AI Front and Center | with Matt McIlwain (Managing Director, Madrona Venture Group)

That brings us all the way back to Seattle where you've got these clear leading players in cloud. And even more importantly, you know, like I was saying, you know, 10 years ago, 10+ years ago now, when we started investing in, in applied AI, the insight was, is more and more of the data was digitized and could be accessible to build recommender systems, predictive systems, and to do that in the cloud with your data so that you could build these kind of models and then deploy them and manage them at scale. What you want to have is you want to have access to the ideas, the people, the experiments that are working and not working in the aggregate learning, you know, more robustly, more, you know, kind of with high fidelity and earlier than anybody else. And I think that's what makes, you know, the greater Seattle area, not only the cloud capital of the world, but I think it will be, if not the, you know, one of the very best, I think it should be, you know, kind of viewed as one of the very best in the world centers of excellence for AI.

Hi, I'm Boaz, founder and CEO of Simply Augmented, and I'm excited that you're tuning into Shift AI, a show that explores what it takes to thrive and adapt to the changing workplace in the digital age of remote work and AI. In today's episode, I'm thrilled to have Matt McIlwain as our guest. Matt is general partner at Madrona Venture Group and has been a key technology leader in the Seattle community for over 20 years. During our conversation with Matt, we explore his background and experience building companies and advising founders to focus on the things that drive business growth and innovation. If you're someone interested in how AI is going to factor into the startup and venture ecosystem, you won't want to miss this episode. It's great to have Matt on our podcast and we can't wait to learn more from him. Let's get to it. All right, Matt, welcome to The Shift Podcast. Thanks for joining. Us today.

My pleasure. Thanks for having me.

I am really, really excited to have you on the show today. Everything that I've been thinking about lately has been about AI, and I know that you've been talking a lot about it and it's been on your mind, and I'm sure you're talking to your partners about it. So it's just perfect timing. I'm just really excited to talk to you today.

Well, I'm looking forward to jumping into the conversation, Boaz, and there is a lot going on in AI and generative AI. It's now kind of the popular topic these days. So look forward to unpacking it.

Absolutely. Well, before I want to do that, I like to always start the show where I tell you your background and you correct me if I get anything wrong. So the show, you know, primarily is about the future of work. And so a lot of the audience that listens to the show is excited about what emerging technology is going to look like in the future, how it fits into their work. And so I always like to talk a little bit about people's backgrounds and about their, their work environment. So let's, let's just dive in. So you studied at Dartmouth College, you got, uh, you got your MBA at Harvard, and then, uh, you came out of school and worked for McKinsey as an engagement manager, did that for 4 years, and then moved over to be vice president of business process Genuine Parts Company, which I'm really interested in hearing more about. So, but after that, you— is it? Yeah. Well, after that, you know, you've spent the last 23 years more, almost 24 years as the managing director at Madrona. And you've seen a lot during that time. And then also, I'm sure that you're involved in lots and lots of companies, but you're a board member at Fred Hutch as well. Am I missing Anything about your past?

Oh, I think those are some really good highlights. I did a couple of years of investment banking between Dartmouth and grad school and actually did a government degree as well in grad school, which was fun at the Kennedy School. And yeah, there's an interesting story arc there of investment banking, management consulting, and real-world operating experience and what's been most helpful as a venture investor. But We could get into that if we wanted to.

Yeah, we might have to. But first question is, what was your first job where you actually got paid?

Yeah. So, I grew up in Miami, Florida, and went to a big public high school and worked as a front-end service personnelman at Publix. Publix is a grocery store, and another name for that is bag boy. But they called it front-end service personnel because we were there to be customer service for our guests, our customers who were shopping at Publix. And Publix has done incredibly well as a grocery store business, mostly in the Southeast over the last many decades. But that was my first job when I was 16 years old. And I'm not counting refereeing and coaching in soccer. That was a lot of fun too. But this was kind of that first sort of quote, real-world job.

That was the first one where you got the paycheck.

Yeah.

Well, you know, it's interesting that when I hear you speak, I hear you counsel founders to really think about customer first, but you were right there in front of the customer during that job. That's for sure.

Absolutely. And the expectation in the hot, sweltering Miami summers was that you were to take an offer to take the groceries out for every single customer. And so, you'd go between the freezing air conditioning of the grocery store out into the sun and back over and over and over again. And so, yeah. You got to be customer-friendly, provide good service.

Yeah. What was your favorite job over the years? Is it Medrona?

Oh, you know, there's, there's really two. I mean, Medrona is, I, you know, I just love the opportunity that we have to, to come alongside incredible founders and be company builders with them. I mean, they're doing so much more of the, the hard work and heavy lifting, but I think our team and the experiences that we've had and being able to coach and mentor and advise and just roll up our sleeves and help is such an awesome job to be able to have. Interestingly, one of the things that informed me very early on when I was in college, I actually started a company and it was building a board game business. It was a Dartmouth game. It was kind of in that era when Trivial Pursuit was all the rage, but this was more of a, like, you know, kind of how do you graduate from college? And it had a kind of a Trivial Pursuit style board that you had to navigate around, and it was the Dartmouth green. And so that was a lot of fun for me to do that, to learn a lot, you know, make some mistakes as the founder and CEO. Ultimately, it was a reasonably successful business. So that was, that was fun from a learning perspective and helps me just in a small way emphasize, empathize with with founders.

