Ep 29: Navigating the Transformational Shift to AI With Former Snowflake CEO and Microsoft President Bob Muglia

Saam Motamedi: Hi there, and welcome to Enterprise Software Innovators, a show where top technology executives share how they innovate at scale. In each episode, enterprise CIOs share how they’ve applied exciting new technologies, and what they’ve learned along the way. I’m Saam Motamedi, a General Partner at Greylock Partners.

Evan Reiser: And I’m Evan Reiser, the CEO and Founder of Abnormal Security. Today, on the show, we’re bringing you a conversation with Bob Muglia, Former CEO of Snowflake and Former President at Microsoft. For over thirty years, Bob has been at the frontier of modern technology and spent over two decades at Microsoft, becoming one of the four presidents reporting directly to CEO, Steve Ballmer. After his stint at Juniper Networks, he was the CEO of Snowflake, a foundational cloud data company, and today he sits on the boards of several technology startups. In this conversation, Bob discusses how enterprises can navigate the transformative shift to AI, the crucial role of data in AI success, and valuable lessons on innovative leadership. First of all, thank you so much for joining today. It's a privilege to get to chat and excited to hear more about your background. So maybe to start us off, Bob, you've had such a storied career in technology, you've been involved in many foundational companies operating at the frontier of innovation, do you mind giving your audience an overview about your career up to this point?

Bob Muglia: In early 1988, I joined Microsoft as the first technical guy on SQL Server. I was a Program Manager, so I was sort of responsible for what the product definition is and working to build, distribute, you know, get it out the door, basically get the product out the door. And back then we were manufacturing products and we really actually had a manufacturing center in Bothell, which is nearby here in Seattle, and I went down there and helped to build materials together for this SQL Server box. It had 72 floppy disks in it. There were, I think, 36 of both the five and a quarter and the three and a half. So we had both of them there, so it could run on the PS2 and on the other non-IBM computers of the day, and the Compaq and whatnot that were around back then. And the box weighed about 20 pounds with all the manuals and stuff in it. So I helped to put that together and spent 23 years at Microsoft, mostly in the back ends of things, server sorts of things, but had the opportunity to be involved in, or run, pretty much every product group at Microsoft, except games. I wound up my career at Microsoft in the last seven years running the server and tools group, and I ended up as President of server and tools, which had Windows Server, SQL Server, Visual Studio, and management products underneath it. Spent a couple of years at Juniper and then went to Snowflake for five years and built that company from $0 of revenue to about 200 million before Frank took it over and just made it into what it is now.

Saam Motamedi: That's an amazing background and set of journeys, and like, we feel privileged to have you on the show and we've been really looking forward to it, just given the vantage point you have. There's a lot we want to double click on, but I want to double click on the latest super cycle we're all focused on, the AI super cycle. And I think your vantage point's going to be particularly interesting, given both the journey you had at Microsoft and then building Snowflake, a data cycle, you likely have really interesting points of views on what's now happening with AI. So maybe I'll kick it off, but we have a series of questions around AI. The first is where are we in the cycle?

Bob Muglia: Well, we're early, first of all, I'll start by saying that we're very early days. But if you want to talk about the hype cycle and the Gartner hype curve, which I think is a very good way to think about new technologies, because frankly, I do think the technologies follow that hype curve. In Gartner's terms, we are at the top of the hype cycle right now and we are entering into the trough of disillusionment. But in August, Gartner put us pretty much at the very top. And I think probably, we've crested the curve now, and I think we're in the period where we're waiting for the applications. We're waiting to see what comes out. We may not have to wait real long in this case. I think the hype cycle was as loud as anything I've ever seen, and maybe even as short as anything I've ever seen. It was like a six or eight month hype cycle. And so I expect a short trough of disillusionment while people are trying to figure out what's real and what's not and we will enter what Gartner calls the slope of enlightenment, I think certainly by next spring, where we'll really begin to see how this technology will be applied. So it's very early days, to say the least. But I think that the days are going by very quickly, very, very quickly.

Evan Reiser: Are there any particular applications of AI that you feel particularly excited or passionate about?

