Astasia Myers, GP at Felicis, $3B AUM, AI Agents & Autopilots, New Infra Opportunities, & Roadblocks

0:00I think the individual will be more empowered than ever before to focus on what they care about and get the most joy from. Maybe that is too big picture, like high in the sky. But if you think about the creativity tools that we see today, it's like democratizing access to creation. It brings people so much joy. If you think about the enterprise products that are autopilots or copilots, so allowing people to really focus on the strategic aspects of their role that are more creative and interesting. And so I think it will enable more choice and opportunity in the future.

0:55Welcome to Sorcery. I'm Molly O'Shea, founder of Sorcery. Today, we have Astasia Myers, GP at Felicis, which recently raised a new core$825 million fund back in March of 2023. That brings their total AUM to over$3 billion. Astasia is focused on databases, dev tools, infra, and AI from inception to series A, with investments such as LaunchDarkly, Hex, and Superbase. We go deep into all things AI infra and even learn a bit about some of the predictions for 2024. I hope you enjoy. Hi, Astasia. It's a pleasure to have you on. Yeah, thanks so much for having me, Molly. Well, I'm really excited about this conversation.

1:41I think we're going to go deep in many aspects of AI, infrastructure, data stacks. But to start, I'd love to start wide and talk about Felicis, its history, and its founding story. Totally. So Aiden was one of Google's first product managers and an international sales leader. And he left Google in 2006 with a deep passion to partner with early stage founders. He went up and down Sand Hill Road looking for a role at a VC company, and he got feedback that he didn't have the right pedigree to become an investor. And this really inspired him to start Felicis and create a venture firm that challenged traditional venture norms.

2:27And so for us, Felicis widens the opportunity for founders from unconventional backgrounds as a team of outsiders to VC, we know that great ideas can come from anywhere. And so our mission is really simple. We want to back amazing companies that are inventing the future and become founders, most trusted partners by investing directly in their growth. And so we really believe in success with empathy. That's empathy to founders, to our LPs, to our employees, and to people across the industry. And so what's really cool about Felicis is even the name. Like we, Felicis means luck and fortune. And we believe that our luck is through preparation meeting opportunity.

3:14And so that's exactly how we like to invest. And so who are the partners that make up the fund and what areas do each of you focus on? Yeah, so we are a generalist fund with specialists underneath. We have 11 investment professionals, and we cover areas like AI, infrastructure, security, vertical SaaS, global resilience, and healthcare and bio. For me personally, I spend time in enterprise software, working with these founders across AI, infra, security, and SaaS. I have great teammates like Sandeep, who work and do a lot with vertical SaaS and bio and healthcare. Viv, who does enterprise software, infra and app layer and global resilience.

4:03Nikki, who has also done global resilience infrastructure and SaaS. Victoria, who has a background as a marketing leader doing vertical SaaS, fintech and consumer. and then Aiden, who works across all the different domains. Your recent$825 million fund brought Felicis AUM to over$3 billion. So I have to ask, what is new with this fund? And what is the team currently excited about? Yeah, so our heritage is being inception and seed investors. And over the span of time, you know, we've evolved our strategy to still love all things early stage. And we do a lot of our work there, but also doing series A's and very selectively series B's.

4:52And so this fund supports that mission of really our goal of just working with the world's best founders that are trying to reimagine very large categories. We're spending a lot of time in this fund looking at clearly AI. I feel like it's a must do, right, in different aspects. So Infra, AppLayer, AI-enabled services, foundation models, and how this could impact frontier categories like biohealthcare, robotics, and connected services. So it's really a continuation of the work that we've been doing for a while with a fund that we think right sizes for success so that we can support very early stage teams at inception and then build relationships to lead rounds in slightly later stage companies.

5:44So is your team mostly in the Bay Area or are you scattered globally across the nation? So we are based in the US. Our headquarters is in the Bay Area. We have offices in San Francisco and the Presidio, in addition to on Sand Hill down in Menlo Park. And we have team members that are also in New York City. So that's kind of emerging space for us. We have a new office out there, which is very exciting. We do invest globally. And we really take this perspective that the best founders can exist anywhere. And so we have partnered with early stage teams like Canva and Dovetail out of Australia, Audion out of Europe.