Yeah, for sure. And how about your family? You know, what were your parents' jobs? And was work and business talked about at the dinner table? Was that a topic that you guys discussed? Was that part of your growing up?

You know, my parents actually met in the military. My mom never went to college. My dad was the first in his family to go to college. And they met in the U.S. Army, actually at Fort Knox. My dad had a, you know, a business undergraduate degree, and he ended up in the technology world quite a bit, you know, back like early modem companies and things like that. And so, you know, I learned a little bit about that from my dad. My dad was a pretty quiet guy. And so, you know, probably not so much. It really wasn't a conversation around the dinner table. But it was something that was kind of in me, I guess, genetically somewhere, that I did have a fascination for entrepreneurship and innovation and always being curious. And so I was in Junior Achievement when I was in high school and different things like that that kind of reinforced that personal interest. And I would say even more so subsequently in our family with our 3 children, that probably was much more the case of kind of having conversations around the dinner table about businesses and startups and innovation and technological trends, those sorts of things.

Yeah, super interesting. I mean, the last 23 years, you know, I want to talk more about Madrona. There's just so much has happened in Seattle. And I've been here for the last 20 years too. I moved here from San Francisco in 2000. And so, you know, there's been a lot of changes in the city and also in the tech community. I really like to think about Seattle as at the forefront of so much technology at these big companies, but also in cloud and now in the startup ecosystem. It's really exciting. So with AI coming, it seems like because of our cloud experience, the time is right for us in Seattle to really, really shine. So tell me about the work and tell the audience about what you do as a venture capitalist and, you know, how your investment thesis relates to the future of work and emerging tech. Generally speaking, about your work at Medtrona?

Well, I'll talk about that, but I completely agree, Boaz, with your perspective on Seattle. I moved here in 2000. And honestly, that year, if it hadn't been for a debt financing that Amazon raised as a relatively new public company, they may not be around still. And Microsoft was kind of on top of the world then, but did go through some some challenging times. And one of the interesting things to note, even if you look at just the past 10 years of Microsoft, since it's almost on the 10-year anniversary for Satya Nadella as CEO, their market capitalization's grown from $250 billion to $2.5 trillion. And so this economic opportunity and value creation and therefore kind of capital and people and talent that can be reinforced into kind of creating a flywheel in the Seattle innovation ecosystem is super exciting for me and for us. And then that leads into what we do at, you know, Madrona. I mean, Madrona's strategy is to be really early stage. It cannot be too early for us. You know, I actually prefer being involved in kind of company formation stage. So whatever we call that these days, pre-seed, seed, you know, stage, we also lead Series A financing rounds. And then Over the last 7, 8 years, we started a whole separate fund that is for more Series B and Series C rounds that will be new investments for Madrona. So, we can really cover investing from a pre-seed stage to kind of a Series C stage as initial investors. But then we have a phrase, being there from day one for the long run. And so, that doesn't just mean with our time and our kind of ability to hopefully add a ton of value to the company. But also our capital and our network and our mindshare for you as founders. And that really does involve a lot of different elements of, as a venture team, we're trying to get to know founders and develop what we call prepared mind thinking, thematic investing before the rest of the world has kind of figured out that that is going to be one of the next big waves. And so, if you go way back, of course, consumer internet in the mid-'90s with Amazon and others, classmates and others, classmates could have been Facebook but was actually a good outcome for us but was back in that era into then turning software into software as a service. It's just taken for granted that software is delivered as a service these days in the 2000s. And then, you got to the later part of the 2000s. We'd been thinking a lot and doing a lot of interesting things in virtualization and realized that this thing that came to be known as the cloud and Amazon Web Services, we did an event in 2007 with the Amazon folks to launch AWS up on Capitol Hill in Seattle. And lo and behold, the last big piece, which came really about, and we'll get into it, but it really came about about a decade ago. We were making our first investments in applied AI back in 2012 and 2013. And we can unpack why that time kind of seemed to be the sort of the first big turning point in applied AI.

Yeah, I want to hear more about that. For those people that don't understand really venture and venture capital, you know, you guys are making big bets, but you're also, Madrona in particular is coming in early. You really understand these founders before some of the other big venture comes in later. And puts more money on the table. And that's an interesting position to be in. Can you talk a little bit about that, that perspective and the way that you engage with founders early?

