Gen AI Series: Intelligent Applications in an Agentic World, A conversation with Sabrina Wu, Investor at Madrona Ventures

In this episode, we talk to Sabrina Wu who is an Investor at the famed VC firm Madrona Ventures. Sabrina shares her thoughts about closed source vs open source, agentic applications and the emergence of multi modal applications. 

Sabrina joined Madrona’s investment team in 2021 and focuses on sourcing and evaluating new investment opportunities from Seed through Series C across Madrona’s core investment themes. She also supports the growth and strategy of current portfolio companies. Sabrina is particularly interested in intelligent applications, enterprise software, and data/AI products.

Sabrina and Madrona Partner Vivek Ramaswami author the biweekly newsletter Aspiring for Intelligence where they share thoughts about the future of intelligent applications.

Brought to you by TiE Seattle
Hosts: Shirish Nadkarni and Gowri Shankar
Producers: Minee Verma and Eesha Jain
YouTube Channel: https://www.youtube.com/@fromstartuptoexitpodcast

Transcript

SPEAKER_03 0:00

Our uh listener viewer base has a lot of aspiring entrepreneurs. What advice would you give them?

SPEAKER_01 0:06

The founder market fit part is really important. Why are you the right founder to be solving this problem? What makes you passionate about this space or what from your background lends you to believe that your way of thinking will be the future is one thing that I would continue to think about. The second is you know just this idea of why now? Like what has changed from a technological perspective that enables you to build this application or or maybe you're building something at the infrastructure middleware layer. Like, why is it relevant if you think about the next 10 years? Why is what you're doing today relevant for the future?

SPEAKER_00 0:40

Welcome to the Startup to Exit podcast, where we will bring you world-class entrepreneurs and VCs to share their hard-earned success stories and secrets. This podcast has been brought to you by TIE Seattle. TIE is a global nonprofit that focuses on fostering entrepreneurship. TIE Seattle offers a great program.org. We encourage you to become a tie member. Get access to become a member.ti.org.

SPEAKER_03 1:22

Hello, everybody. Welcome to another episode of our podcast from Startup to Exit. My name is Gary Shankar. I'm here with my co-host Sherish Natkarni. The podcast is produced by Thai Seattle, a not-for-profit aimed at fostering entrepreneurship. Uh we both are board members of Thai Seattle. Shirish uh is and I live in Seattle and is an author, actually author of two books already. And his first book, from which we borrow the name for the podcast, from Startup Exit, is a go-to guide for every entrepreneur that's uh thinking of jumping into the startup world. Uh we like to thank all of you for supporting us over the last two years. Please share, please subscribe, and uh hopefully you're enjoying us on uh YouTube where uh the video is also available. Uh thanks a lot, and uh look forward to hearing from you more through comments and uh enjoy another great episode of our podcast from Startup to Exit. Now, a word from our sponsor. At JP Morgan Private Bank, wealth is understood to be more than just numbers. It's about creating a legacy, achieving dreams, and securing a family future. With over 3.1 trillion in client assets under management globally, JP Morgan Private Bank is committed to providing customized financial advice that aligns with each client's unique goals. Clients benefit from a personalized approach, working closely with experts in philanthropy, family office management, fiduciary services, and special advisory services. The firm emphasizes building lasting relationships and ensuring that every financial strategy evolves with the needs of clients and their families. JP Morgan Private Bank not only manages wealth, but also helps clients create a meaningful impact. Whether individuals seek to support charitable causes, manage a family legacy, explore new investment opportunities, the private bank team is there to provide guidance every step of the way. With the presence of key financial markets worldwide, JP Morgan Private Bank offers unparalleled access to global insights and opportunities. Its commitment to excellence and innovation ensures that clients receive the highest level of service and expertise. Time is recognized as precious, and JP Morgan Private Bank values every moment spent with its clients. The mission is to make the financial journey as seamless and rewarding as possible from the initial consultation to the ongoing management of assets. To learn how JPMorgan Private Bank can help clients achieve their goals, visit privatebank.jpmorgan Connect with a dedicated team of specialist. Again, privatebank.jpmorgan.com. Thank you.

SPEAKER_02 4:38

Welcome everyone. Please welcome Sabrina Wu from Madonna Ventures, where she's an investor. Sabrina joined Madonna's investment team in 2021 and focuses on sourcing new investment opportunities from C to Series 3. Sabrina is also particularly interested in intelligent applications, enterprise software, and data AI products. She and Madonna partner Vivek Ramaswamy author a bi-weekly newsletter called Aspiring for Intelligence, which I'm a subscriber to as well. And I would highly, highly recommend this newsletter. It's a great way to keep in touch with what's happening in the industry, in particular with AI. So welcome, Sabrina.

SPEAKER_01 5:19

Thanks so much for having me. Yeah, really great to be here.

SPEAKER_02 5:21

Great. So let's start with your background. You were at Vector Capital before joining Madrona and then joined Madrona Ventures in 2021. So tell us a little bit about your journey.