6:32Really, we look across the world when we partner with founders, but we're based here in the US. Amazing. And I do want to dig into your particular background. So you're a name to know in enterprise software, and you have quite accomplished in many different seats. So I'd love to understand from your perspective, how did you get your start and what hooked you to keep you building your career and becoming a GP? So I was born and raised in the Bay Area and kind of grew up in a single parent household where I was very fortunate because my mother worked in the semiconductor industry, always at early stage startups.

7:17And, you know, dinnertime conversation was really us talking about her work. And, you know, back then dealing with design cycles and tape outs and trying to get these contracts with chip providers. And so it was very exciting to me. And one of the most seminal moments in my childhood when I was, you know, later in elementary school was she was working at a company called MIPS and they designed this chip that went into a Sony robotic dog. And as someone that loved dogs and animals, I was like, oh my God, this is totally amazing. I can't believe you have a robot dog pet now. And I can't believe my mom was a small part of making that reality happen.

8:01And really through these types of experiences, I gained a deep appreciation for hard tech and doing something that has never been done before. And so in undergrad and grad school, I was always researching the role of software and technology in society and economies and coming out of grad school, I really realized that I wanted to be part of the action. And so I evolved my career to being in sell-side equity research, really understanding publicly traded companies and IT networking and security, what were the levers of their success, the management teams, product lines, and how they manage the financial positions of the company.

8:49The opportunity to deep dive into businesses and really appreciate what success looked like and how you got there deeply informed how I view companies today. From working in sell-side equity research, I joined Cisco. It It was a company that I actually covered, so knew their product lines really well and had an opportunity to get to know some of the management team. And I was deeply inspired to join them because they always had really amazing NPS scores when we did buyer surveys. And they were super creative when it came to thinking about venture, M &A and strategic partnerships. I mean, Cisco was the most acquisitive tech company of all time and had very cool innovations like the spin in where you could have a call option to buy amazing startups.

9:42And so I wanted to get closer to the action. And so joined them for a few years, focused on M &A and venture investing for their core business lines of servers and networking. Through that experience, I naturally gravitated to the investing side of the house. My sub team was the most active across all the BUs, and it was incredibly fun to partner with businesses like Cohesity and Gardacore. And, you know, through that experience, I realized that I wanted to focus on early stage investing full time because you get to be there for a lot of the first. It was very enjoyable to tie founders into Cisco's go-to-market engine through doing events together, co-designs, and really trying to help move the needle for businesses.

10:34And I felt like I could do that at the earliest stages. And with the love of focusing on deep tech and hard tech, I stayed in infrastructure and joined Redpoint's early stage team for a number of years. I had a great time partnering with businesses like Hex over there. Most recently, I was at an early stage team called Quiet Capital, where I was helping build out the enterprise software practice. And then earlier this year, I joined Felicis as a GP to continue those efforts, as I said, across AI and infrastructure. So I think really for me, the North Star has always been working with really early stage founders.

11:18It's an incredibly fulfilling process where you can be a first believer in them before everyone knows about what they're building and their big vision and really try to tangibly help these teams. Even yesterday, it was very fulfilling. One of my seed stage companies hired their founding engineer. And literally right before this podcast recording, they said they closed their second engineer. And I was just like, so excited for them. It was like, we were talking about these candidates. I worked with them to help qualify them. And they did a great job closing. I'm like, oh, my gosh, like, that's so cool to be there for some of the firsts and be a sounding board and kind of be that rock and champion for them.

12:00So you don't miss the days of being a hardcore sell side analyst? You know, I have to say, I love the experience because you get, I feel like the skills that you learn there of like understanding financial modeling, having the opportunity to really dig into product lines, do primary research. Those skills are directly applicable to what we do in venture. and some of my favorite work when I was there was deep diving, like emerging technology. And at that point was like, what are the late stage startups that are going to disrupt the publicly traded companies? And like software defined networking and software defined storage back then, hyperconverged infrastructure.