Yeah, we, I mean, we, part of it is this idea of having, you know, prepared minds in these, you know, these categories. And what we find is that if we pushed ourselves, you know, to understand why some big changes are afoot or some big problems might be able to be solved better than they could be solved before, that leads to really engaging conversations with founders, sometimes even before they're founders. You know, we have a whole bunch of people that we kind of jokingly refer to, not really, but we kind of jokingly refer to as future founders. We just think these are folks that might find something someday. We want to get to know them then, you know, before they've even said, oh no, I'm feeling compelled to go start a company. And so, we really enjoy having those kinds of conversations about, you know, I kind of think of this as sort of the triangle of, you know, why questions, you know, talking with founders about why is there this problem that's out there and why, you know, are you obsessing about how to solve that problem better than it's ever been solved before? And then the all-important question of why now? I mean, we, We know that we're going to be too early, you know, metaphorically speaking, in a number of categories. Back to your point about getting out in front of these trends and opportunities and problems to be solved, maybe out ahead of others at a pre-seed or a seed stage. And we're going to sometimes be wrong about that. We're going to sometimes be too early. We were probably too early in VR and AR, for example. And as a category, you know, I don't think we were too early in AI, but, you know, we had some early wins that maybe we sold too early in AI. That were companies we founded back in that 2012, 2013 timeframe. But why now are there disruptions from technological changes, societal changes, business model changes? And then ultimately, kind of why this? Why can this kind of thing be the better solution to that problem? Of course, the final question that founders are asking themselves, and it's really a two-way street, is why us? You know, why us? Why are they the team that can bring together a bigger team and go solve that problem? What's their unique background, their unique insights, their unique capabilities? And then they should be also asking why us about why can Medtrona be a part of that journey and help them have a disproportionate, you know, kind of advantage and ultimately being, you know, one of the winners, if not the winner in an emerging category.

Yeah. Yeah. Is there a fit both sides?

Yeah.

Well, that leads me to the question about AI. I mean, you're in a unique position because you're seeing deals all the time. People are pitching you all the time. I imagine that you're seeing AI in pitch decks all over the place these days. And but you've been thinking about this for a while on the automation side. And so, yeah, let's talk a little bit about your thesis around AI, around automation, around machine learning. You know, a lot of people this year have gotten really excited about AI and it's some of the, you know, they're thinking about it for the first time, but this has not, it's been around for a really long time. So yeah, let's talk about your sense of that and how does that relate to the frenzy that we're feeling right now?

Well, if you go in the way, way back machine, you know, artificial intelligence as a term broadly viewed to be coined back in the late '50s at a summer gathering of professors and academics at Dartmouth College. It was a group that was well beyond Dartmouth, my alma mater, but that there was this gathering. And so it's definitely been one of those technological areas that has been around for decades and it's had its ups and downs. And then I think it's also helpful to recognize that data has always been a critical part of how we kind of can improve the human condition. And we first had data stores, databases that were largely helpful, even in the simplest sense of them, to be able to be systems of record. And systems of record are important because we could keep track of things that previously maybe we had to use much more manual processes and books and booklets and ledgers and things like that. And then we moved on to, in the last 30, 40 years, what happened is those systems of record were being queried a lot for things that were more analytically oriented. And so there was a whole world that evolved around systems of analysis. And largely speaking, we think of those things as data warehouses. And even the more— we've been investors in Snowflake for many years. And Snowflake was kind of the innovation there was being a cloud-native data warehouse. But it still was, at least in its early years, much more focused around being a system of analysis on top of all the data that was gathered in these systems of record and sometimes across different systems of record, hence the data warehouse construct. But now what's happened is that as more and more in the past, you know, really couple of decades, more data has been generated and often more data that can be generated than how humans could sort of kind of typically process and more data types have been generated in digital forms. Including interesting types of data like biological types of data or even audio types of data that are increasingly captured from a digital form versus an analog form, is we could apply some of these algorithms that sat on the shelf in fancy research papers and apply them first in an embedded way into interesting applications. And that's really where data and data warehouses and these different kinds of data stores and now data models have become more systems of intelligence. And so I think that's that evolution in a broad sense and puts a little bit of context around what's happening more in the here and now. So getting into the systems of intelligence a little bit, then you say, well, wait a minute, hasn't that been around for a while? And at some level it has, but again, mostly embedded. And we experience those systems of intelligence when we do a query in Google, we do a search on Google and we get a result. They're using algorithmic techniques to basically predict through what started out as the PageRank system but has built out into far more sophisticated algorithmic kind of speedy algorithmic approaches, what's the best results for me from a search perspective. But the same thing applies to Amazon. I was talking to some folks yesterday about how many places within Amazon AI is embedded. Whether it's product recommendations, whether it's Alexa, whether it's all kinds of things and all the finished services and the underlying services in AWS, their own advertising business, which is a $40 billion business. So, they're doing quite a bit themselves as well and have been embedding it. Or of course, we get the benefit of these things with recommenders and Netflix and Spotify and so on and so forth. So, there's lots there. What's different now in the last set of years And by set of years, really kind of when some of these transformer models, things like BERT, now we've got this whole evolution of different kinds of frameworks and so-called large language models, foundation models have emerged, is twofold. One, we could interact with that predictive capability through natural language. We can just write or speak and sort of ask it to do something for us. Versus it being embedded within the recommendation that Netflix gives us. And secondly, those systems are appearing to and actually are generating things. So, they're not just giving me recommendations, predictive systems, but they're actually generating something new. Now, they're actually under the covers using very fancy forms of predictions to predict what word to put next to the next word or what code and what sequence to generate as code within something like a copilot system from GitHub, Microsoft, or the CodeWhispers that is Amazon's version of that. And there are others, but also, of course, pixels. That's all that like these image, and I'm not trying to trivialize it at all, but a Midjourney and a Stable Diffusion and other such services are DALL-E. Are generating images based on pixel data that's been, again, been predicted in terms of what you're looking for, you know, from a generative image.