SPEAKER_01 5:31

Yeah. So as you mentioned, prior to joining Madrona in 2021, I spent uh about three and a half years at Vector Capital. Vector is a later stage private and growth equity firm. During my time there, I was primarily focused on investing across kind of the B2B technology landscape. And many of the companies that I looked at there were, you know, once kind of VC-backed companies. They were the darling angels of the time. They had maybe found product market fit within OneWedge, but really struggled to expand kind of beyond that. So many of those companies got to, you know, call it 50 million or so of ARR. Sometimes they were growing, sometimes they were declining, mix of profitability and burning capital, and really just needed somebody to come in and help kind of with the next stage of growth. And so that's those are the types of businesses that I tended to look at prior to joining Madrona. And, you know, we also looked at much later stage companies that had been around for, you know, call it 10 plus years. But I found myself, you know, always really excited about two things, which was tech, product, and kind of future vision. And that's when I ultimately decided, you know, I wanted to move earlier stage and do more early stage investing because you get the opportunity to partner with founders at the early stage, help them ideate, help them incubate, and really kind of roll up your sleeves and think about what's the future of technology and how is that going to change the world. And so joined Madrona in 2021. Um, one of the investors here. I spend most of my time investing across B2B software, specifically focused mostly around applied AIML and up and down the stack, but a lot of my time is spent mostly at the application layers. You know, it's been a really, uh really, really fun journey so far. I joined um, you know, just before chat, I guess it was over a year before ChatGPT launched. Um, and then, you know, um when the Chat GPT unlock moment happened, it's been really interesting to see kind of the juxtaposition pre-that and post-that and how how much um capital has now gone into AI investing. So yeah, that's a little bit of the of the journey for me from Vector to Madrona.

SPEAKER_02 7:27

Great. So we'll focus our discussion primarily on AI since you write a lot about it in your newsletter. Uh one of the things that has um become apparent is that the competition for LLM foundation models has heated up uh with new models from DeepSeek and of course Meta and Google and others. Um what is your sense for how these models compare with OpenAI's model? And um, you know, which models do you see your portfolio companies primarily using?

SPEAKER_01 7:58

Yeah, so I'd say one thing just to maybe level set. I think over the last year, we've really seen a step function change with all the models. You know, you look across all of them, um, both open source models and closed source models have gotten a lot faster, they've gotten a lot more efficient, and they've gotten a lot better and they've gotten cheaper. And we see that with Deep Seek. I think that was a big the big news not too long ago, just about how how the model was performing on much cheaper costs. I mean, there's a lot you can dig into, kind of what was the last training run, et cetera. But all these models have gotten a lot better in the last year. And I think it's really impressive just the step function change that has occurred. I think within our portfolio companies, um, we're still seeing them use a lot of the open aim, open AI models and the anthropic models. Those are kind of the two that I would say continuously come up. Anthropic Claude models have been um probably more performant on the coding tasks and some of the reasoning tasks. Um for so depending on what specifically you're building, what application you're building, um, a lot of our founders really love using those models. And then um with OpenAI, I would say GPT-4.0, um, some of the um and other, you know, if you're doing reasoning tasks or other um multimodality um application use cases, people really love using open AI models for that. Um, I would say with Deep Seek, Meta, Meta's uh Llama 3 and Um and Mistraw, I would say people are tinkering with those models for sure, um, but they still take a lot to actually get them to be performant and get them to be usable in in certain settings. And it just takes more to get um to get the model to get to a place that you want it to be as performant relative to out of the box using um open AI models or anthropic models. So that's kind of how I see the landscape. But it's I think that you know it's constantly evolving and changing. And so it would be curious to see how um how this evolves in the in the next year or so. But I think the models itself have all just become really performant and you can use them at you know relatively cheap costs.

SPEAKER_02 9:59

Yeah, there have been a lot of predictions that uh the open source models will catch up uh with the closed source uh models like OpenAI and Anthropic and so forth. Uh, do you share that uh prediction?

SPEAKER_01 10:13

I think that the the there's definitely a chance that that happens. Yes. I think that um, you know, at some point there is a point in which all the models have a baseline level of performance or capabilities. And then to actually get to the last mile, you will have to optimize the models in some form or fashion, be that, you know, for your the application use case that you're using um or trying to solve for. So so I think it really depends on what you're benchmarking it to, for example, or what types of tasks you're doing. But I do think that the open source models are getting better and stronger and they will continue to um to do so in the next year. Got it.

SPEAKER_02 10:50

And I assume the uh costs, the costs have gone down quite a bit and they'll continue to go down um over time. Is that your prediction as well?

SPEAKER_01 10:59

I uh yeah, correct. I believe that will continue to happen. And I think that in in a lot of ways, these models will become commodities. Um and then, you know, the the value will be derived um for for how you um build the applications on top and how you create a delightful experience and how you fine-tune it to do a very specific task or how you you know get the model to to retrieve certain sort of, you know, uh do it like we're I think we'll talk a little bit more about like agents, right? But I think that the we can the models at a baseline will become a commodity. And then the real value will be what's what's layered on top at the app layer.

