AI Agent & Copilot Podcast: Donna Sarkar of Microsoft on Moving AI Agents from Experimentation to Production

Hey everybody, welcome back to another episode of the Agent and Copilot Podcast, where I get a chance to sit down with executives over at Microsoft, other leaders in the industry, to talk about the opportunities, the impact, and the outcomes possible with AI today. Um, I am thrilled to be joined by Donna Sarkar today, who is one of our keynoters for the AI Agent and Copilot Summit going on March seventeen to nineteen there at the Torrey Pines Hilton in San Diego. Hey Donna.

Hello. Hey everyone. Thank you, John, for inviting me to this podcast and to the event. I'm very, very excited about this. And I'm excited to meet with so many of you in person. This conference sounds like a thing, not just a conference, but seeing all the pre-day stuff, the postday stuff, the evening stuff, the daytime stuff, I'm like, there's a pickleball tournament. It's going to be fun. It's going to be fun. I'm really, really looking forward to it.

Ah, well, we're so thrilled to have you be a part of it with us, Donna. And um, you know, gang, uh, in Donna's role, she gets an opportunity to work with a whole bunch of different folks inside Microsoft. um as the chief troublemaker of um uh enterprise ai advocacy. And Donna, do me a favor, maybe we can take a step back and up for a second. Maybe you can fill folks in on on what that role means today and what some of the stuff is you're driving.

Okay, so advocacy itself is falls into this category called cloud advocacy. It is also known as developer relations in developer service. But because we don't only do work with pro developers. We work with IT pros, we work with builders, we work with data people, we work with all categories of technical people, right? So think of it more like tech rel, but tech rel doesn't make any sense. So that's why we call it cloud advocacy, even though our work is basically all work is cloud work, right?

So our specific role, my team specific work is focusing on people who work in an enterprise. When I say enterprise, I don't mean five hundred thousand people. I mean an organization of some sort. So we work less with free range devs on the internet. We work more with people who have a job of some sort in a company.

So the areas that my team specifically focuses on are exactly what you think of. So that would be M365, the Power Platform, Copilot Studio, Azure AI Foundry, all of the Azure Core tools, think of the Azure Management Tools, et cetera, et cetera. Basically, everything people nowadays need to build and deploy AI workloads into their business in some way or the other.

So the three areas that my team specifically focuses on are We take the content coming from the product team and we create our own, well, we create the products coming from the product team. And then we create our own content and spin on it. So for example, what we would do is build a really interesting agent for Copilot Studio for something that relates to us. So for me, I would go build an agent to help me manage.

my textile inventory for my fashion line. And I would go out there and demo it on stages all over the world, showing people how you can use Microsoft tools, build on it and deploy it in your company. So you can do something similar. So that is the kind of work we do. We take the product team work and make it more applicable to other people's businesses. So that's the first thing we do. The second thing we do is deeply um we invest in community relations.

So we work with people like all of you to really understand what is your business trying to do this year? What are your goals? Where are you running into roadblocks and where can we help? And sometimes that is, hey, we need this product to be a different way, or we need more information on how to use the product, or actually, like we don't even know which product to use. So a lot of it is enabling the success of your customers.

Because again, we're not sales. We we're not trying to sell you anything. We assume you've already got a thing. Either it's free or you bought it. You may not know exactly how to use it. And that's very much our job. We're not here to sell you anything. I've had customers come up to me and say, how do I buy this? I don't know. I have no sales target, no way to accept money. Um go see your salesperson or buy it off the internet. Do not know. So we really help people use the thing.

And the third thing we do is we make the product better. So when we see a large amount of product feedback saying like, hey, this MCP thing is super confusing for Copat Studio or M365, we go work with the product team to say, we need to improve this workflow because people are extremely confused. So my team specifically creates a lot of content like hands-on workshops.

breakout sessions, step-by-step labs. We have a big uh Copet Studio Agent Academy people can follow along. So we really are the liaison between Microsoft product teams and the communities that we serve. It's awesome, Donna. And one of the things, gang, that that um I was so inspired by with that sort of explanation and overview that Donna was sharing is it makes all of this extremely relatable to us.