Yeah, that's a little bit of a no-brainer. Yeah, yeah. It's amazing to think about the ways that people who don't know how to write code can now query this data and ask it questions using their own— using English and using their own language. And that's really powerful. We've been talking to a lot of businesses who are sitting on a lot of data And there's a lot of people in these companies that would really want to ask it the right questions. And now they have an opportunity and it kind of democratizes things, don't you think?

Well, I think that it does democratize access to, you know, unlocking the power of the data. You know, it wasn't very many years ago, and I imagine this still happens in many companies where you have a set of, you know, whether it's sort of analytical questions or questions where you're actually looking for kind of insights and intelligence, and you have to sort of set up the questions and then go ask the people that are in the data warehouse group or data analysis group. And then they get— takes weeks and they get you some answers. And well, gosh, either that leads to new questions or it turns out we didn't quite ask the questions the right way. I mean, all this sort of high-friction, slow-turnaround process. So that in and of itself, I think it's dramatically improved. In a world of underlying much more flexible cloud-based data stores like a Snowflake or a Databricks, but also, you know, these new ways to use natural intuitive language to ask for something. And so, you know, your point on code is a really interesting one because in that case, you're actually helping people be more productive as code writers. And you'll notice that there's plenty of kind of AI skeptics. We could talk about that. I'm more on the AI optimist side of this equation, more Marc Andreessen than Elon Musk, I guess, if you want to frame it that way. And so, there is quite a bit of case to be made. You can look at some of the work that GitHub, Microsoft's done, Amazon's done already. Of showing the productivity gains with, hey, why don't I have, you know, my, you know, foundation model and my Copilot draft a bunch of the code? Of course, I'm going to then inspect it and edit it and complete it and refine it, you know, before I ever put anything into production. But the productivity gains that you're hearing about are in the, you know, 30 to 50%, you know, time efficiency types of pickup. And that's, You know, you talk about the future of work, you know, and you think about coders being a logical early adopter of this kind of a technology. We're really just getting started.

Yeah, I agree. We're seeing this in lots of different departments, customer success, coders, marketing. You mentioned VR/AR before. Does this feel, does this hype cycle feel different to you? I mean, we've been through a lot of hype cycles. Here in the last 5 years, but does this one feel different than some of these others?

Well, we do go through these hype cycles, and I think that this is a recurring challenge for the venture world, not just the investors like ourselves, but the founders. My college soccer coach, when we would run a drill poorly, in his Scottish accent would say, "Ah, lads, would you do the same thing, only different?" And what he meant by that was, you know, do it better and do it better now. Like, how can you do it better now? And so, you know, there were, you know, there was grid computing in the '90s. It didn't work. But cloud computing clearly worked. AWS is a $100 billion business today, you know. And so trying to understand why the same thing, only different, is going to work now. And part of that is because it's actually not the same thing. You learn things, you do things differently. but timing really turns out to matter. And so in the case of AI, I think we're benefiting from, once again, something that feels like an overnight success, but it's actually decades in the making. I think the thing that we and others got wrong in AR/VR— and by the way, the final story is not written there. I think a lot of what people call augmented reality or mixed reality is having transformative things. In a number of respects in different kind of companies and business models. We're investors in a company called Rec Room, for example, that is a really exciting— now, they started out saying we're going to be VR only. And one of the things they learned is that this kind of a next-generation Roblox, which is what basically Rec Room is, it would work better as a community if people could access it from any device. So rather than being just a VR headset orientation, you can go from your PC or your Xbox or your mobile phone and your VR/AR systems. And so I think you've seen that in different contexts. You've seen interesting industrial contexts where it's been successful, but it certainly has been slower going and taken longer. What's the neat thing about AI is that we are at this point with these new architectures and the pace at which those new architectures are being innovated on. We've got the series of GPT models from OpenAI. We've got these much, ironically, much more truly open source models, like Stable Diffusion, like what Cohere and Anthropic are working on. Certainly what what Meta has done, I mean, isn't it notable, right?

Yeah, Llama 2 is amazing.

Meta comes out with Llama 1, Llama 2. There's a little bit of folklore about kind of the release of Llama 1 out in the wild. But now with Llama 2, you not only see Meta putting out this thing that I think they're trying to create a de facto standard around, but you've got Microsoft and Amazon and OctoML, which is a kind of models as a service business, all embracing quickly that model and hosting it for you as a managed, you know, kind of model as a service type of thing. So this gives choice and selection to the people that are the builders that want to take advantage of these models, and they can figure out the combination of models, their own data, and their own use cases, you know, what level they want to build for themselves. Do they want to be a do-it-yourselfer, which takes a lot of pretty specialized skills? Do they need to be kind of preferring sort of a do-it-with-me? Kind of like my example of, hey, OctoML is going to host the model, but I'm going to then fine-tune the model or adapt the model with my data to what I need. And then in a lot of cases, I think you're going to still have people that are do-it-for-me. I mean, I think underappreciated, for example, about Apple is there's a whole lot of machine learning and AI models that live inside your iPhone. And you just take them for granted. They don't always work perfectly. Autocorrect and type completion, image, voice, Siri. I mean, I'm hoping now I'm not going to invoke it on my watch that is sitting on my wrist right now. But these are a lot of kinds of things that are already already out there. And so I think there's a lot that's been built up and there's a bit more of a kind of, I can incrementally adopt these things. And what's, I think, most interesting, just to add one more dimension to that, this is why deservedly so ChatGPT was quite a breakthrough. And OpenAI had been working on different kinds of natural intuitive interfaces for a couple of years, But when they came out with ChatGPT, and it wasn't perfect, of course, but it was quite impressive, they had something that, again, had this combination effect of really was accessible to almost anybody, back to your point about democratization. And it had this magical property of generating something, even just a, write me a 4-paragraph essay about the history of Seattle. And it came back with something pretty good.