Bob Muglia: Yes, there are for sure. And to some extent, I think that it may have to do with where I come from too. I think the sorts of things that are really interesting are how we can change business processes and the types of things that change fundamentally in business processes. So the kinds of things I've been focusing on are, for example, one of the companies that I've invested in and I'm working with is a company called Docugami, which is building tools, which has built an application that lets people take the contents of a contract and turn it into essentially a semi-structured document, where the data can be accessed directly and you can understand what's in it. And they're in the process of reinventing the life cycle for contract creation, because those are the kinds of things that are possible, and I think every sort of business, the workflow inside that business will be transformed by AI over the next three to five years. So the interesting thing is, wherever your favorite area is, and if you have expertise, it's now possible to bottle it essentially and to put it into systems that we've never had before.

Evan Reiser: How do you see AI transforming accessibility of data or insights or what impact that has on the workforce or the opportunity for businesses to accomplish more or be more efficient or effective, given your background both at Microsoft and then at Snowflake? I have to imagine this AI is gonna be transformative from getting insights out of data and right into people's brains.

Bob Muglia: Probably the most incredible thing that's happened this year, if I sort of look at this and just look back on it and say, well, of all the things that happened, what is the programming language of 2023? English. English is the new programming language, which is crazy. I mean, it's a crazy thing to think about, but now basically you're programming these models by talking to it in English, and pretty soon they'll be multimodal and you'll just literally be talking to them. So it's making things much more accessible. At the same time, there's a realization that with the emergence of this technology, that it's not complete in itself, and you can think about it in some way, these models are intelligence that can be applied to a business problem. But to be effective, that intelligence has to be combined with knowledge, knowledge about the business, about what's actually happening. And that information is in all kinds of documents, it's stored in databases, and it has to be made available to these large language models. So those two work hand in hand together. But if you put them together, you can get some pretty incredible results. And that's what people are starting to figure out how to do. And when I talk about the tools improving, what I really mean is how can you take your knowledge bases that you have within your organization, your databases, your information, and use that information either to fine tune the models, or, in many cases, to provide information into the prompt to help it find the right answer. There's this thing called retrieval augmented generation, where you combine a database, typically a vector database, with a large language model, and you take data knowledge that's relevant to your business and combine it with this intelligence.

Evan Reiser: I like that analog of thinking about the model as an intelligence that you have to then combine with some sort of knowledge then  actually apply it towards some application.

Bob Muglia: Pretty universal, actually. I mean, if you think about it, even a smart person, if they don't have the background in something, they can't give you a reasonable answer. And while these models are trained on the whole internet, I mean, they don't know what happened yesterday, so you have to augment them in some ways.

Saam Motamedi: Yeah, and I think the other thing you said, Bob, that really resonated with me was just natural language as the interface. And like in some sense, now everyone can be a programmer because the accessibility of the interface has completely changed. I want to double click on the last thing you said, you started talking about techniques like fine tuning and retrieval augmented generation. And we have a lot of CIOs, CTOs of large enterprises listening to this conversation, thinking to themselves, a lot of the use cases Bob is describing could be relevant to me, but how should I actually go build that? Like OpenAI is telling me one thing. I have other vendors telling me fine tuning is important, and you put on top of that the backdrop of just how fast the space is moving. And I think for many, you know, CIOs and CTOs I talk to, it feels a lot different than other technology arcs, let's say, over the last decade. And so what would your advice be to the kind of large enterprise IT teams listening to this podcast, completely believing that AI can transform their business, but also wondering, given how early and fast moving everything is, what the areas they should make their bets on are?

Bob Muglia: What I sort of said at the beginning was, it's time to experiment and to learn. It's not necessarily the time to make the giant investment to build something that you think is gonna last for five or ten years, because I think the technology is changing so fast. There's a significant likelihood that whatever you build today will get replaced in a few years. So just take that into account. But the techniques, they're fairly different. If you want to build a language model that really understands a corpus of data and has a background in a general corpus, that's typically where fine tuning is applied. And what's happened now is you can, of course, fine tune the frontier models that are coming out of companies like OpenAI, GPT-3.5, GPT-4. We now have a wide variety. It's kind of a stunning variety of open source language models to choose from, including very powerful ones like Llama 2 that was recently released by Meta, and those can be fine tuned. And it makes sense for really large scale sorts of things to do this fine tuning. And it's actually come down to become a lot less expensive now. I mean, it's thousands of dollars. It's on the order of thousands or tens of thousands, not on the order of hundreds of thousands or millions of dollars to do fine tuning in many cases.