12:44This is all like so logging up. You were like, what are these things? But it was a ton of fun. And it's the principles and best practices from that I try to apply as part of my VC role. Yeah. And now it's an active role, right? Something you've been doing for many years now, but you're actively involved in investing into research and into the innovation class. Exactly. And that's a ton of fun. And it's also much more enjoyable to build deep relationships with the founders themselves. I like to view myself as an extension of the team being available 24 by 7 to think through any situation and try to support them as best as we can.

13:31And that level of interactivity with founders and team members, you just don't have in self-sidged research, right? You're so removed from that. And so, yeah, it's just fun to have the personal relationships as a key component of the role. So I'm curious, what other types of resources does Felicis offer your portfolio companies? Do you have a platform team? Is it you? Do you kind of handle most of the relations? Like how does that work? Yeah, Felicis is fantastic. You know, in addition to the 11 investment professionals who work collaboratively to support our companies, and we also have a founder success team that helps with events, content, talent.

14:18And so we really try to bring to bear ourselves as investors, our success team, and our network on behalf of founders. And yeah, it's been a ton of fun being here and just seeing ways that our success team has moved the needle for our founders. A great example is one of my companies collaborated with one of these seed stage companies to do some events with them. We do buyer dinners. We try to identify, qualify, and provide leads for founding roles and executives. And so, yeah, it's a comprehensive approach to helping teams. Hey, we'll get right back to the conversation after a word from our sponsor.

15:04Sorcery is brought to you by Archer. I'm genuinely amazed at what Archer has been able to accomplish. Archer's goal is to transform urban travel, replacing 60 to 90 minute car commutes with estimated 10 to 20 minute electric air taxi flights. They are safe, sustainable, low noise, and cost competitive with ground transportation. Archer's Midnight is a piloted four passenger aircraft designed to perform rapid back-to-back flights with minimal charge time between flights. Learn more about how Archer is set to open up a new world of opportunity for passengers by providing safe and efficient access to people, places, and events across the communities they live.

15:40Visit Archer.com. Wonderful. So in preparation for our conversation, we talked about this a little bit, but you mentioned that Felicis is kind of taking AI in three buckets. So I'd love for you to explain out what those three buckets are and maybe dig into some examples for each of them. Yeah, so we kind of think about three different aspects of AI today. One is AI infrastructure, which has been very hot over the past two and a half to three years, really trying to reimagine what AI engineers and builders need as part of their tool suite to build the Gen AI applications. This is everything from data management to looking at the models, frameworks, RAG systems.

16:40And this has been very active for us. So an example of that is myself and my great teammate Viv led the Series A in a business called Dottology that's helping with AI data curation for teams. How do you make that process of deciding what data is part of the training data set as effective as possible? We work with other businesses like Predabase and the fine tuning space. Strong belief that with the rise of open source models, teams can now take their proprietary data and create a better service through fine tuning. So AI infra is one big bucket for us. A second bucket is AI-enabled SaaS. And we think about it in two different ways.

17:32One is autopilots, kind of think GitHub Copilot and how it helps accelerate code generation for engineers. And the second bucket is autopilot. So any software that can actually create the entire program application on behalf of engineers, we apply these kind of principles to different domains within SAS. So office of the CFO, the sales stack, even for marketing or particular roles like data engineers or data scientists. How do we offload as much work as possible using AI agents on their behalf? Kind of thinking about it as areas of opportunity to automate with AI agents as anything that's when you're acting like a execution machine or workhorse, kind of like rote tasks that are repeatable in nature as compared to strategic and creative thinking that we think will still be in the hands of the employee and is really their value add to any process or business.

18:41And then the third bucket that we're looking at is AI enabled services. So something that is very interesting about the services market is, you know, for every one dollar of software that they charge for, there's usually four or five dollars of human capital value that is also being created. And so it's very cool to think about a future where you are building new services organizations that are using AI agent enabled software to change the margin profile and deliver a completed product faster and more effectively than traditional alternatives. So we've been looking at legal services where you have businesses like Lighthouse that are using AI agent software to help with the immigration process and application.