Yeah, delightful. Yeah, it's really kind of, it feels magical when that happens. Even though it's just recommending the next word, it just feels magical that that is possible. The other thing that strikes me too is how much cloud is playing into this phase of AI that we're talking about now. I recently heard you talk about the new Silicon Valley partner and presence that you guys have. In the Bay Area. And, but I know that you feel strongly about Seattle's place in this conversation. So, and I also, as someone that's lived here for a while now, kind of see the future and see Seattle being a big part of that. Can you talk about our city, Microsoft, Amazon Cloud, how that, how the audience should be thinking about that as we think about this AI revolution?

Well, you know, I think you and I've talked about this too, you know, that Seattle really is the cloud capital of the world. And that's not just because of our winters, which have plenty of clouds. It's because we have, you know, clearly the number 1 and number 2 players in cloud innovation. And especially at that infrastructure layer, you know, all of the compute and the storage and the networking and the data stores and all the security and governance and all that enables that, you know, with AWS, you know, clearly the leader, even still today, clearly the leader. And then Microsoft doing an absolutely amazing job to be a fast follower and increasing innovator in a lot of different areas. What's sometimes also overlooked is there's close to 10,000 employees of both Google and Meta up here. And as we were just saying, Google's done some kind of amazing things and Meta's done some really interesting things, including kind of applying some of these kind of cloud-based, opportunities to AI with some of these open-source models. But we should make sure that Google gets good credit here as well. Areas like the tensor processing units are kind of very customized chips to do better kinds of AI and deep learning around. They've been innovators with BigQuery, which is kind of their competitor to Snowflake, and they've been evolving that to take advantage of AI. We already talked about the search algorithms, but I'll give you one more example. You know, Google's done some really interesting things with their group called DeepMind, which is more based in London, but it's got, you know, good ties here into Seattle as well. And one of the interesting ties is this whole area of designing proteins and understanding the structure of proteins. So, here we have this area in life sciences where a team in Seattle out of the University of Washington, the Institute of Protein Design, has been kind of collaborating and competing with the DeepMind team out of out of Google to build better and better predictions of the structure of proteins, proteins being the applications of life, basically. And so that brings us all the way back to Seattle where you've got these clear leading players in cloud, really running the whole gamut. Even Apple has a decent presence here. And even more importantly, like I was saying, 10 years ago, 10-plus years ago now, when we started investing, In applied AI, the insight was as more and more of the data was digitized and could be accessible to build recommender systems, predictive systems, and to do that in the cloud with your data so that you could build these kind of models and then deploy them and manage them at scale. And again, you look at all of those companies, Apple, and they have a center of excellence here in Seattle around Applied AI, partly because they bought 3 of our companies and then have built up that center. And so I think all of these businesses have, what you want to have is you want to have access to the ideas, the people, the experiments that are working and not working in the aggregate learning, you know, more robustly, more, you know, kind of with high, you know, kind of, you know, fidelity. And earlier than anybody else. And I think that's what makes, you know, the greater Seattle area, not only the cloud capital of the world, but I think it will be, if not the, you know, one of the very best, I think it should be, you know, kind of viewed as one of the very best in the world centers of excellence for AI.

It's really exciting. We're recording this episode at the end of Seattle Tech Week. And I've been going to a lot of events this week, some that you've sponsored, and others in the community. And I'm just struck by the energy that's out there. So, you know, we have a lot of big company energy, as you've just described, but the startups and the founders in this community are really on fire right now. And so let's talk a little bit about what you're excited about in some of the young companies, the startups, including those that are in your portfolio. I would just love to hear, you know, what your thoughts are about that.

