout the enterprise? How do you and Microsoft think about that?

SPEAKER_01: 3:39

Well, yeah, I mean, and and you know, to be fair, I mean, I'm gonna be talking about what Microsoft is doing, but also what I'm talking about, everybody's doing this, right? We're doing it, Google's doing it. Yeah. And I I get to use everything, right? I use Claude Code, I use Gemini CLI, I use everything. Uh so I'm not limited. But what is happening right now, and other things I was doing this weekend was I was using a team of agents. And this is going to become a thing. Now, you know, most people, by this point, most of you here in this room have some number of agents in production today. Let's say double-digit number of agents. Some of you may be, you know, uh a couple of hundred. But the the concept of a team of agents is a little bit abstract right now. Well, in the world of coding, uh, there exists already this notion of a team of agents. You can provision one in GitHub Copilot, you can make one in Claude Code. And that means that you have multiple agents doing different things at the same time. When that happens, the challenge is you need an organizational principle. And it there's a little bit of a mental load there. Now, fortunately, in the world of development, we already have an organizational principle that works, and it's the notion of bugs or issues in GitHub. So these are a logical separation of jobs. This bug is your bug, this bug is his bug, and so on. And so I can do that. I can make a whole bunch of bugs and simultaneously assign them to a bunch of agents which will go work on each one at the same time, and then I become the bottleneck. That's the challenge. So you have to think about the organizational principle of how do I organize the work among these agents, but then you have to think about all right, well, now how do I review that work? How do I, as the manager of this team of agents, how do I make sure that they did it right? So when I do this coding, and I do this all the time, where I will, in fact, assign four different bugs to four different agents at the same time. I can make more. I can I don't even know what the limit is, to be honest, but I can make more and I don't make more than four because I become the limiting factor. They start throwing the stuff back at me fast and furious, and I have to review their work. I have to check for sometimes they conflict with each other, they code in the same place and they make a merge conflict. I have to deal with that. And I just have to test it and make sure that they actually did what I told them to do. Yeah. Uh and so that these are the challenges that you're gonna find now as you all will start to get over the next two years or so into this world of teams of agents.

SPEAKER_02: 6:00

Yeah. And I think, you know, just for the audience and connecting to some of the topics we we went through yesterday, you know, we think about that uh in our business, you know, both from uh an offensive and defensive perspective within cyber. We think about it within the productivity of what we're trying to drive within the organization, and certainly in terms of how we evolve or think about evolving our business models, those those agents um, you know, and I'd say more bought into this today than even last year when we were talking about it. The idea that an employee, you, a human being, is going to have, you know, 10, 20 or more, maybe doing different things, you know, for you, is a we believe strongly that's a reality. You know, then it introduces, okay, well, that's going to require a lot of imagination from and and thinking from everybody of how do we redo or rethink how we do work. Right. And I would say even within the coding assistance space, we cover the technology, we talk about uh the impact that it can have on businesses and companies, but there's a there's a human and and and you know, there's a people and process element of this that's critically important to go with just this is the best tool, that's the best tool here uh for those different use cases. So as you think about organizations or what you're seeing within Microsoft at scale, how how do you all handle that from an from an Android? You're you're kind of you know, you're tinkering with the stuff on the front end, but at scale, what are some of the things you see Microsoft doing with it?

SPEAKER_01: 7:22

Well, certainly, I mean, one of the key things that we have to do to scale our tools, and we do have some of the most popular tools in M365 Copilot, GitHub Copilot that are used in the world today. Uh, you know, and and I gotta say, it's unsexy, but the evaluation stuff on the back end. So we are exposing these agents to the world. And we have to evaluate that these things actually work, or at least mostly work. And so we uh have invested a lot in uh making sure that we can evaluate these tools at scale. You know, you kind of have to do, for those of you who are familiar with development, I mean, we have always done what we call unit testing, this kind of automated testing that tests every time you check something in, make sure you didn't break anything. Well, now we have to do this kind of unit testing for AI, for something that is inherently non-deterministic, that gives you a different answer every time or does a different thing every time. And that becomes increasingly important now, especially now, that we're moving past the era where just agents answer questions. I mean, there's a lot of question-answering agents out there, but now we're getting to a point where they're starting to do things. You know, they will file a ticket in ServiceNow or they'll go take some action in SAP. And as we get there, it becomes increasingly important that these things act correctly.

SPEAKER_02: 8:37

Yeah. And and we've talked so far really about um uh you know, digitizing kind of you know, digital or non-physical things. Uh robotics is a is a big trend out there, right? You know, and in the industrial base, you all understand this very, very well. Use cases that we're seeing or contemplation into, not just autonomous vehicles, but but beyond industrial manufacturing use cases, uh, a little bit maybe as these agents from go from sort of digital assets to physical assets. How do you think about that? I got a friend up here.