SPEAKER_02 11:35

Yeah. So let's talk about agents. One of your predictions uh for 2025 uh was that it'll be the year of agents. We've we've actually spoken to two companies already uh who have built agent applications. I'd be interested in examples that you might be able to share that you've seen in your portfolio companies uh where you've been particularly impressed with their capabilities.

SPEAKER_01 11:55

Yeah. So I think AI agents, it's really interesting. I think this, you know, there's been a lot of discussion about it in the last year. This year we're starting to see even more buzz, more activity around it for sure. I think that agents fundamentally can help you with two things. One is as a company, it's one, you know, driving top line revenue. And the second one is helping with cutting costs, right? And so you can think about agents as either a coworker, they they work with you or alongside you to help you achieve a certain task, or you can almost think of them as essentially an employee, somebody that you give work to, and they go and in the background are doing that task, completing that task autonomously for you, right? And so that's kind of how I, you know, think about agents. I think that a lot of people have different definitions of them and there's different types of agents, and there's autonomous, fully autonomous, semi-autonomous. So there's lots that you can dig into there. Um, but I'd say within you know what I'm seeing, there's a couple of things. The first um is around the vibe coding space. I think that's been talked about a lot of of recent, but basically this idea that um, you know, these agentic AI tools, you can ask them to um spin up an application for you in natural language. So me, you know, as a non-technical person, I can go to a Vercel V0 or a cursor or Bolt. Uh, Bolt is one of Madrona's investments, um, and and you know, ask it to create, you know, call it like a fitness app for me that tracks my meals, tells me how much I should be working out based off of, you know, trying to achieve certain fitness goals and objectives, or maybe I'm trying to run a marathon. How do I create an app that helps me do do that in a certain time? And it will go and spin up that application for me. Um, not only will it um create kind of like the UI UX layer, it also sets up the back end. It's it's it's a full stack application that you can um have um in coded in any language that you want. Um and it's it's all done for you in natural language. And so I think that's one thing that agents are really great at doing in the code gen space. You're seeing that a lot today. And that's probably one of the bigger use cases that we're seeing. Um in the B2B setting, um, you know, seeing a lot more vertical and horizontal use cases of AI agents. Um, one company um I call out in the more go-to-market space is a company called Clarify. It's a next-gen agentic CRM system. It really unifies all your customer data and connects into different data sources, systems like your email inbox and also different sales tools that you have and creates a unified um view of your customer. And um, you know, what it does is it abstracts away a lot of the tedious manual tasks that sales teams are usually tasked with doing, be that uploading um a new lead or understanding, like getting data entry into the CRM system, trying to figure out who's connected to this person. And then also just doing things like you know, automatically doing follow-ups. They're all done by the agent itself. So the agent understands all the context and it's doing that in the background for you. And so really it's it's the it's this idea again of abstracting a lot of the low-level tasks and allowing the humans to work on more analytical and important things. And I'd say the agents really act as a, like I mentioned in the beginning, either a co-pilot or a basically another employee that you can go and task it to do things um for you.

SPEAKER_02 15:05

Uh, do you see that there'll be a um transition from fully autonomous from semi-autonomous autonomous to fully autonomous over time, uh, just because people have to feel comfortable that the agent is going to perform the task properly and so forth?

SPEAKER_01 15:21

Yeah, I think we're still in the early innings of that. And I think it'll really depend on the use case. Like it depends on how customer-facing it is or how much sensitivity there is around it, or how much access you're giving the agent to do, if it's something more simple or if it's something more complex. Um, and I think that there's there's issues today where we're not fully at a at a spot where the agents itself can reason. It doesn't have the best memory to understand specifically how to what it should remember to do or what you've told it previously, like long context memory windows, like those sort of things still block the agent from being fully autonomous without involving the human in a way that it's thinking like the human would or prioritizing like the human would. But I think once we get there, it will be able to go and do tasks fully autonomously for certain types of uh of work. Um but I think that we're still pretty early in that and we still need to see some more infrastructure unlock, if you will, for the agents to go and actually execute on a lot of the tasks that they're that they're given.

SPEAKER_02 16:19

All right. One of the other things that you've written about is multimodal applications and that you believe that voice will become the primary interface. Uh uh, can you give some examples? Have you seen any examples so far of some interesting multimodal applications?