Evan Reiser: I can imagine actually two different futures, right? One where that kind of AI stack gets more vertically integrated, right? You end up buying kind of like one integrated solution. There's a different version where it's a little more kind of horizontally integrated. And there's like platforms built on top of like, open source components. How do you see that landscape evolving between like what components of that AI stack are open source that are then kind of optimized by enterprises versus there's more proprietary models that become commonly deployed through some of the notable players?

Bob Muglia: Well, yeah, I think the good news is we're going to see tremendous advancements in both the proprietary frontier models, you know, which are starting to become multimodal now as we move forward, incorporating things like the ability to draw and create images, as well as to be able to see images. And we'll see exactly the same thing happening with being able to understand speech and to be able to talk. So those sorts of things will come out over the next six to nine months, I think. In the meantime, open source will follow right behind that, and it's not that far behind. You know, the current open source models are pretty much as good as any of the models were a year ago, and they continue to improve almost on a day-to-day basis. I think everybody thinks the frontier models built by the big companies will stay ahead. But because Meta, at least, is willing to release these in open source, we will see continued open source, I think, advancing. And I think that's really great because it means that companies of every size can participate in this and build unique sorts of things on top of it. It won't just be coming from the big companies. I believe.

Evan Reiser: If we kind of posit that, open source will always be a little bit behind the big companies.

Bob Muglia: Maybe not always though. Open source tends to catch up. Eventually it tends to catch up, but it's behind for now. For now it's behind.

Evan Reiser: It's only like a little bit behind, but over time. In the future, right, the open source will kind of always be as good as the best thing was six months ago. It was not that far behind the grand scheme of things. So it kind of begs the question, you know, where does value accrue to in the AI ecosystem? And how much is it really about the models and the operationalization and application of those models versus the underlying data assets that actually power those? Right, because yeah, how much of the AI strategy really is about figuring out the right models to use versus, hey, what are the data sets or the knowledge base that we're going to feed in to apply the intelligence you talked about earlier, right? Does that mean like the importance of data and organization and data agility becomes really the key to winning in the future?

Bob Muglia: Well, we don't know yet, do you fear, right? So far, like most new gold rushes, the storefronts, you know, the guys with the picks and the shovels are doing really well. Nvidia seems to be the current big winner in everything. I don't know that that'll be true three years from now, but certainly the early winner, they were the early biggest winner and Microsoft's done pretty well as well. The big companies have definitely won to begin with, but I think as AI gets incorporated into everything, one of the attributes of this technology is that it can be built into pretty much any existing application and it can create new categories of applications as well. I think that's where most of the value will go. It's mostly where it always goes is in the solutions and the applications that people use. And given the variety of things you can do with this technology, it's gonna appear everywhere. Yes, if you're a platform vendor, if you have a cloud, you're helping to provide it, you'll probably do pretty well. Obviously, if you're the chip vendor that's behind it, you've got something going here. But I think in the long run, it'll be the applications and the value will accrue to the organizations that actually leverage the technology. And to your point, the one thing that is really important is that to leverage the technology, you need to have control of your data assets. And so everyone should be implementing one of five platforms in the modern data stack and moving to one of those solutions, whether it's Snowflake, Databricks, Microsoft, AWS, or Google, go to one of those platforms and begin to consolidate all of your data assets in one place, manage that properly, because that management of that data is a foundation for actually leveraging it and using it inside these models.

Evan Reiser: Why is the data aggregation organization indexing, the data agility and accessibility, why is that so important for the success of future AI applications?

Bob Muglia: Well, the AI only knows as much as the data that it is trained on or fed with. It has to be provided with data in order to provide answers. And so if you can't come up with that data in a coherent way, you're not going to be able to use these models very effectively. The other thing that's super important is governance associated with it and making sure that the data that you're feeding into the model is appropriate to feed into the model and doesn't have proprietary or confidential information that's relevant to customers in it. And so understanding how all the data is set up and governing it properly is a prerequisite really to using these models in an effective way. And in a way, it's what my book that I just released, The Datapreneurs, is really all about, which is that there's been an arc of innovation over decades that have gotten us to where we are today. And it's these data systems that were built, SQL databases in the 1970s and the 1980s, some of the NoSQL databases that came out later, working with text in the internet and search that happened, all of these advances that happened over time have brought us to where we are today where we can use the vast amount of data that we've collected to build and leverage these language models.