SPEAKER_01: 9:09

I mean, I think you're gonna see a lot of the same evolution that we saw in AI. You know, I've been doing AI for like 15 years. I mean, it's hard to tell how old I am because it's not a gray hair on my head. Uh but I've been doing this for a while. And, you know, the evolution of AI has been we started with these purpose-specific models. So these things that do things like OCR, they read documents and, you know, language models, stuff like that. And we move towards these general purpose foundation models that we all know and love today. Now, in robotics, we are in the purpose-built part of that story still. So the other day I was just talking to the exec team at John Deere. And John Deere has had a lot of success lately with their autonomous tractors and autonomous harvesters. So they have these things that will like basically crawl through a farm and pick all the corn and all that kind of stuff without the farmer even having to be in it. The farmer could be standing alongside it, just watching it go. Yeah. Uh, but that's a very purpose-built kind of a robot. Now, I mean, at Microsoft, we do have a robotics lab, and they are putting these things together. So, what is happening now in the world of large language models, we are starting to see the rise of general purpose agents. Microsoft 365 Copilot is one of them, Perplexity is another one, Menace is another one. There are a set of them. So, you start to see these general purpose agents coming on the scene. There will be an age, although I don't think it's that close to it. Yeah, what's the time? Yeah, when you think about it. It's like three to five years. Yeah. I mean, y'all, I don't know how many of you saw that OneX is a neo-robot that they just launched. It's uh kind of it's all decked out and felt, it's like five foot six, and it can kind of take the dishes out of your dishwasher and put them in your cabinet and that kind of thing. But it's still super nascent. And I don't think that that uh neo robot will be that reliable yet. But yeah, over the next three to five years, we probably will be entering a world where we put the reasoning capabilities of the agent orchestrators that we have today for large language models together with robotics, the physical capabilities that that robots have, and we create more general purpose robots towards a future like Bicentennial Man. If you haven't seen Bicentennial Man, the movie Bicentennial Man with Robin Williams in like 1997, it's actually probably how this is going to go.

SPEAKER_02: 11:19

Yeah, it's amazing. I mean, some of these move, these films, you know, one, they set a high standard for what we all have to do, you know, in terms of imagination. But um, you know, it it does feel like the momentum around robotics. And I'd be interested in how you think about the impact of model development on robotics or physical AI use cases and some of the advancements that we've seen in world models and other things that are assisting with that training. Digital, uh uh digital twins has been a big topic in and amongst this audience as well. So your perspectives in that area.

SPEAKER_01: 11:47

Yeah, well, I mean, one of the things that we are moving towards now, you know, we started with large language models. Today, the term large language model is kind of a misnomer because the fact is that for most large language models today, you can feed in at least language and images. There are other types that we now call omni models, where you can feed in language, images, and voice, like audio. And so we're moving towards this realm where these models can take many more modalities as input and can output many more modalities. Those modalities can include robotic movements or sensor data in, robotic movements out, and those kinds of things. So I anticipate that what will happen over the next few years, like there are standalone action models, there are standalone like world understanding models. And the challenge is they only get a piece of the puzzle. So what I anticipate will happen over the next few years is that we'll start to see omni-models with greater modality. And those modalities will include the ones that are used by robots. Yeah.

SPEAKER_02: 12:42

No, very good. And uh, you know, one thing um uh I wanted to, you know, uh throw out there. And then if if we have any questions, we'll have time for one or two. Um, you know, as I look at the the evolution of the models, the the development of these, I think inference and edge use cases. So today we've talked a lot about these gigawatt data centers and you know, big, big, uh, big facilities that are supplying the commute, the compute and ultimately the tokens that we're all using. As robotics will be one influence, but I think there are a number of other industrial and enterprise uh use cases out there. The distribution of these workloads out to the edge and considerations or thoughts you have on what we all ought to be thinking about.

SPEAKER_01: 13:23

Yeah, I mean, this is the thing. As part of my job, I do a lot of work with the space agencies and I hang out with the NASA folks a lot. And, you know, on their spacecraft, they're looking to put generative AI literally onto spacecraft, but it has to be done on the edge because there's no clouds in space. And so uh, you know, we're we're working with them on that, and so is Google for that matter. They may have more data.

SPEAKER_02: 13:41

Yeah, it sounds like they're gonna do a spaceship thing, I heard.

SPEAKER_01: 13:44

Yeah, I mean, we're you know, we're we're we're trying to do that, but I mean, in space, I mean, power is limited and your processing capability is limited. So, still, as it stands right now, small language models, the kind that run on the edge, are still very nascent, not super capable yet. I mean, you could try them. We we have one ourselves called Five 4. Uh, and so you could try that yourself. It'll run on most of your laptops and some of your phones, even. But I mean, relative to the the really frontier models like GBD-5, eh, you know, not as good. Yeah, yeah. Not yet.

SPEAKER_00: 14:12

No, very good, very good. Okay, what today's conversation makes clear is this the future of AI won't be defined by any single model, tool, or even a breakthrough, but how we learn to work alongside a new class of digital teammates: agentic systems, multi-agent workflows, natural language automation. These aren't experiments anymore. They're the early signals of a workplace where humans stop wrestling with complexity and start directing intelligence. This episode of the AI Proving Ground Podcast was co-produced by Nas Baker, Kara Kuhn, and Diane Swank. Our audio video engineer is John Knoblok. My name is Brian Felt, and we'll see you next time.