SPEAKER_01 16:33

Yeah, so I think that voice will be one of the interfaces. I think that we are moving to a world where multimodality is critically important, and we're starting to see that with a bunch of applications that are in the market today. I think the Chat GPT unlock was a big moment for people because they were able to interact with these systems. Um, but I don't think that text is the only form of communication for everything. And so I think that we're really at this, at this the start of a shifting point of how UIUX will be and the experience of what UIUX should be in an energetic world. So I think that, you know, first and foremost, I believe we should have more personalized experiences. So the idea of going onto a website or logging into an application, it shouldn't be a static experience that you and I have the same landing experience. It should be very dynamic. So based off how we interact with the system, based off of our preferences of interaction, maybe that's voice, maybe that's text, or maybe that's, hey, I want to see an image. Um the application itself should should spit out something different for me. And so I think we're starting to see that. A lot of applications are embedding voice, for example, as uh as one modality where you can talk with the system. You don't just have to chat with it in the in the natural setting. Um, but I'm seeing that with companies, you know, like companies like Character AI, for example, are a great example of multimodality. You can have an interactive discussion with a chatbot-like person, or you can have a real-time conversation with an avatar-like persona, and that emodes a different kind of um experience that brings it more to life. They can understand how you're reacting, and you can have this kind of conversational video interface with with um with somebody else. And so I think that we're going to start to see more, more of this in the recruiting space, for example. You're starting to see companies like Mercore that have AI avatars that are um emulate kind of human behaviors and can understand, you know, when I'm more emotive or less emotive, what does that mean? Um, how can I get them to be more engaging, less engaging? So I think these sort of, you know, um multimodalities will be embedded more in applications. It'll be voice, it'll be more of this kind of AI avatar conversational experience, and it'll also be text and image. So it won't just be one modality, but it should be kind of a mix of all, though, similar to just how you and I interact on an everyday basis. It's not always chat or text. It should be all of the above. So I think that we're we're starting to move more and more in that direction. Great.

SPEAKER_02 18:51

That sounds that sounds uh quite right to me. Yeah. Uh let me turn it over to uh Gauri. Uh he has some more questions around uh uh vertical and horizontal applications of AI.

SPEAKER_03 19:01

Uh thanks, Sherish. Uh I echo Sherish's thought uh to all our subscribers and viewers, please uh read the newsletter that Sabrina co-authors with the Veg. It's a must-read. Uh great uh great material. So thank you, Sabrina, for uh keeping us well informed through the newsletter. Um so let me start with uh uh you touched on one thing, which is uh as you look at uh investments today, um it's pretty clear that there's the infrastructure layer or the silicon layer and everybody else. There's the foundation layer, which is uh now dominated by a few, some deep seeks and others that may come about. And then there's the intelligent application layer that you really talk about and have extensively talked about. And uh so go first going to that, right? The in the intelligent application layer, originally everybody thought, hey, you're building a wrap around ChatGPT or something like that. Turns out you needed workflow or operational expertise to exploit the foundational models, right? And so when you evaluate investments today, how do you separate where incumbents they're not going away? You know, you talked about a CRM application, Salesforce isn't going away, or any other applications, right? Uh they're all not going away. They're also investing continuously in their own capabilities around AI, whether their own models or other models. How do you uh look at uh uh a team and say, hey, they're on to something that could actually build value both as an investment and as a solution to the enterprise that they are uh that they are uh selling into?

SPEAKER_01 20:50

Yeah, I think that's a that's a good question. I think a lot of it comes down to do they have first principles thinking. So I think the one of the obviously there's a lot of advantages that the incumbents have. They have distribution advantage, they have customers, they have a lot more capital, they've been around, they've built that trust with their customer base. And they and they have a lot of um uh investment of ability to go in and and and build out new features and products and functionality. Um, but I think what the incumbents sometimes lack is this ability to think, um, think kind of on a clean slate, if you will, right? And so I think that the first thing that we look for is how is the founder thinking about the problem fundamentally? Is there some difference in how they think about it from building this product bottoms up that you couldn't have done before would be very hard to rip out legacy systems as you know once systems are in place and you've set up an infrastructure in a certain way? It's not as easy to just plug in or swap in different models, right? And so thinking about it architecturally, is it is there some differentiated approach that they have? Have they thought about the problem? And are they using AI in a way that's different, differentiated? That's one. I think um and then from there it's just how how fast. Are they moving? Are they using these AI tools to become more efficient? I think that's one of the advantages that many startups have is that they're actually using a lot of these tools in ways that incumbents aren't or don't know how to. And they're, you know, just seeing a lot more advantage from trying new new things. So that's kind of what I'd say we we look for in when we're evaluating some of these early stage, early stage companies.

SPEAKER_03 22:22

So uh just to build on that, right? Madronas are always advocated uh not only product market but founder fit, right? This is a principle that they have ever since I've known Madrona for a long, long, long time. They've they've really zoned in on is this the founder to solve this problem at this time? Um when currently you mentioned cursor, right? They zoomed into 100 million ARR with a small team and and uh it's a younger team to have solved the problem they solved, right? So is that changing when a founder comes in because they have this infrastructure advantage that they can um that they can navigate, especially in enterprise, uh a workflow or a solution? That is that uh is that being challenged? Is this the founder for the right solution? Because in the SaaS model, you needed that knowledge to really solve that problem. Today that seems like the LLMs could give some advantage to the founders. Are you seeing that? How are you uh evaluating a founder in this current setup?