Saam Motamedi: So Bob, maybe continuing on from the data to one of the other big conversation topics around this wave of AI, which is ethics, right? And as you allude to like the capabilities of these models, they're quite significant. And on the positive side, they have the potential to really transform how we work and live. But as part of that transformation we're all going to collectively undergo, I think we all agree ethics are kind of a crucial consideration, right? And I think that's particularly true for a lot of the people listening to this podcast who might run large global enterprises that have the reach and impact that those companies have. And so how should folks like that be thinking about the ethical considerations around their AI development and making sure they're doing it in a responsible way?

Bob Muglia: Well, first of all, I've always believed, you know, in creating companies and growing companies where values and ethics and standards are very appropriate and an important part of creating every company. And I encourage every company that I work with to create a set of values and to really live by those values and to demonstrate that in many deep ways throughout the organization. That's something I focused on throughout my career. And one of the observations I've had is that in the technology space, the values of a company are reflected in their products and it shows through in a number of ways in products. And this is only going to become reinforced and more true as AI takes on more and more decision making capability within these products. And that means that the values that a company has will be present in the products. And so it's really, really important that we be thoughtful about that. I think, because we're going to see thousands, hundreds of thousands of companies building solutions that leverage these models, we're going to see everything because it's what the human race does. We're going to see use for incredible things, incredible good, and we'll also see use for bad things and some things that some would call even evil. And that's people leveraging these things as tools. AI is a very powerful tool, and mankind has used every tool that it has created for every possible purpose, and AI will be no different. But the way it will be used is all based on the values that go into it. Now frankly, I'm a person that believes that there should be a diversity of values and that we should allow that in the world. And so I'm really pleased that it doesn't appear that this is all gonna be controlled by three or four companies and that we're gonna see thousands of companies reflecting and creating products that reflect their values. Now, as I said, some of those will not be so good. And if you love to create spam, well, this is a good tool to create spam mail. I mean, it's a great tool to do it. And so we'll continue to see it used for nefarious purposes, but that's people using it that way. And again, the values we'll have to enforce, and ultimately there is some regulation that governments around the world should do that's appropriate.

Evan Reiser: So Bobby, I'd love to talk maybe a little bit about work on your leadership style and kind of how you think about building a culture of innovation and critical thinking, or even kind of like whatever the values you think are appropriate for the organizations you've been in. Maybe starting with innovation, like are there rituals or traditions or habits or behaviors you try to reinforce in your leadership teams to really kind of drive a culture of learning, experimentation, innovation, any things that's worked there for you that you'd like to share?

Bob Muglia: Yeah, what I would say is that I have always found, I mean, leadership is really three fundamental things: it's strategy, structure, and people. Having an appropriate and coherent strategy; putting in place effective structure that's more than organizations, certainly involves organization, but it's also tools and resources that are required for people to get their job done; and then having the right people, the people that have knowledge about the area and can scale and can help grow other people. If you put those three things in place, you have the foundation for a successful organization. I have always been focused on building products, technology products have been my area of focus throughout my career and so part of it is really understanding the customer and really getting a deep understanding of the customer and then targeting your product to meet that customer need. And here again, values come in and are so important, but at Snowflake we made the first value that we put our customer first. And our first words underneath that were, we only succeed when our customers succeed, and that was very, very true. And we very much live that. And if you do live that, it makes a very big difference. And so focusing on listening to the customer, and yet at the same time recognizing that you don't necessarily deliver what the customer asks for, you deliver the solution the customer needs. And so being able to translate the customer request into an understanding of the problem they're trying to solve. So often I've asked people, what was the problem the customer had? I said, I understand this is the solution they want, but what's the problem that they had? And at least 70% of the time, when you drill in at that level, you come up with a better and very different answer. The customer expectation, customers don't imagine what's possible, typically speaking. They want what they know. And sometimes that's right, but it's not always right. And so that was always a large part of my job. That's what I did when I was at Microsoft when I was a program manager. It's my background. So focusing on that's important. And then really getting teams on track to deliver it and just laser focused on that delivery. And when you have a hard problem, when you're trying to ship something that's difficult or you have a hard problem that you don't know the answer to as a leader, when you're not sure about what to do, basically, I have had one technique I've used for decades, and it's a very simple technique. It's called a weekly meeting, and you take the stakeholders of the problem, of the business problem, and you bring them together on a regular basis, and weekly turns out to be a pretty good cadence. And you talk about the problem, and you talk about how the team is gonna work together to solve it, and it creates cohesiveness within and across teams, because it's typically a cross-team sort of thing you're working on. And what I have found is that the answer that we come up with is always much better than the one I would have come up with myself on my own.