Well, thanks on the Seattle Tech Week. It was kind of the first year for that. I think it really resonated with the community. I think some of us that were involved with kind of creating it were like, wow, if we can have 15 events and then we had like 45 events and to your point, you know, hundreds and, you know, overall in aggregate, thousands and thousands of people turning out for them. And so it is a very robust ecosystem here in Seattle. And, you know, I think it is because there's a lot of startup energy around what can be possible. In a world that's going to have such profound disruption through data. I mean, there's an element of it, which is really the next-generation data stack, which is, of course, super kind of intertwined with applied ML and AI but is different. And so, you look to— a lot of that BigQuery team from Google was based here in Seattle and has led to a company called Motherduck, for example, that's a really a really promising company. And you've got other companies that are working with, you know, kind of data and data warehousing techniques, data stores, data ingestion techniques here. And we are excited about that. But I think you just keep coming back to either directly or in a supporting role, how a lot of these companies are gonna help with the embrace of applied AI. And let me give you a kind of an example that may be a little surprising. So we've got a company called Statsig, once again based here, came out of a team that was at Facebook, an exceptional team out of Facebook Meta. And they really help you with kind of experimenting with your applications to make your applications better, you know, A/B testing if you want to think of it that way. But, you know, kind of a very robust suite of things that you can do. Well, lo and behold, a lot of their most interesting and bigger customers now are companies like OpenAI. And why? Well, ChatGPT has all kinds of knobs and variables that they can play with to get more engagement, more usage, upselling people into the paid version of ChatGPT. Take concepts like temperature, which is an applied AI concept for How much determinism versus predictiveness capabilities do you want built into your model? Or even what kinds of the different variations of the underlying GPT models are you going to use as a mix in how you deliver the ChatGPT service? And how does that impact, again, results and usage and monetization? A company like Statsig that wasn't initially founded 3 years back to focus on applied AI scenarios has learned their way into that area as an extension of what's already a very good business. We call that generative enhanced scenarios, where in this case, a company that was doing one thing is kind of enhancing other capabilities and extending into the applied AI world. There's also a lot of software companies, software as a service companies. And again, you think about all the great ones up here in Seattle beyond Microsoft and Amazon. You know, think about the DocuSigns of the world or the Smartsheets of the world or the Appdios of the world. You know, all these companies are saying, how do we leverage the context, the data, the workflows that we have and apply generative AI in compelling ways to solve our customers' problems better. I mean, I'm on the board of Smartsheet and it's a public company, it's approaching $1 billion in revenue and they're doing very interesting things to create kind of an AI assistant for the teams that are using them for collaborative work management in the enterprise and getting some very exciting early and promising feedback around how they're enhancing their application embedding some of these capabilities into it. Now, in the startup world, what's also true is that there's companies that are building what we call, you know, gen-native, you know, services.

So yeah, tell me more about that.

So a gen-native service would, would be a company that was built first and foremost understanding that the really the future of software isn't really software at all. I like to call it dataware. Because you're going to have models that are trained with data that are embedded into your software that make predictions and recommendations, inferences, and generate things. And that's just gonna be native to the service, the software, if you have to call it that, that has been built. And so companies like RunwayML would be a good example of that where they can help you create and edit and refine your videos all as a gen-native solution, which at some level is an alternative to an Adobe where Adobe is having to live with, you know, kind of, you know, and doing a nice job of it, I might add, with their Firefly offering and all, you know, they're trying to, you know, enhance their legacy non-AI native, you know, solutions with applied AI, you know, or a company in our portfolio called Read.ai. You know, what Reed does is they help you get instant meeting recaps. So, you know, you could, you know, you just say, say, I want Reed AI to be participant in all of my Zoom meetings or my Teams meetings. And I get these instant meeting recaps with here are the 3 main topics. Here are the couple of key questions. Here are the action items. Well, how are they doing that? Well, part of what they're doing there is they're using multiple models of transcription, and then they're creating their own model of transcription that optimally can create these instant meeting recaps and make it effortless for you or me, you know, who wants to be able to, you know, just get the summary view of the meeting that you just had. Again, back to the future of work. Think about how much more productive I can be. I could have even missed that meeting potentially, you know, if I had another conflict or another thing that was going on. But certainly, I mean, I'm always trying to figure out how to most productively follow up from meetings. And this service is, you know, doing it, you know, helping do it for me.

Yeah, that's amazing. I mean, all the meetings that we're doing now post-pandemic, it seems like that's accelerated and the ability for us to have summaries of those meetings, the ones that we're in and the ones that the assistant went to the meeting for us is, is just amazing. It's really amazing. You know, it kind of makes me think about some of the hybrid work situations that we're finding ourselves in. And I think it's an interesting, you know, this is about the future of work. Since the pandemic, all of us are now having work meetings in a very, very different way than we had pre-pandemic. And it's really changed the way that companies are able to go out and get talent. But now I'm finding that a lot of founders want to be back together. Again and want to be innovating together again. So I'm seeing this kind of hybrid model of startups, you know, founders are in the early days coming together in the garage, quote unquote, but then having the ability to hire in other regions that, you know, in the past they may not have done that before. So maybe we can talk a little bit about hybrid work because I think the future of work is also going to be about that as well.