SPEAKER_01 23:29

Yeah, so I think founder market fit is still really important here because I think that you still fundamentally have to be passionate about the problem that you're solving, and there has to be a reason in which you are trying to solve that problem because again, it's it's always going to be a long journey as a startup founder if you just decide to build something that's not something you're passionate about. I think that comes through when times are challenging. I do think there's advantages to building today that founders, you know, call it five, 10 years ago, didn't have, right? Like if you're a non-technical founder, you could put together a mock-up using cursor or bolt or one of these different things and have a V1 of a product faster than you ever could before. So I think there's advantages to building in the modern day, but I think this notion of founder market fit is still extremely important. Um, understanding the problem space that you're going after, understanding why you're the best founder to tackle the problem, understanding kind of what experiences you had that led you to believe that you have a differentiated way of solving the problem. You know, I think Cursor's example is a really great one too, of just uh how an incumbent can't, or sorry, how a gen native company can beat out incumbents like GitHub, even right. And GitHub was around for a really long time with their um uh co-pilot product out in the market that had a lot of adoption, but cursor fundamentally understood what the developer was looking for, what the modern developer that coded in in new languages wanted to see and the interface in which the product should be. And they just took a first principles approach thinking around that and made a very, very delightful experience. And that's that is very hard to do. So um, yeah, I think this idea of founder market fit is still very relevant, but I do think that there's some things that have changed, you know, from call it five, five, ten years ago that these founders can tools the founders can now leverage.

SPEAKER_03 25:09

So so just kind of building on that, so they built something horizontal, pretty much every developer and every um uh you know vertical could use. So are you seeing the differences where when uh founders come to you that there's a tendency to build a lot more vertical because you could, you know, you could do something that's very clear in its proposition to the solution versus horizontal, which is hey, I have a multimodal application, I could use it for recruiting, or I could use it for debt collection. Both are you know two extremely varied, but both are talking to humans and understanding emotions. The horizontal could be applied around either vertical, but I could build a recruiting only vertical AI application and come to you. Are you seeing any shifts today uh in terms of how founders approach it as models get better and better at uh things that they uh are doing between their various gens?

SPEAKER_01 26:09

So I think that there's both vertical and horizontal applications being built in the market. What I would say that I'm seeing more of is founders starting with a specific wedge. So even though you could be building a horizontal product, call it within the go-to-market stack or within the software development lifecycle, founders are starting with a wedge where they're winning in that specific narrow wedge and then expanding after they've landed the customer. So it's this idea of more of that land and expand model, start with something specific, solve a very um niche or not niche, but narrow uh problem, and then expand from that problem statement as you start to get gain trust of your customers. So I am seeing that there are a lot more vertical application use cases that are appearing. Like, for example, in the legal space, you have companies like Harvey and Even Up and others that are uh focused on kind of a vertical application use case. And again, they may start with with even within that vertical, they may start with doing one specific thing and then expanding and doing things that that that um uh expand across the entire uh job of a you know lawyer, call it. Uh, but I I'm seeing both, I would say. But I think that the wedge strategy is definitely one that is resonating with people because it's hard to tackle a bunch of things at once, right? If you win within one wedge and then expand from there, it's much easier than if you just say, I'm trying to do everything under the sun, and there's a lot of different companies that are also competing trying to do something horizontal as well.

SPEAKER_03 27:34

Got it. So uh let's just flip it the other side, right? So the enterprise buyers that uh we've talked to, uh the executive, they're of course bombarded with uh a lot of Gen AI startups, uh incumbents that they have from SaaS uh from the SaaS uh era, uh, and then they have the foundational layers players also, whether it's uh Microsoft or AWS or anybody, also coming up, right? Um there seems to be a distinction uh emerging between those that can be productive and those that can impact outcome, right? As as they as they look at buying things, and it requires a lot of stitching. How do you see enterprises reacting to because there's a whole business model, a whole uh lot of changes that they have to do, and they really haven't set aside a gen AI budget, they have set aside a business budget that some of it's getting allocated to that? How do you see enterprises reacting to uh the various approaches by startups who are born in the AI era but have could make an impact for them, but unclear how they scale or how they uh go across the enterprise and so on and so forth? How how what's your uh read from the market from the enterprises who are buying products from your from your startups or anybody's work?