Evan Reiser: So Bob, at the end of every episode, we like to do kind of a bit of a lightning round, just to get some short, punchier, almost like one tweet length, kind of quick hits. Saam, you wanna kick it off?

Saam Motamedi: So Bob, we've been talking a lot about AI today, but maybe I'll ask a non-AI question, which is what is an upcoming new technology that's not AI-related that you're personally very excited about?

Bob Muglia: Knowledge graphs. We're going to see the creation of relational knowledge graphs that enable people to model their business within a database for the first time. And that'll actually be very important, because it'll provide the context to these large language models. It's the other source of knowledge, besides vector knowledge databases.

Evan Reiser: Bob, what is one piece of advice that you wish someone gave you when you maybe first kind of stepped into an executive role that you could offer to share with other folks today?

Bob Muglia: Push hard on what you think is important, but if it's not working, if the team is not agreeing with it, good chance it's not right, what you're trying to accomplish has a problem and you should listen to that feedback. Listen to that feedback from your team. When I push hard on something and the team doesn't respond the way I would think it does, in looking back in my career, there were always good reasons for it, and usually I was wrong or quite often my timing was wrong. Not that what I wanted to do was wrong, but the timing wasn't right. So listen to your team and respect the fact that when they push back, they may have really good reasons.

Saam Motamedi: Maybe switching to the personal side, what's a recent book you've read that's had a big impact on you and why?

Bob Muglia: Oh, well, for sure. I reread all the Asimov books. It was fantastic. I read them all when I was younger and I've gone through and reread all the Asimov novels and found a book, I think it's Robot Dreams that he published, Asimov had a whole bunch of essays that he wrote that talked about his ethics and morals back in the 1970s and earlier. And it was just amazing to look at that because he was such a prophet. He saw this world that we're entering into in the 1940s, before there were digital computers, Asimov was talking about robots that lived amongst us and talked with us and worked with us, and he thought about the ethics of it and that's why he came up with the laws of robotics. And while you can't apply the laws directly on the large language models, they're directionally correct. And we will need to do that. We will need to have similar things in what we're doing.

Evan Reiser: Bob, what do you think will be true about technology's future impact in the world that most people would consider science fiction today?

Bob Muglia: I'll just go to what I was going to with Asimov. We're about to enter the era of robotics where our lives are surrounded by robots. And I think in the 2030s, robots will interact with us in a wide variety of ways. They will be delivering packages for us. They will probably be driving us around. I mean, I think self-driving cars will start to work and be good enough to ride in, certainly in the next 10 years. And I think we will begin to see humanoid robots that help us and help elderly people and work in a variety of elderly care situations at home or in nursing homes and things, as well as, frankly, household robots to help people,  as well of course as all the industrial robots. It's gonna be a very different world.

Evan Reiser: Bob, great to chat with you. Super enjoyed having you. And thank you for sharing your wisdom and advice for the world.

Saam Motamedi: Thanks for joining us.

Bob Muglia: Great to be here. Thanks a lot.

Evan Reiser: That was Bob Muglia, Former CEO of Snowflake and Former President of Microsoft. 

Saam Motamedi: Thanks for listening to the Enterprise Software Innovators podcast. I’m Saam Motamedi, a General Partner at Greylock Partners.

Evan Reiser: And I’m Evan Reiser, the CEO and Founder of Abnormal Security. Please be sure to subscribe so you never miss an episode. You can find more great lessons from technology leaders and other enterprise software experts at enterprisesoftware.blog

Saam Motamedi: This show is produced by Luke Reiser and Josh Meer. See you next time.