Who are putting these things to work every day for our organizations? You know, um, Donna, we we talk a lot about at our organization, we've got one primary mission, and that is to connect what we like to call the four points of light in this community or in this ecosystem.

It's the creators at Microsoft, like yourself, right? It's the partners that are bringing these platforms to market every single day out there through to customers. It's the buyers who are investing in these platforms to transform their businesses, modernize. their workflows and all that sort of thing. And then it's the users that put these things to work every single day inside their companies.

And as we think about the um uh the role, Donna, that you just explained there with advocacy and you think about what we're doing with the AI agent and co-pilot summit, boy, it's a really nice peanut butter and jelly sandwich that we put together there, right? Because We've got not only Donna keynoting an amazing session for us there on Wednesday morning at the event, but we've also got industry accelerators that are going to be coming up on the stage and telling their stories.

of how they are transforming their businesses and putting AI to work every single day to not only accelerate their companies, Donna, but also their industries. Mm-hmm. I'm excited about that. Um, I know one of the things that we've been talking about a lot is what are some of the things I want to do at this event? And I honestly am there to learn more than teach.

because I find events like this, I call this a professional uh a professional people's conference, right? So this is not a necessarily a community event. It's more of a professional development conference. And at these kinds of conferences, I love mixing up with senior people at other companies who've been in this cloud AI drama for the last three years.

And we're now reaching that point where people have done the experiment, they've built the thing or bought the thing, and now they're trying to put it into production in their company and they're running into all kinds of issue, right? All kinds of issue, whether it Hey, we don't know how to do permissions or hey, we don't know when to take it down or hey, who should have it?

Or did we get our data right? There's so many topics that people run into when they actually try to put something into production. And I'm here for that. Right. That is what I'm here for right now because sure I can help you build an AI agent or copilot in five minutes. We've done that demo for you hundreds of times. But let's be clear, the best agent is probably not built in five minutes.

And they're probably not built by a solo person. It's probably your finance agent's probably built by four people in finance along with a lot of help. from other people, also your IT team to figure out when to deploy it, what to deploy to, who should have permissions, who's the agent boss, who's the second agent boss in command. There's a lot of uh process and tooling.

that goes into actually putting agents into production and having them be useful in your business. So I'm here to hear how you are doing that. And if you're having trouble, what can we do from a product Content point of view. And if you are doing it well, I want to learn. I want to learn how did you do that?

Because I'm also training my AI model on what is the best way that companies can benefit from using AI. So it's not some like nonsense hype tool that we all check the box on, but it actually move your business forward. So I'm very, very excited to come learn from you. Uh huh. I've done it. It's uh

Thank you for putting it that way. And um boy, I I think again you can see why I was so um Kind of excited and energized with the potential of having Donna come be a part of this special experience that we're creating here with year two of the AI Agent and Copilot Summit. And um, you know, so I Donna and I were going back and forth, gang, on uh Donna's forty five minute keynote that's gonna be taking place there on uh on Wednesday morning.

And um Donna came back with what I thought was one of the coolest titles for a session that uh that I've seen in a long time. And it's Who's Afraid of Little Old Copilot? And uh Donna shared a few of the learning objectives and those sorts of things that are gonna be a part of the program. But Donna, do me a favor, fill folks in. What can they expect? Okay, so this Who's Afraid of Little Copilot? It's a Taylor Swift reference for those who celebrate. Um

The the gist of it is this, right? Where everybody in the world, everybody, I think someone told me I've had more pressure to adopt AI than I got pressured to like take drugs in high school, right? Like you all are intense. I say we are very intense. And here's the thing though, Mr.

We we tend to overindex on something that is new because when we're going from zero to a hundred, if we land at 50, that's pretty good. So you almost have to over-index to 100 and because 50% of the time it may or may not work. So fine. Um, but I'm at a point now where I am a true believer that there are some scenarios that are very good for AI and a lot of scenarios that are actually quite bad for AI.