SPEAKER_01 29:01

Yeah, so I think ultimately, even you go back a year or a year and a half ago, there was a lot of uh what I call experimentation budgets, willingness from CIOs or enterprises to uh deploy AI solutions because it w everybody was trying to find ways to be more productive. I think that is still true today. There are definitely still budgets that are just focused on experimentation or trying to find ways to be more productive. With that being said, I think more so now than a year ago, there is a mandate to make sure that there's actually our lie with the spend um that is occurring from Gen AI solutions. And I think that um there needs to be a way of showing that there is some value derived from buying these early stage startup software solutions or AI solutions. Um, and that can be measured in a variety of different factors or ways. And I think that that measure is based off of what product you're you know trying to sell uh to the enterprise. In some ways, it's more clear what the specific measure of value is. Um, and some it's harder to derive what the value is. Like, for example, in a content creation use case, maybe you're generating a bunch of images or you're generating a bunch of content copy. What's objectively good to one person may be different than what's objectively good to another person. So that's sometimes harder to measure. And so that's a you know, pricing discussion of, okay, well, how do you price a product like that? Is it just it that and so you know, is it consumption? Is it usage? Is it is it um some objective form of you decided to use this form of copy? Um, it that's a harder thing to measure. But then in other use cases like recruiting, for example, um, if you have a solution that's like an agentic AI recruiting solution that helps place your next candidate and the candidate came directly from that AI recruiting service, then you know, okay, the outcome that was derived was the placement of the, you know, the person, the employee. And so it's a little bit easier to tie some sort of value to. So I think this discussion around um business models or just how people are justifying spend at the enterprise level is evolving. And I think that it will continue to evolve. Um, but I am starting to see more different types of business models around outcome, outcomes-based pricing or services software start to emerge and even consumption-based usage, which has been around for during the software era as well. Um, but these, but these types of business models, more so now than ever, are becoming more relevant than seat-based pricing, uh which is the more kind of legacy way that I think SaaS was uh priced.

SPEAKER_03 31:29

So this kind of bleeds into my next question on emergence of business models, right? There's consumption-based, seat-driven, some are outcome. Recruiting could be, I mean, some verticals lend itself to outcome. Hey, you help me recruit somebody, I'll pay you on success. Some are sure if you're a CRM, it was a seat-based model because of the way it was. What business models are being experimented? Uh, which ones are sticking, and which ones are kind of uh, you know, not at all going anywhere. They're failing because just not it's not fits into the budget makeup of the enterprise buyer.

SPEAKER_01 32:07

Yeah, so I think the ones that you mentioned are the ones that people are trying today. So, you know, outcome, outcome-based pricing. Um, is there some value that you can tie the outcome to to price the product that you're selling? Um, that's one that's that's common. Consumption is another one that's common that people are are trying in the market. A lot of our portfolio companies are testing around that. Um, how much you use, how many queries you have, how does that result in um certain types of consumption um usage? And and that's also a hard thing to measure because you have to figure out what is it that's being consumed and what are the different factors. And I think as you think about AI agents, they're doing a lot of things for you. They could be doing a lot of tasks both in the background, or they could be doing it more, you know, upfront, but they're doing a lot of, you know, a lot of different types of tasks. And so how do you actually um measure measure that is the question. And like like we were talking about for some verticals, it's easier to measure, and some verticals, it's harder to tie some atomic unit to that. And so I think um those are two things that I'm seeing more and more. You know, you still see platform-based um uh platform-based models, subscription-based model still, because that's just things that people are used to seeing and used to kind of saying, like, um, I I historically allocated 50K to this piece of software, and now I'm willing to allocate 50k to that. So let's just like price it as this for now and then figure out what the usage is over time and then adapt, um, adapt from that. And I do think with more AI agentics types of um solutions, buyers are expecting that the reason they're buying it is because there's all these AI features around it, right? So they don't want to necessarily go and spend a bunch of subscription dollars to not get the AI features, or else they would just rely on their legacy software. And so I think there's some triangulation that's still happening in the market. There's some things that people are trying to figure out, um, but there's just a lot of testing going on to see what lands. I think pricing is always a little bit of a trial and error type of thing. And then you keep testing and experimenting to see when your customers are happier when they actually end up walking out of the door. So I think it'll continue to really evolve in the next um next few years here.

SPEAKER_03 34:14

So when you look at the uh 2025, given the changes that are happening in the market, whether it be at the foundational layer or whether it be adoption uh levels of the enterprise or anybody for that matter, are there particular areas that you guys are really looking for or look at more than the others, or is it as they come in, we evaluate and see where it goes? Because some verticals seem to be adopting more, adopting faster, I should say, not more, faster than others, or at least from the outside, it seems like, oh yeah, yeah, we we are all in on the AI. But it's unclear their adoption is any faster than anybody else's. It's just their perception. So as an investor, are you looking for certain vertical solutions, certain areas more than the others?

