Speed, safety, and scale - Microsoft's playbook for health AI

[chimes] If I say like, I mean, and no one has a crystal ball, so I could be totally wrong on where we are in [chuckles] 12 months from now. But I would say like advising startups today, it's, I, I think build fast, like speed is the, like name of the game. And I, I know I'm saying that in a... Coming from tech, it's like, of course, you build fast, break things.

That's not what I'm saying. But I think just the speed of innovation has increased so much that I think there's an expectation of, um, like faster product cycles in general, no matter which vertical. And then I, but then I couple that with scaling smartly, right? So it's not just about building quickly, but it's making sure that you have a plan to grow, um, that is sound and makes sense.

And I think in particular in a vertical like healthcare, making sure, you know, responsible AI, security, kind of safety measures are all in place. And if that slows down speed, I think it's okay. Uh, but it's, it's that tension between those two things, um, I think that's gonna be the focus like in the next six to 12 months. Um, right?

I, I don't, I don't see the pace slowing down. Um, uh, in fact, I see it ever increasing. Welcome everyone to today's Startup Health Fireside Chat. I'm Unity Stoakes, Co-founder of Startup Health, and I am thrilled to welcome Kimi Grewal from Microsoft.

We excited about today's conversation because we are in the middle of, I really think one of the most consequential shifts in healthcare since the dawn of the internet or of the electronic medical records. And that is because AI is really moving fast. And I think critically, the window for founders and healthcare companies to build the right AI-enabled companies is open right now. I think the big question is: How do innovators and builders turn AI potential into real-world products, partnerships, scale, and impact?

And well, Kimi has a front row seat to that answer. Uh, Kimi is Vice President of AI Ecosystem Partnerships at Microsoft, where she helps shape how leading frontier AI companies and innovators connect with Microsoft's global platform and reach. Um, Kimi also brings more than 15 years of experience in health and life sciences from her time at Accenture. And I think most importantly for us, Kimi has been a Startup Health Impact Board member since, I think, 2021.

So many years. She knows this community well. She understands the innovator journey, and she is here today for an interactive conversation with all of us. So Kimi, welcome back to Startup Health.

How are you doing? Thank you. That was a very generous and kind introduction. I'm doing well, and thank you for having me.

It's great to be here with you all. So in Microsoft, I've had a few different roles, but the role I'm in as of the beginning of this calendar year is effectively leading our partnerships with the leading, I'd say, AI ecosystem partners. So think the silicon partners like NVIDIA and others, as well as all the model labs such as Anthropic, uh, Meta, xAI, Mistral, and really thinking about, um, how do we partner with those model providers and ecosystem providers on behalf of the ecosystem overall, and how do we make sure that they're serving customers, enterprises, and startups well.

That's really important, and you've got this unique view across so many different things today, but also, you know, over the course of arc of your career. You mentioned frontier model companies. There's infrastructure players, startups. You've been on the family office investor side, enterprise partners and, and customers.

I guess maybe frame the moment that you're seeing within that seeing. How do you describe [chuckles] the moment, and maybe some of the things you're seeing that's working in this new world and some of the things that aren't? I mean, I think just to put into context this journey, right? So like the ChatGPT moment that really made, I'd say, like AI today mainstream was last November, about four years ago.

And so I think just in four years how far the market has come is, is really amazing. And I think then if you take stock of like how quickly it's been evolving year over year, like this year alone, I'm sure this community knows, but like the amount of venture investment in AI in Q1 surpasses the amount for the full year of 2025, just to give you like a benchmark of like the pace of investment. And then also as I look at, you know, from a Microsoft perspective, like I think for a long time, everyone's looking at the frontier labs, which are still really important, like your OpenAIs and, um, Anthropic of the world.

But now we're seeing open models really close the gap from a cost perspective. So I see, if I think about the ecosystem, both enterprise, but really at startups like building a lot more with open models. I think that's a big shift we're seeing because of that, that performance gap closing. A, a few other, I think, observations this year in particular, I would say, is like context and data being like the primary moat around which folks are building, whether they're building solutions, um, within enterprise or where startups are building, right?