So what I want to do is have a very rich conversation with all of you in this keynote about scenarios that businesses are doing that are absolutely right for AI and you should do them tomorrow. And some of them are going to be like, Oh, this helps me in my day-to-day. Some will be like, this helps me in my profession. And some will be this problem used to be impossible and now it is impossible.

So I've got examples of all of these from various sorts of businesses. And I want to share them with you because hopefully you'll get some ideas on things you actually can be investing in. A mistake I'm seeing a lot of companies make. Is that they're trying to replace existing human activity with AI. They're saying, Donna and John get on a call. How can we have AI do this? Okay, where we're gonna have a big robotic arm come and click join and have Avatar Donna talk to Avatar John.

No, that's a terrible solution. Don't do that. That should be a hum a uniquely human activity that we should keep going. Instead, please prioritize all of Donna's work so she has more time to talk to John in the back, right? So we really need to go draw a big old line in the sand and say these should be uniquely human-to-human activities, these should be uniquely AI to human activities, and these should be uniquely AI to AI activities.

Yeah. And I don't think most people know that yet, right? I don't think most people know that. So the entire keynote will be based around this. What I want is for all of you to walk away with a plan and say, you know what? I know what I want to do in my company. Or here's what we have been doing. And I'm going to do a little course correcting, correcting, and try these experiments that this slightly insane person are bringing. I love it. I I think um I think it's it's interesting because

Like with any big transformative thing that gets everybody's attention, right? You know, that World Economic Reforum was just going on in Davos and everything was AI, everything, right? Um and it's cool, right? And you get people getting up talking everyone's gonna have a robot and everyone's gonna have these I get it right. There's there's the hypey sort of stuff that goes on.

One of the things, and I love to hear you use the words uniquely human. That's something I talk about a lot. I actually think there are three things that are uniquely human that that we have as humans. Now it doesn't mean that AI doesn't do parts of this as well, right? Um, but it's curiosity, it's creativity, and it's critical critical thinking. Right. For us to be able to to apply our curiosity, then get creative around what we can actually do and think about and create and and et cetera.

And then think critically about some of the other stuff that you just talked about, Donna, which is making sure the data is right, right? Making sure you've got the right governance built around access controls and those sorts of things in these environments.

So we can create those A to A type of environments, like the agent to agent or AI to AI kind type of environments that then lets us do more of this stuff, right? Which is again uniquely human to like sit down and shoot the breeze and talk about cool stuff. Yeah. And I'll tell you, people are craving craving more human to human. This is why every conference is passed.

Right. Everyone kept telling me, oh, with AI, everyone's just gonna learn everything, you know, through a a copilot of some sort. Copilot's just gonna teach them and they're not gonna need human interaction. I say, friends. We have had more packed conferences in the last three years than we've had in the last 10 years because people are saying, nah, I want more human-to-human connection. I want, I don't want less.

Right. And again, not all people, but a huge majority of people. And the number of people who are very interested in your conference, Don. Um, then I have people sliding my DMs every day, Microsoft people, customers being like, hey, can you get me to this conference? I'm like, No, but go ask John. Why are you asking me? So people are very curious about this because they're saying this feels practical.

Right. This does not seem like AI research hype. This does not seem like AGI. No, this feels like how do I use this in my business to make more money and have more time to do other projects that we've never been able to do in our life. So I'm excited for this event because I saw the lineup. It's all very technical, first of all, and second, very, very practical. And the speaker lineup is off the charts cool. So I'm excited to hang out with all of you as well.

Uh, Don, it's so nice of you to say. And um I think it's it's one of the one of the fun things, you know, like I I did a um a little video for our internal team at our company, Galia. Probably six months ago, nine months ago, something. And um, and I said, I think what we're gonna start to experience is a face to face renaissance.

based on what's happening with AI. Because yes, we need AI to be able to do a lot of these things like like Donna, you were just talking about and the like. But if we can still figure out ways to harness the power and the kind of honor of being together in a room and sharing ideas and sharing perspective and also

Just kind of helping each other out a little bit. Yeah, you know, golly, I did this thing over here and meh, you probably don't want to do that thing, right? You know, like those kinds of conversations, they're just they happen on stages, they happen in hallways, they happen over meals, right? And it's just a it's I don't know. To me, it feels a little bit like a renaissance of bringing people back together to decentralize intelligence.