19:34Businesses like Crescendo who are in a contact center space that are really helping agents be more effective to take on more cases at a single time. So both of the three buckets, we're equally excited about all of them. But we're really looking for teams that are trying to take over entire units of work so that there's a deliverable at the end that someone would have been manually doing in the past. yeah that makes sense and it's also really helpful to to cut it up that way because i feel like even earlier on it's just such abstract of a category and even like it became just the term people were using of just ai ai ai but breaking it down and focusing on these three categories of ai infra ai enabled sass ai enabled services really does help kind of contextualize it and understand like, oh, this is where I get it.

20:41Or this is actually how I should be thinking about a business model. Totally. And I think, as I mentioned earlier, there was an extreme interest in AI infrastructure over the past few years, because you kind of need the picks and shovels to move up the stack and create more value in SaaS or AI-enabled services. And there's been a lot of investment in AI infrastructure, which has been fantastic. And people kind of have those tools now. And so we really believe on the SaaS side of the house that we're kind of having a similar moment to what we saw in mobile, which is over the next few years, we're going to see massive AI for SaaS and services companies come together, just like we saw when mobile came out and you had Uber and Lyft and Shopify and a whole others that emerged.

21:34So we're paying very careful attention to what's going on at the app layer. And if there's any app layer founders out there, please reach out to us. Noted. And we will put that in the notes. So I actually want to break one of your points down a little bit more. So given your background in research covering this category enterprise for a while and uh now it's you know become ai infrastructure rather than just infrastructure because ai is now gonna like i think i heard a statistic that it's gonna be like two times more power intensive than what the u.s already consumes or generates uh so that's kind of crazy but in terms of that like working from endpoints that's the critical point to start figuring out how to build or rebuild.

22:30So I'd love to get from your perspective, what was the arc of infrastructure and software stacks throughout time? How did enterprise really evolve and what were the main critical points? Totally. Yeah. So loving infrastructure and studying the evolution over time, You know, really coming out of the late 70s and 80s when we started to see companies building applications for themselves and sometimes more infrequently for end users, either employees or consumers. You know, you have the core pillars. You have networking, storage and data management, compute, and then you add on security to make sure everyone is safe.

23:20And all of that was on premise, right? People were building their own data centers. Everything was appliance-based where Cisco or Dell EMC would go and build these like full stack services that you'd actually implement. And that was happening in the 90s so that more people could have the resources to go build software. What's been really interesting since the early 2000s to late, yeah, into the 2010s is the rise of cloud service providers like GCP and AWS and Azure, who are now democratizing access to infrastructure services, because now the CapEx required to procure them is so much lower, and is all self-serve and usage based.

24:11And so that's really lowered the barrier for builders and even creating companies, because now you can get all the services you need right out of the box. Something that's been very interesting to see in the past 10 years is certain categories of infrastructure becoming more popular, kind of like the ebbs and flows. And so when I was starting my career, a huge trend was software-defined infrastructure. so that you could have more flexibility in which hardware provider you were partnered with, because the intelligence layer was a separate service that could run across all different types of endpoints.

24:52Then we saw a shift from a buyer perspective, from the CIOs making technical decisions about what infrastructure is procured, to more of the engineering teams and developers themselves, shifting from top-down sales to bottom-ups go-to-market, and that shifted interest in having developer tools. And so you saw the emergence of GitHub and GitLab, HashiCorp that had amazing bottoms-up go-to-market motion because they had strong developer ergonomics and a lot of decision-making was happening after there was usage in an account. And so it was no longer the traditional buying just steak dinners to close an account and force technology onto teams.

25:43It was really bi-directional with the end users as well. When I was, you know, kind of right before the 2020s, there was a huge push from looking at developer tools into the modern data stack. Really with the rise of Snowflake and providing a cloud data warehouse, everyone was now able to ingest numerous data sources into a single backend where they could perform analytics and processing. And so there were many amazing companies that were able to benefit from that in addition to Snowflake, like the ingest layer with Fivetrend and Airbyte, analytics with Looker, transformation layer with DBT, and then Hex on like self-serve data workspace for teams.