It's the model is not the differentiator, but it's like the, the workflow and the data and the proprietary data set that you have that really is the differentiator, and I'd say the, that drives stickiness in the market. And I, and I think if I look forward, I think physical AI is where the puck is going, and so it'll be interesting to see how that plays out, uh, particularly in the health, like the health space. Um, but I think if you look, if you were - if I had a prediction of like, y- you know, next year, this quarter, what we'll be talking about more, I think it's physical AI.

I think it's, it's really an exciting opportunity for all of us in, in health. I wanted to go a little deeper into, I know you were just at Viva Tech. You were at Microsoft Build. You're traveling all over the world.

You're seeing kinda what's the big enterprise is doing, but you're also- At these places where you're seeing a lot of different startups, and one of the things you were kind of touching on is, and I saw some of your observations that you shared insights online as well, like, values moving up the, the stack, right? And, you know, you mentioned models matter, but the winning startups, what, what are they doing? Yeah, I would say a few things. I think also, like, the, the tools that we have available today, I think the speed of product development and then shipping of that product.

I think the, the leading startups are just shipping a lot faster than they used to. And then also shipping product and getting real time customer feedback on that product, and then being able to iterate and release the next version of that product. The, the speed of that feedback loop, I think, is, uh, really different than it was 12 months ago, and I'd say that some of the leading startups... I know from a Microsoft perspective that we look at for investing in, it's, like, how quickly they can ship product, iterate, and then get that first POC with a customer.

I think that's, that's definitely an important area. I think also scaling smartly. So for anyone, I think obviously at the end of the day, depends on who your end customer is, but anyone who's looking to sell to hospitals or to large, you know, businesses, making sure that the speed is not getting in the way of being kind of enterprise-ready, security, governance in a regulated industry like health, that's so important. So it's balancing making sure you're shipping product, but in a safe and secure way, um, so that you're not just getting customer feedback on something that can never scale.

So what you're actually putting out there is something you can get to production scale, I think is another really important piece. And then I would say, like, ecosystem, super important. I think today, like, Microsoft's point of view certainly is, like, our success hinges on the ecosystem of, you know, partners and startups and VCs. And I think what I see the leading startups is making sure they're investing in up- like, a intentional partnership strategy upfront, and making sure that that's not an afterthought post product and initial customer feedback.

Yeah, that's, uh... Healthcare tech companies versus just tech companies shipping fast, it's almost like two different worlds in, in the context of safety, security, privacy, all these other issues. Tension of needing to keep pace with how fast you're able to build with, you also have to balance validation and, and kind of the, all the safety, and the components as, as you're building. Are you...

So that speed to shipping, are you seeing that also with, with healthcare startups or, or just tech companies more broadly? And then kind of a follow-on to that is when you say ship, are, are any type of company, are they shipping weekly, monthly, daily? It's just... Has the cycle...

How fast has the cycle gotten from what you're seeing? Yeah. I would say broadly in tech it's weekly. Um, in regulated industries it's not quite as fast, but also it depends, like, which sector in the regulated industry.

So I think in healthcare, for example, a company who is looking at documentation and workflow automation can ship very quickly. Like, the barrier there is much lower than if it's looking at, like, drug discovery or the diagnostic space, right? So I think it depends on, like, the slice of healthcare that you're operating in and how quickly you can get a, a product out to market. And that's why I say that balance is important because it's not just about speed, it is also about being able to scale post, like, getting something out in the market.

And so if, if you don't have the appropriate safety and security and the governance or, like, alignment with regulations, you're not winning by getting a product out quicker, right? I think spending the time to do that right up front will probably serve you better. So I would bifurcate, like, even within health or ev- you know, there is a difference between kind of l- lower regulated verticals. If you're doing a revenue cycle or a payer operations type solution versus...

Or a patient engagement virtual care solution, a, you know, deeper clinical workflow solution. That context matters. And as you compare, you know, to your days maybe working with life science companies or, or big enterprise healthcare, and then you come across those same types of partners, customers, enterprise healthcare companies today, are you seeing th- those big partners and customers operate at a different pace and speed? Or, like, what's ha- what's changed, if anything, from, you know, when, when your days at Accenture, say?

Yeah. I- it's really interesting, and I think a part of the, what I've observed is not what we expected. So the part we expected is you obviously see all the software development companies and the AI native companies, like, quickly adopting, right? And, and leading from the front.