It's very true. You know, I'm on this big AI tour right now, AI AI tour and AgentCon. And one of the things that I've been hearing from a lot of people, senior executives at Microsoft, is that information is now ubiquitous. You can get any information in the world with a query or an API call, like in one minute, right? Under a minute. You can get any information in the world, but point of view still comes from humans. Sure, there's a thousand ways to build an agent, but should you?

Right. And which kind of agent should you build to solve problems? That is something that I would trust a human who's done the thing rather than an AI with options. We'll tell you exactly what you want to hear too. Right. So I think I think that's why conferences are so useful, John. And I'm so grateful you're doing this.

And I'm very excited for this customer roundtable thing you've got going on because I want to hear customers talk to each other and I want to go and like harass them a little bit and be like, and then what? And what went wrong and what went right and tell me more about this. And let me tell you mine because I built I've built a hundred agents. I've built more than a hundred agents. I've been building agents for three years and I can tell you

Some of them did not go well and I had to take them down. I the number of agents I've had to take down is probably like 50% of the agents I built because I didn't know what I was doing or I had an agent for the wrong thing. And I want to share those with you.

Because I don't want you to go down that path. Please learn from my mistakes, right? Go do the things that I know are gonna work. Because if it's gonna work in Microsoft, because we have 250,000 employees, some number, right? Huge number of employees. Uh, in how many countries? And I know your businesses are all different, but there's something to be learned about us having, you know, drunk our own champagne for the last three years or so. So I'm excited about that too.

What a great way to say it. And I I think it's funny. I was um at a conference uh uh like Q three last year and um uh uh people were asking when we had started creating agents at our organization, right? And uh and so the first one we created was in Q four of twenty twenty three. And um

Uh and on stage I actually c talked about it as the dark ages of building agents, right? Because it was so hard. And it was me and our our CIO and and uh another another individual who was giving us a hand figuring it out. And To your point, The stuff I did then, I would never do today. I wouldn't have even done it 12 months ago, right? Um, or what have you, because it's just changed so swiftly.

And uh, but it did give a really nice foundation to have been building it back then to now go, golly, look how much easier it is, right? And look how much more thoughtful we can be about the application of these. And the tooling's much better. We we did what we could with the tools we had. Back then we had like agent builder, connect to this data source, deploy. Like, okay. Then we learned maybe not that. Like you need authentication of the data before you hook it to an agent.

before you deploy in one big chunk to everybody you know. Don't do that. But now we're tooling for that because we learn from our mistakes. And the thing I want everyone to know is Microsoft rolls out tools to you because we made that mistake. Cause we built it for ourselves. People are like, how come Purview exists? I'm like, cause we needed it. How come SharePoint Restricted exists? Cause we needed it. How come Foundry Plane exists? Because we need it.

That's why. And that's where all these tools are coming from. That people are always saying, Oh, you guys release a tool every hour. Like, yeah, because we needed it. We thought maybe you'd need it too. So I really want to share some of the latest of those tools and what might be useful for you based on the workflow of, you know, agent building or buying or deploying or whatever it is that you're doing. I'm excited for the event run.

I'm very excited. Thank you, Donna. Thanks so much for being a part of this discussion here just to fill folks in on on what they can expect and that kind of thing. Can't wait to see you there March 17 to 19 uh at the Hilton Torrey Pines in San Diego. Hopefully we see all of you there as well, gang. Um and uh Donna, any final words for everybody here? No, but I will tell you, if you are coming to this event and you're wondering what is a piece of advice I would give you, mine would be.

Bring your own stories with you. Figure out what you can share because you're going to get a lot more value if you've got something you can share along with things you can take away. Because remember, you're not just coming to the event to learn from others. You're coming to share what you have learned because others can really benefit from your lived in experience in the AI version.

That is a great point to end on because what we do is we decentralize intelligence, right, in this amazing world that we get to live in. Donna, thanks so much for being a part of the program here. Can't wait to see you in San Diego. Thanks for tuning in, everybody. We'll see you next time. Thank you.