26:38And so that was a very big category and theme for a number of years. And then more recently, to your point, it has been around AI infrastructure. The old world of how we built machine learning stacks is less applicable today because of the model types that we're running. And so teams are no longer necessarily using like PyTorch and TensorFlow to develop a prediction or recommendation models, but really using these foundation models to have conversational AI or agentic workflows. And so the infrastructure that you need for that is significantly different. And that has been why over the past few years, there's been huge interest in that category.

27:31One area of infrastructure that I feel like is emerging that I'm very excited to see is anything at the edge. So by moving compute and storage to the edge, you can have much better performance and responsibility from applications. And so I'm looking for startups that are working in that domain because I think it's the new frontier. Well, you should definitely check out Armada. I'm a huge fan of them. We are investors. Oh, there you go. Okay.

28:09That's awesome. So, Astasia, is there any particular edge computing company you're excited about in your portfolio? Yeah, it's great. That's so funny. That's awesome. Okay, so that's super helpful. And thank you for giving the context over time and how it's evolved. I have looked at recent modern data stack maps, and they are horrifying. It's kind of like the same as like ad tech. It's just thousands of logos. It's crazy. It's crazy. And so I think that during that era when there was a lot of investment in the modern data stack, people were really, some people were really thinking about, gosh, if Snowflake is a 50 plus billion dollar company and this product is X dollars, how to put this, if someone is paying a hundred dollars for Snowflake annually, theoretically, and this startup can take X dollars of that, you can extrapolate that it can also be a multi billion dollar company.

29:22And for some companies that has completely been true, but then there's been some other segments that are smaller than people anticipated. And the larger higher flying companies have brought in their portfolio scope to kind of eat up some of those smaller domains. And so now there's been a lot of talk actually recently, and you may have seen some of these blog posts about like the end of the modern data stack, essentially, where it's pretty clear who the emerging winners are. And hence, as early inception investors like myself, we're looking to these other categories of like AI infrastructure and edge technologies.

30:02Yeah, it's really interesting. I'd say like over the last year and a half, the GPT moment, that was like a critical component and point in which everyone really realized, like, oh, this is a big deal. I'd love to get into it a bit further. But from your perspective, How is AI different today than what we saw last year? And I'm not talking like just on like a marketing basis, but maybe down like below the surface. Yeah, I would say that there's two things that seem very different this year than last year. One is at the beginning of last year, I felt and I think the industry felt that there was a lot more hegemony with open AI, where it was like one model company will win it all.

30:54And that model company is open AI, which has clearly done fantastically well. And we should all wish we were investors in that business. However, I think that we've also seen the emergence of great alternative generalist foundation model companies like Anthropic and Mistral, where we're seeing increased interest and even Llama from Facebook increased interest because certain models perform better for certain tasks. And so from a builder perspective, they're more agnostic to the models that they were using today than previously. We're also seeing the importance of foundation models that are tuned and oriented to specific tasks.

31:44So, you know, we are investors in RunwayML that has built their own foundation models for the creativity space for text to image and text to video. We're in Poolside as a specialist model for cogeneration. we're seeing really amazing things for foundation models in the world of AEC, in the world of doing physics modeling. And it's very cool to see that there is now a number of different financial models, models that are either generalist or for specific verticals. And I don't think there was that much activity last year. A second thing that's been a very big difference this year compared to last year is the rise of AI agents.

32:43OpenAI talks about their five tiers of getting to general intelligence. And last year was really level one focused on conversational AI. There was a lot of interest and pressure from an executive level to do something with AI. And conversational AI was a great entry point for teams to get started building their rag stack. And we saw some early innings with AI agents for auto GPT, but the resilience and reliability of those services just weren't there. And while we need better models for reasoning to go to level two and level three, it's been very cool to see teams start building services that are enabled by AI agents.