What I didn't expect is, like, regulated industries, like financial services, but also, like, healthcare, being super fast followers in the enterprise in terms of adoption. Um, like, I expected them to be laggards, and maybe retail and media, telco, kind of some of the other industries to be faster followers, but we haven't seen that to be the case. I think, a- and I think there's multiple potentially hypotheses for that, but I think a big part of it is a lot of the problems, I think, in regulated industries have been known, known challenges for a long time.

And I think now with where the AI capability is, you actually have a real opportunity to solve some of those challenges. And so there's a eagerness to figure out how to cut through the regulatory tape in a way that makes sense, but to adopt. So I think we're seeing that adoption faster than we would've expected. Even on open models, I would've thought open models would be, like, lower on the list, and I think they're, at least on the experimentation front, there's a lot of pickup in, from healthcare companies- Slower shift to production still, because I think you have to test kind of the, the safety and governance around all of it, but definitely lots of experimentation.

And j- going back to something you said at the beginning, I mean, the context is this is, we're only a couple years into this. Yeah. And I, I, I suspect if we're talking even 90 days from now, it'll... A lot will have even changed since today.

So that's, that's the pace we're talking about, which is quite extraordinary. One of the other things I've, I've - insights that I've seen you share or talk a little bit about is capital efficiency becoming the default now. Yeah. I don't know, maybe unpack that a little.

You know, more investors pushing startups to, to really build on existing models rather than reinvent. You talked about open source models people are building on. What are the... What are you seeing today, and maybe put also your hat on from when you were at a family office and, and other places where you're advising startups to, to build, and how you really have to be thinking about your capital very, very carefully as you're building.

Yeah. I think, I think a few things. So one thing at Microsoft is we are thinking about whether it's companies we're investing in or whether it's startups we're partnering with is, like, kind of our rubric we look at. I think one is, like, start where the money exists today, right?

So where is money already being spent in healthcare? And so, like, who's your buyer, like, thinking about that upfront. And I think the other piece is the point on capital efficiency, like we see startups are gonna win on the workflow, not models. 'Cause I think if anyone who's following the state-of-the-art models, the leaderboard changes daily on who's on top, new releases.

And so your differentiator is not gonna be the model. Maybe if you have a proprietary model and then you're appending that with your proprietary data. But I think in general, like, the fastest path to a solution in the market is building on something that exists today and then owning and pointing it to a specific workflow where you know the buyer already, and then what is your differentiated data that you're - is really your moat for your solution. And so I think...

And then I think being eyes wide open on, like, the regulatory lane that you're picking, right? So my point on, are, is there a solution around, like, rev ops or kind of patient experience virtual care, or is it like a drug discovery or something much more involved? Because that's gonna really impact time to market and time to ROI as well. That's really important, companies not rely on one model.

Because the leaderboard's changing- Yeah... you know, you gotta really be thinking not about a multi-model and be flexible. Would you recommend people go all in on one or just be very adaptive? I guess I'm kind of leading the witness here, but, you know, what, what's your thought on that?

Yeah. I would say you don't wanna be over-reliant on any singular model. So you wanna, I think, have the flexibility in your, in your architecture to be able to swap models in and out. And I, I'm being a little pedantic.

Like, it's not so easy to swap, like, but, but you should be able to swap. And so the differentiate... 'Cause I think if you have too much lock-in with one model, I think you're assuming too much risk for something that's not stable in the market. Meaning today, Anthropic's models are on top.

I think tomorrow it could be a totally different model, and I think we're still... I'm quite bullish on, like, some of these open models being at the top of the leaderboard for quite some time at a much lower cost. So I think the other thing is, like, even if you have super high performance with a specific closed model, is it at too high of a... I think that really matters for a startup.

So I think my, my general view would be design for being able to be flexible on the model. But I think what you really wanna do is make sure you protect your IP, and that's why I go back to, like, data. And I think enterprises and startups are, um, this is the piece that they should be focused on are, you know, and making sure is, like, no matter what model you're using, making sure that your IP remains your IP and you're not exposing that anywhere because that truly is your differentiator.

Yeah, I think that's, that's essential. And but shifting gears a little, I'd love to talk about Microsoft for, for startups and, and I think the Pegasus program, for example. And for those of who may not know, what is Pegasus? Let's start there, and then I'm very curious to learn, uh, for those who are, are tapping in the right way, what trends are you seeing?