SPEAKER_01 35:05

So I would say that we're as investors at Madrona, we are interested in all areas of tech application. And I think that we see uh AI affecting all verticals in all industries. So I think that there's actually truly opportunity across all different sorts of industries. I would say that we do have a thematic approach to investing. So, you know, different quarters, we may dig into a certain area or have a belief or thesis that um a certain area is more right for disruption, be that some be that driven by some technological step function change that occurs that allows these models to do things that previously they weren't. Like reasoning is a great example of that. Um, and then you're starting to see more vertical use case applications around things like in legal, for example, because these models are really good at understanding large corpus of data. Um, you're starting to see the models get more performant around things around math. And so perhaps there's more opportunities that happen in the finance industry or other industries that are related more to doing actual math, which is a year ago they were, you know, very comically bad at doing math. And maybe still to today they're not not the best at that. But we start to think about, okay, well, as these models get better and faster and are better at reasoning and are better at memory, what are the types of things that the applications will therefore be better at at in in the next five to 10 years? So we always have to like to have a prepared mind and prepared thinking around um different areas of interest for us. But I would say in general, we also like to hear about what founders think is really interesting because for us it's you know, the founders are the day-to-day really in the weeds, thinking about this deeply. A lot of them have research background or operating background, so they're you know thinking about the problem space in a different way. And so we're always open to hearing kind of new and differentiated ideas. So um that's kind of how I would describe describe it.

SPEAKER_03 36:52

Uh so last year or last couple of years, at least since um your uh, you know, your joining Madrona, if you look at the last, say, four years or so, uh Gen AI has dominated the investment thesis and cycles. Uh it still happens. I mean, uh, I saw the uh that uh uh you know well-known, well uh uh uh founders are still getting a lot of investments, a lot of dollars chasing a lot of people. But is there a is there a shift from moving from that away? Because uh the there's enterprises that are clearly saying this is what I'm going to spend my money on, so that becomes very tactical, as opposed to you know, is it a problem that's solving for the long haul kind of a thing? Is there a shift moving away from pure Gen AI investments to other things that also involve AI and evolving from there?

SPEAKER_01 37:50

Are you saying within is that a trend that we're noticing within the enterprises or or more so invested from dollars?

SPEAKER_03 37:56

Invested dollars. Uh more that rather than the buyers, uh the invest investors such as yourself saying, hey, we've made a ton of bets on Gen AI surrounding applications. Now the next evolution of this is could they build something on top of Gen AI and therefore slow that down while we shift dollars towards the newer applications that could emerge, um emerge.

SPEAKER_01 38:23

We're still in the Gen AI era of investing, and I think that there is maybe more there are still dollars being invested into more of the enabling infrastructure, not necessarily the model layer, but just enabling um layers around how do you use these models more efficiently? How can you get more out of the models? So that could be around, you know, different kinds of frameworks around things like RAG or things like you know, MCP protocols. Like are there more um enabling pieces that enable you to build a better application on top? And I think people are definitely thinking about that. But I do think that there's still a lot of investment happening at the Gen AI application layer in particular, um, because we're just scratching the surface of that. I don't I don't think that we we've we've really moved on past past that. Um I think we're in the middle of of people still being excited about investing in um in this broad space because it's impacting every vertical, every industry, and there's a lot of opportunity to go after.

SPEAKER_03 39:24

So our uh listener viewer base has a lot of aspiring entrepreneurs. What advice would you give them about how they should go? Think about the new AI era, and therefore, what would be appealing to you if they bring something as an as an investor saying, these are things to think about? What advice would you give them?

SPEAKER_01 39:44

I think, you know, going back to what we were talking about, the founder market fit part is really important. Why are you the right founder to be solving this problem? What makes you passionate about this space, or what from your background lends you to believe that your way of thinking will be the kind of the future is is one thing that I would continue to think about. The second is, you know, just this idea of why now? Like what has changed from a technological perspective that enables you to build this application or or maybe you're building something at the infrastructure middleware layer, like why is it relevant if you think about the next 10 years? Why is what you're doing today relevant for the future? And so really just having a crisp understanding of the kind of why now as the world is shifting, as the world's changing, be that, be that kind of your thesis for the longer term, or be that a technological shift, I think that is incredibly important. And then just how you're thinking about the overall like UIUX experience of it, as we talked about earlier too. I think that as we go into a new era of of agents, perhaps, um, what does that mean? How do these systems all interact with one another? What does that experience look like? And so I think just um those are kind of the the high-level things that I would I would think about if I were you know building a company today. And and and um I think that it still just goes back to the fundamentals if you're building a company, you know, 10 years ago, but it's still um it's still all very relevant.

SPEAKER_03 41:03

Well, one thing that you said in your answer, I think is very relevant. I want to sort of highlight it to uh a lot of aspiring entrepreneurs. Shirich and I talked to a lot of them who are either thinking or who are going through this, and we tell them the cycle for them to think about is eight to ten years, not one to four years. That uh there's some you know, it's it's very uh you know uh fascinating to say I'm gonna go to a startup in four years, I'll do something. We try to tell them in four years you would have figured out whether your product is relevant or not. That's about it. The scale is in eight to ten years, so it's daunting for them to think, okay, I'm this age right now, and in ten years, I'll be that age, and you know, a decade has passed. Will I want to be doing that? You know, uh whether it's AI startups today, SaaS startups in the past, the timeline hasn't necessarily changed for success. You gotta you gotta put the time in. Totally.