Yeah, well, I think that I think this is where we know some of those why now questions. One of the big why nows is there's been a massive societal change. I think a lot of that was from COVID and the reality that we went from largely a world which was physical first in terms of where and how we met and worked to digital only out of necessity for a period of time. And that's actually, I think, an easier shift, as hard as it was, than what we're trying to kind of work through now, which is this world of hybrid work. Because now you have this need to have mechanisms so that we can be coordinated in different ways. And one of those mechanisms, of course, is things like Zoom and Teams and Google Meet and others. I think we all find that to be less than fulfilling if that's the only mechanism we have. And even companies that were pioneers like GitLab in, you know, so-called, you know, kind of a distributed-first running of their organizations, they have different mechanisms for how do we meet up in person. And so, what do I mean by that? Well, you know, I think of this concept of XBRs. So, an X could be a day. Daily business review, well, that could be a daily standup. That might be something that's just fine for those daily business reviews to be done through Zoom, right? But what about quarterly business reviews? And do we have an expectation and a practice that those are going to be done in person and those are going to be maybe, you know, a little longer in time because in a hybrid company, not everybody gets to see everybody all the time. And so, those quarterly business reviews or maybe even you do them in greater frequency, monthly business reviews, need to be in person. And then that doesn't presume that people are going to be all distributed, right? I think that we're finding in many, many situations, both in startups and in bigger companies, that having some kind of weekly cadence around, you know, 2 or 3 days in the office, ideally the same 2 or 3 days in the office. And so that, back to again, like creating kind of both a cultural and, you know, and a process set of expectations really helps you with certain kinds of work, the brainstorming kind of work, the prioritization kind of work, the kind of decision-making kind of work often is much better done synchronously in person, ideally with most everybody, if not everybody, physically in the same rooms. There's this Amazonian expression of disagreeing and commit. Well, it's possible to do those kinds of deep discussions and spirited disagreements that ultimately need to align around committing to a direction, you know, in a hybrid or distributed way. But I think those are generally a lot better in person. So then you got to say, well, but yet the expectations have changed. So what can I do while still getting to the results that we want, you know, both on a functional level and across functions in companies, you know? And so setting expectations, I think this is a very important role from not only senior leaders setting those expectations, but living by that themselves. I actually have kids in their 20s. They want to be in the office, but they want the senior people in the office, which was some of the magic of learning from each other and having that mentorship kind of relationship. So, if your senior folks aren't leading by their own actions of being in the office frequently, I think that makes it a lot harder. And then finally, is, I think, going to be an evolution in a whole other area, which is physical space. And, you know, are you going to have more of an office space that's kind of got a work element and a hoteling element to it? So it's easier to have these XBRs with more frequency. I don't think we've really begun to see how that— because, you know, physical space, obviously, always harder to, to innovate around. And I think that's going to, you know, potentially play out here over the next, you know, 3 to 5 to 7 years.

I like the way that you think about that. I mean, work is changing for so many people and the way that they connect with each other is changing. When they show up with each other is changing. And, you know, and with AI, the work itself is changing. And that's— we talked about ethics at the beginning. We already touched on it a little bit, but I think now is a good time to kind of go deeper on that. I've been reading a lot about a world where maybe there's not as much work. You know, what does that look like when the computers are doing a lot more of the work? Or, you know, maybe we spend more time being creative and we're doing different kind of work and the computer's doing more of the mundane tasks. But I also think it's an opportunity for us to talk about just the ethics in general around what happens when computers start doing things for us. I really believe that humans should be in the loop always, but there has been a lot of talk about sending computers off to do things. And then, you know, there's some issues around that. And can you talk a little bit about some of the ethics that we should be thinking about?

Well, I think you— I'm happy to. And I think it does start with, you know, what are your expectations of how things are likely to evolve? I happen to agree with you. The humans are always going to be in the loop. But I also expect that the machines are going to be able to do certain functions and certain tasks, not only as well as humans, but in some cases better. An area of applied AI that's very near and dear to my heart, both having been on the Fred Hutch board for many years and increasingly things that we're investing in is how do I use AI-trained models to do better jobs at detecting cancers? And, and it's not just cancer, not cancer, but it's what kinds of proteins are on these cells, which distinguish them as a specific type or flavor of a cancer, which will lead to a different prescription, a different strategy for treating that cancer. Yeah. And so I, not only do I want that to be more efficient and more accurate in America, but candidly, where there are many places in some countries where there's very few oncologists. And can we take those kinds of capabilities and make them way more accessible around the world? And so I, I think that what this does lead to is it leads to hopefully better decision-making. It also leads to the evolution of what a, an oncology radiation oncologist or a pathologist or others is gonna do. I don't think because similarly in their case, there's a lot more precision medicine to figure out exactly what's the best treatment for patient A or patient B. So that I think gets to your point of they're gonna need the time to have the sort of the creativity and the insights and the judgment to say, okay, which of these different precision medicine options is actually gonna be best for, you know, the patient A, patient B, patient C. I don't think we're gonna run out of jobs to be done by humans. And I know that it's, nervewracking, right, when some of the jobs that have been around for a long time that have employed a lot of people are going to evolve and in some cases go away. But to your point about humans in the loop, and there's so many places where the humans will be in the loop, again, and this is a matter of belief. And so then you've got to say, well, where do the ethics come in, right? And I think there's multiple levels where they apply. One is to the extent that jobs are evolving, you know, are we providing, you know, you know, as society, and are we providing this kind of individuals and the people that we know and care for, you know, kind of the motivation to be able and equipped and being equipped to go learn new skills and to evolve into new jobs and new opportunities so they can have the meaning and dignity of being, you know, you know, gainfully employed. I think there are some interesting questions. Candidly, I think they're a bit over hyped around content and derivative content. I kind of take a little bit the Picasso view here, which is that all art derives at some level from other prior art. Yes, is there kind of a spark of creativity and innovation? But Picasso, who was one of the most innovative artists of all time, believed that he was borrowing from the inspirations of the past. And I think that all artists do. I think people that do create innovative content need to be treated fairly if others violate their— actually do violate their copyright or other intellectual property rights. So don't get me wrong, but we have laws for that today. And so those laws will ultimately need to be adapted. The regulations will need to be adapted. And that's the ethically right thing to do. And then you could go down a bunch of other paths. And I'll just mention one more, which is just the ethics of when do the humans need to be in the loop? And the example people give is good old, I think it actually was set in Seattle, this movie 40 years ago that maybe inadvertently we were triggering nuclear wars. I think it was WarGames, I think it was called.