What are they doing to really get traction? But Microsoft for startups, Pegasus. So for broad education, there's, there's two big programs. One is Microsoft for startups.

So Microsoft for startups, anyone can apply, startups.microsoft.com. You set up your Azure account, and you immediately get startup credits, and you can unlock up to 150K of credits as you build and grow your solution with Azure.

And then the, uh, Pegasus program is kind of within the broad pool of Microsoft for startups, like a, a selective group where we nurture and provide much more hands-on support for co-sell and selling in the, in through our marketplace. And so I think, uh, they're typically the f- the companies that got selected for Pegasus, you know, are aligned to, I would say, Microsoft priorities in that they're selling to enterprise as first criteria because that's our primary customer. They're continuing to build, test, and deploy on Azure.

I think there's, there's focus on verticals within health life sciences, in this case, that are important for aligned to our kind of industry strategy and our, um, [lip smack] go-to-market motion. So I would say broad brush, like anyone can start with Microsoft for startups just from startups.microsoft.com.

And if you're, if you're part of that cohort, people can discover any of your solutions through our marketplace. Um, and I think the other big trigger we look at is how much marketplace traction you're having. So how much are you selling through our marketplace, and the more you're selling, the more hands-on support you'll get from Microsoft. This is important and it's, you're, you're one of the key players out there.

Are there... Is what we're talking about now the, the right, best pathway in? Are there other pathways? Are there other opportunities for health tech, healthcare companies specifically that in addition to what we talked about be looking- I would say for, for startups, we try to one single front door.

So the single front door should be Microsoft for Startups. That being said, I think once you're in the door, and particularly if you are into the Pegasus program, we can connect you with, like, Microsoft Research, for example, depending on your businesses and how our research organization is, um, investing in that area. So I think a lot more doors open, but that's the right front door is Microsoft for Startups. And then I think from there we can help with enterprise selling, joint marketing, reaching more customers through our distribution and sales channels is, like, the primary benefits, right, that, that most companies get beyond the credits.

Yeah. Okay. You know, let's, let's talk about the future for a second. Um, and by the future, I mean near, near future, [laughs] I guess, you know, over the next 12 months.

Where, where do you see this all going, and how should builders, innovators be thinking? How should they be... It's, it's, it's moving so- it's so dynamic. It's moving so quickly.

Um, it's hard to keep up by week, but yet we also need to plan. We've gotta think about capital allocation. We gotta think about fundraising. We gotta think about product cycles.

Is it, you know, just hold on by the week? Or, or how do you... For the companies that you advise or the ones you're seeing, the builders that are really doing it right, into the future, you know, say 12 months, and how we should all be thinking about that? Mm-hmm.

Yeah. I think one thing I will say is, like, even as an exec in Microsoft, I find it very hard to keep up with the pace of innovation. So I would just say broadly, it is impossible for any human to keep up with everything that's happening in the marketplace. I get a behind-the-scenes view sometimes, and I still can't keep up.

And so I think that's one, like give yourself grace on, you know, what are you trying to keep up with. I would pick kind of like three or four spots where, that are really relevant to what you're building that you wanna keep up to date with. I also find it really useful to, like, if there's certain thought leaders or builders that I really admire, like just keeping up with what they're doing because they're kind of scanning that segment of the ecosystem on your behalf in many ways.

So if you stay in tune with them, you generally are kind of up to date on the, the evolving landscape. But if I, if I say, like, I mean, and no one has a crystal ball, so I could be totally wrong on where we are in [laughs] 12 months from now. But I would say, like, advising startups today, it's, I, I think build fast, like speed is the, like, name of the game. Coming from tech, it's like, of course, you build fast, break things.

That's not what I'm saying. But I think just the speed of innovation has increased so much that I think there's an expectation of, like, faster product cycles in general, no matter which vertical. And then I, but then I couple that with scaling smartly, right? So it's not just about building quickly, but it's making sure that you have a plan to grow that is sound and makes sense.

And I think in particular in a vertical like healthcare, making sure, you know, responsible AI, security, kind of safety measures are all in place. And if that slows down speed, I think it's okay. But it's that tension between those two things that's gonna be the focus, like, in the next six, 12 months, right? I don't see the pace slowing down.