SPEAKER_01 41:58

You always have to be iterating, and the first couple years can just literally be trying to think about how to iterate, how to listen to customer feedback, how to take that and and continue to build a a great product. And you know, we talk about this concept of flywheels all the time at Madrona as well. And it just it takes time for the flywheels to get going. And a lot of times it's a really long-term journey, and it's not for, you know, it's not one to two years or even three to four years, it's 10, 10 plus years in that journey. And so I I totally agree.

SPEAKER_03 42:24

Yeah, awesome. Sherish, back to you.

SPEAKER_02 42:27

All right, thanks, uh, Gauri. Uh uh, I have a few more questions uh regarding the investment uh climate. Um, you know, a lot of money was uh put into Gen AI companies in 22, 23, even 24. Uh, and it's now time for these uh companies to come back for air and raise more money. Um do you see them uh a lot of them have a cheap product market fit? Do you see them being able to raise more money, or are you going to see a lot of consolidation and accuracy?

SPEAKER_01 42:56

I think you're gonna see a mix of both. If I were to predict, um as you alluded to, there was a lot of capital that um was was put into these early stage companies, some of which will we're starting to see succeed. Cursor is a great example of that, publicly known. Um and uh and then there's a number of them that have struggled to find part product market fit and have struggled to really uh find kind of that narrow wedge that they're landing with and expanding to. So I think that um it'll be a mix as it always has been and it and it will continue to be. Um and I think that's just early stage investing as well. And so some companies are able to find that wedge and expand from it, and some some struggle to do that. So I I I believe it will be um a mix of both.

SPEAKER_02 43:36

And then um looking ahead for new companies, uh, you know, we are certainly entering some turbulent times with the tariffs and um interest rates still being relatively high. Talk about recession, etc. How is that affecting uh your Madrona's appetite for um new investments?

SPEAKER_01 43:56

Yeah, so I think for Madrona, um we are primarily early. Stage focused firm. So when when I mean early stage, I'd say, well, we invest from pre-seed or incubation all the way through Series C. As we were just talking earlier, this is a long-term journey, right? So, you know, think 10, 10 years plus. And so a lot of what's happening in the macro today doesn't affect us, that doesn't affect our early stage companies or kind of our appetite for investing in really early stage companies, maybe nearly as much as I would say if you were a later stage growth investor or company that was closer to IPOing in the public markets that are going to be more impacted by some of the macro that's going on, some of the ability to even go public, just given everything that's happening with macro uncertainty. So at the early stage, I would say uh relatively insulated towards some of this. And, you know, we continue to be really bullish on the opportunity around technology and AI investing. And so we haven't necessarily slowed down our pace of investing by any means. And we're still very, very active. We just raised our uh TED Fund, announced it earlier this year. So, you know, 770 million uh of capital and continue to deploy that very actively. So I think um, you know, with any uh with a macro, there's always going to be ups and downs as early stage investors. Um, we're we're here for the long-term journey. So a little bit less impacted than I would say late stage growth investors or IPO public investors.

SPEAKER_02 45:18

Okay, one final question is um what is the impact that AI will have on jobs in the next uh five to ten years? Uh, you know, we recently had a conversation with uh Arwen Bala, who's the CTO at Seakout, and they have just released some agentec software to help with recruiting. Uh, eventually it will replace recruiters in doing their jobs. Um, so do you see that happening uh really taking steam in the next few years that it'll actually impact uh the job creation market?

SPEAKER_01 45:48

I think it'll change the job creation market. I definitely think that it is true. There will be um AI systems or AI agents that are doing things that a traditionally an employee may have done. Um, but I think it'll also enable more creativity of the human to go and actually do things that previously they couldn't do. And so I think it's always going to be somewhat of a trade-off of how these systems are used. And um in some ways, maybe it's taking over certain types of jobs, but in other ways it's enabling you to do jobs that you couldn't do. So it's it's kind of a mix of both. And perhaps, you know, a non-technical person can now go be a coder and do and and and do coding to some degree and then be able to be, you know, more of a product manager as well and kind of layer on those hats where whereas before maybe he or she would have been doing something different because they didn't have those capabilities. And so I I think that um it's not as easy to predict in a kind of black and white setting, but I do think that it's it's gonna be both, it's something that we're watching closely, I would say. And and I think a lot of people worry about it for good for good reason. Um, but um I think that there will always be opportunities to kind of think about things newly or creatively um that historic that are hard for us to imagine today because we just don't we just don't see it.

SPEAKER_02 47:01

All right, excellent. Fascinating conversation. Uh always a pleasure to both uh read your newsletter and then listen to you talk as well. And uh we wish you all the best in your endeavors.

SPEAKER_01 47:13

Awesome. Thank you so much for having me. This has been a lot of fun. Appreciate it. Thank you so much. All right.

SPEAKER_03 47:19

Thank you for listening to our podcast from Startup Big brought to you by Dice Seattle. Assisting in production today are Isha Jay and Minnie Barbara. Please subscribe to our podcast and create our podcast wherever you listen to them. Hope you enjoyed it.