Yeah, yeah, yeah.

Well, ultimately, if we fully empower the machines to make the call, we are going to put society at risk. So, the ethics there are what are the mechanisms that say that there's still human judgment, you know, in this case, at the most senior levels of the government and in the military, so that we are confident in the information that we have and can actually make the decisions that hopefully will allow society to last and thrive for a long, long, long time to come.

Well, I appreciate that approach, and I agree with you. I think that I want to see humans stay in the loop. Both at the beginning of having, you know, if you have an AI agent, I want to be there at the beginning to help set the tone for what that agent is going to do. And I want to be there on the other side too, to check and make sure before anything goes out the door. But you touched on this before, as we get close to the end of our conversation about our relationships with other humans and mentorship. And I just, I want to know the folks in your life that have been mentors that have been impactful to you. And I'm sure you've mentored others, but who in your life has been a real mentor?

Well, I've been fortunate going back to early in my career to have a number of terrific mentors, you know, and all those different companies, including, you know, like Genuine Parts, which was a holding company with a lot of different businesses that were threatened by this thing called the internet back in the mid-'90s. And Larry Prince and Tom Gallagher were the CEO and president of that company. They took on this young whippersnapper McKinsey kid to try to help navigate all that. And I learned so much about how you treat people with respect and honor, even in the midst of technological disruption and change and how you think about running a business for the long term by them. But I think I'd have to say that my most impactful mentor in the last couple of decades is Tom Ahlberg. And Tom was the co-founder of Madrona. He also had led that very early round of Amazon, was on the board for over 20 years there. And his office was right next to mine. And so, I would spend a lot of time listening to Tom, who was very soft-spoken, but he had an incredible curiosity for many, many things. And he had this great long-term vision. The things in our ecosystem in Seattle from the Technology Alliance to Oxbow Farms to Novelty Hill Winery, I could go down this incredibly long list of things that Tom had his hands in, but he never sought the credit for them. He always was just trying to kind of in a quality, curious, quiet way make an impact on the world. He was a great mentor on so many different fronts. Very honored to, you know, for him and the other founders, you know, people like Paul Goodrich and Jerry Greenstein and others of Medtronic, you know, over the years.

Yeah, that's amazing. Well, when you think about the next 5 years and, and maybe, you know, the conversations that you've had with Tom over the years around emerging technology in the future, what are the things that do you think are going to be the most impactful looking into the future?

Well, I hope we don't have it wrong on this kind of this GenAI front, or more broadly, kind of applied AI. But that really is the major technological driving force of the next decade. I think the modern data stack is a big contributor. And I think one of the manifestations of all that, that's just early and is starting to get more conversation, but it's this notion of personalized agents. Right now, we're thinking about large language models as these monoliths. I think soon we're going to be hearing more about kind of domain-specific models. We talked about proteins, for example, but what about financial services? What about creative areas and marketing? But I don't think we're too long off from having these customized personal agents that are not only relevant in our individual life, I want a buyer's agent that helps me plan my travel better or buy a car better or whatever that might be, but also in our business lives. And I think you're seeing lots of experimentation coming right now. It's too early to know what's going to emerge. But I think that it's not just going to be the model side, but it's going to be the agent side and the agents often at the edge, you know, on our phones, on our devices that will become much more pervasive in the, in the years ahead.

That's exciting. Well, I always like to finish the show asking you to describe the future of work in 2 words. You can elaborate a little bit, but how would you describe the future of work in only 2 words?

Gosh, that's a great question. I would say creative collaboration. And the creative part, as you said, there's going to be more things that are more kind of the, the heavy lifting basic work that we can let others do or let really the machines do. And so, we're going to have more opportunity to be creative. And I think in an incredibly networked world and even in a hybrid world, the power of kind of collaborative work and collaborative work being unlocked is greater than it ever has been. And so, I think creative collaboration would be my two words.

I love it. Well, these are very exciting times, Matt. Thank you for taking the time talking to us today. It was really, really wonderful spending time with you.

Thanks so much, Boaz. Really enjoyed the conversation.

That's a wrap. Thanks for tuning in. Such a pleasure to have Matt as our guest on today's episode. Matt's knowledge and experience growing successful businesses from zero to one and his insights into how AI is transforming our lives sets the stage for how emerging technology is changing the way that we work. If you want to learn more about Matt, you can find him on LinkedIn and Twitter. If you're interested in how he thinks about venture investing, definitely connect with him and keep up with his posts and activity online. I continue to be amazed by the guests we've had on the show, and I'm excited about the ones joining us in the near future. I truly appreciate you spending your time with us. Thank you for listening to this episode, and don't forget to subscribe to the podcast for more exciting interviews and to the Simply AI Weekly newsletter for updates on how AI can help you scale your business. The Shift Podcast is produced and sponsored by Simply Augmented, and our theme music was created by Dave Angel.