Uh, in fact, I see it ever-increasing, um- Right... like at the past four years. We're, we're, we're all in this together, so you said give, give yourself grace because I, I think everyone's going through this all at the same time. So you and I are both, you know, Bay Area, West Coast, in the bubble.

Yeah. But you're also traveling all over the world. You were just in France for, for Viva Technology. You're meeting with global companies all the time.

Are you... Give us a little global perspective. We've got a lot of companies from all over the world that are builders and innovators. Are you seeing the same trends and patterns all over your travels?

Or are, are you just mainly what you're describing, kind of a West Coast, a US base? I would say the, the shift is global, broadly meaning, like, I think if you look at any enterprise globally, you're either being AI native or figuring out how to AI into your workflow and businesses, like, is looking at that 'cause it's a topic in every board discussion. I think there's nuances in an application by geography. So, like, if I look at Asia, some of the best open models are coming out of Asia, and so naturally you're seeing much higher traction from then building with open models as a starting point versus the frontier labs.

If I look at Europe versus Nor- like the Americas more broadly, like Europe is, has always been a leader on regulation. Um, and so it's sovereignty, data residency. Vertically having all the infrastructure in the region is super important, much more important than I'd say in the States where kind of a global standard is, is acceptable. So I think those nuances exist.

So, like, if you're building in Europe, I think trust, and I think I put this in my, on LinkedIn, like trust is a part of product market fit. Like, you can't later on think about trust or sovereignty because that's an expectation in the market, which means perhaps slower speed of product cycles, right? Because you're, you're building that in upfront versus in the Americas, I would say our venture community largely rewards pace first, then figure it out later in terms of what needs to happen next.

So I think those are the, some of the big nuances I see. I, I would also say- There is a, like, definitely much faster adoption in the Bay Area than there is broadly. Like, I see a lot of enterprises come through the Bay Area and do a innovation tour of us, Google, and all of the VCs here. And it's interesting, everyone here is talking about OpenClaw and all of the latest, you know, and then people come in from other parts of the world and are still trying to figure out, like, how do you apply agents in the four walls of an enterprise, right?

So, like, the, there's a broad understanding, but, like, the delta is quite wide based on where, where you sit geographically. But while, you know, we only have a couple minutes left to end a few minutes before the beginning of the hour, but Kimi, what would you say to startups? What should one thing that a startup should do, like, today or tomorrow that maybe they're not doing in order to get in the game? 'Cause it's, it, it, there's a lot of, of innovators that are, you know, they've been so heads down on kinda their plan from the last 12 months- Yeah...

that they haven't yet recalibrated from, as far as I can tell, the world changed again in January and, you know, a month ago. And so what would just be one practical thing that would be maybe useful to early builders, early innovators that they should just get started? Yeah. I, I think one thing I would say is, like, because the field and the space is evolving so quickly, like, you're actually not behind if you start today.

Like, I think people like to say, like, "Hey, you'll be, you're, should've started four years ago." But even if you're a startup today that is trying to figure out how to leverage the latest tools, like, like, you can really teach yourself almost everything, uh, right now. And so I think one thing is, like, I think some folks feel like it's daunting, and I think you, like, the, the s- entry point is really easy if you're a self-starter. If, if I was a f- a founder today, I think the opportunity, like, to exit has never been brighter, because enterprises are not AI native by nature, and they're looking, uh, to tell their boards how they are investing in AI capability.

And if there's a really great, focused startup that's operating top of their lane, like, I think your ability to find a path to scale and exit is, like, much greater today than it ever was before. So I think just being mindful about tha- if that is the plan for you or the strategy for you, being mindful about curating that today and thinking about that may not be five, 10 years down the road, may be much sooner. That is such an important insight. I think that's a, a great place to end the conversation.

Kimi, we are so grateful for your wisdom today, and also just your, um, all the support and help you've given over the years and how you are of our whole ecosystem. Thank you, thank you, thank you. Again, Kimi, thank you. Everyone, thank you all for being here.

Have a great day, and we'll see you all soon. Thank you for having me. Have a great week. Okay, bye.

[jazzy music] And finally, an invitation to all of the extraordinary health transformers, whether you are a startup, someone building the future of health, or whether you are an investor, funder, or buyer of the solutions out there in the marketplace, this is an invitation to you to join the StartUp Health member community. Learn more at startuphealth.com. Look forward to hopefully seeing you soon.