Infra Investing | Astasia Myers, GP at Felicis

Astasia Myers is a GP at Felicis, an iconic VC firm with investments in companies like Shopify, Canva, Adyen, Notion, Mercor, Plaid, Supabase, Flexport, and more. 

Astasia's favorite books: God's Bankers (Author: Gerald Posner)

(00:01) Introduction
(00:26) Astasia’s Infra Thesis
(03:59) Golden Age of Infra & Innovators Network
(06:22) RL Environments & AI Agents
(08:57) Disruption Opportunities: Data & Observability
(11:31) Where to Find Infra Founders
(16:31) Early Signals & Thesis-Driven Investing
(18:01) Picking & Decision-Making Process
(20:11) Red Flags in Infra Investing
(22:20) References & Diligence
(24:35) Proof of Usage & Production Signals
(26:24) Building Edge as an Investor
(28:01) How Felicis Helps Founders Post-Investment
(30:05) Consensus vs. Contrarian Views in Infra
(32:09) Tourist Traps in Infra Investing
(34:43) GTM & Sales Motion in Infra
(37:25) Pricing Strategies for Infra Startups
(40:09) Ecosystem vs. Core Product Focus
(42:15) Lessons from Outlier vs. Good Companies
(44:30) Infra Wedges to Fund Today
(45:23) Commoditized but Promising Categories
(47:06) Exciting AI Advancements
(48:21) Rapid Fire Round

--------
Where to find Astasia Myers: 

LinkedIn: https://www.linkedin.com/in/astasiamyers/

--------
Where to find Prateek Joshi: 

Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-infinite
X: https://x.com/prateekvjoshi
Research column: https://infrastartups.com 

Transcript

Prateek Joshi (00:01.271)
Astasia, thank you so much for joining me today.

Astasia Myers (00:04.408)
Thanks so much for having me. a joy to be here.

Prateek Joshi (00:07.865)
Let's start with your view of the infra world and you're one of the few people who I follow and in terms of infra and you post very interesting things. So let's start with your infra thesis. How do you look at the world of infra?

Astasia Myers (00:26.796)
Yeah, we actually have two theses right now. First, we think that the most valuable part of infrastructure is actually the model weights. So you want to invest either upstream or downstream from the model weights themselves. You're probably thinking like, how do you do that? The model weights are part of these really large AI, infra companies like open AI and entropic. So we think about it in terms of the data supply chain.

that goes into the model weights. And we've been investing across this thesis. We've partnered with businesses like Eventual that offers Daft, which is an open source data processing engine that's best fit for multimodal data that can be used for data prep or pre-training. We've invested in Mercor, which offers chain of thought data, and also Datology that offers AI data curation for teams that are pre-training.

The derivative of investing in the data supply chain for model weights is actually investing in the model context layer, the information that the model has at inference time. And so we've invested in Letta, which is AI memory, as well as Chalk, which is a real-time feature engineering platform. So that's been a big thesis for us over the past 18 months, really, since I joined Felisis.

The second thesis, which is more emerging that we're excited about is how we're shifting from the users of infrastructure being people to AI agents. And with that shift, there is a change in the amount of load and stress and strain put on these systems. You know, we work with a company called Superbase, which is a Postgres database service, and they support vibe coding apps like

lovable where AI agents take prompts and build applications. And with that, you have different needs like incredibly high provisioning, fast provisioning time, the ability to do branching and snapshotting so you can have rollbacks. And so we think this is just the beginning of how AI agents will be influencing the infrastructure stack. It's exciting. It is much more frontier.

Astasia Myers (02:50.53)
To think about that layer, I think we're just at the beginning phases of AI agents at scale, but we think in the span of time, it will be incredibly disruptive and really a re-imagining of infrastructure, taking it to this next level of scale.

Prateek Joshi (03:08.697)
That's amazing. I love the crisp framing of how you look at it. And also Ari Morkas of Datology. He has been on the podcast, big fan of just same thing. It's just very clear in how he thinks. And also SuperBase. I love the product. Obviously Postgres is, it's hard to do new things in Postgres, but SuperBase has this delight to it where you just sign up and it's surprising how many dev tools just forget about the delight.

So I think it's wonderful. Now, Felices, and you mentioned that we are in the golden age of infra and you run the infra innovator series. So what themes consistently emerge or surface during these conversations that outsiders may miss?

Astasia Myers (03:59.17)
Yeah, as you mentioned, we have a Felices Infra Innovators and AI Innovators Network here. It is a really amazing group of builders and buyers from hyper growth companies to publicly traded ones. And we get together regularly to talk about key themes, pain points, and where the market is moving. If anyone listening is interested in joining, please reach out to me. We'd love to have you at one of our events.

and what's really great about organizing these individuals and kind of like a safe space for sharing as they talk about what's kind of on the frontier for them, not just what they're working with today, but where they see the market going and the needs that they will have in the future. a few themes that have come out recently is of course all things AI agents. I think once again, we're still in the early innings of

production grade AI agents that are highly reliable. And we believe that this next phase to making them even more performant is going to come from reinforcement learning, where you can kind of train these AI agents through a reward system to make them more performant. And we've broken down the RL ecosystem today into two core components. One is

RL as a service startups that go into enterprises and large scale businesses who may not necessarily have the talent in house to do RL themselves. They go in and identify unique data sets and use cases and then bring to bear an RL agent that they will serve to their customers. The second aspect of the ecosystem are RL environments. And actually this week,

Andre Capati was highlighting how he is a big fan of RL environments. He thinks that they're actually maybe even more enduring than post-training RL, that it could be one step in how we make these agents better. And RL environments allow the agents themselves to be trained. And the RL environments can come in many different forms. We've seen browser-first RL environments, full computer use, even

Prateek Joshi (05:54.659)
Yeah. Yeah, yeah, yeah, yeah.

Astasia Myers (06:22.141)
RL environments that focus on co-generation and software development and security use cases. And so while we know that the model companies have been doing RL and have environments as a priority, we also think that in the span of time, while nascent today, this will be top of mind for enterprises themselves. And that RL environments will be key for any business to improve their AI agents.

just as Datology unlocks the ability for any company to do pre-training for really amazing specialized models for their use case, we think that RL environments in the span of time will do that for enterprises as well. And so there are few startups in the RL environment space, Mechanize, Kaizen, and a number of others, but we're very excited about this new technology.

Prateek Joshi (07:20.205)
Yeah, I was about to reference Andre Garpati's tweet exactly where he talked about how personally he's bullish on environments and agent-tech interactions and how... I love the analogy he uses. It's like a gem for the agents. Like, go, go, workout, get better at whatever part you want to get better at. And right now there's no good way.

for people to do that, for their agents, unless you're a very, very big company or a research lab. And so I think this response was to Prime Intellect, the environments hub, the model environments hub that they released. But yeah, I think this is a very exciting area. Now, if, yeah.

Astasia Myers (08:05.089)
And it's interesting because environments can be quite specific to an application or a service themselves. And so that's why we think in the span of time, it will move beyond just the model companies who are requesting environments to the broader global 2000.

Prateek Joshi (08:13.902)
Yeah.

Prateek Joshi (08:19.745)
Yeah. Yeah.

100%. I think that's a huge dimension along which hopefully it'll unlock the next level of scale, is, that's why it's very exciting. Now, if you look at the different layers, like data layer, orchestration, inference, safety, and if you had to redraw this stack for the AI native companies, which layer do you think is more open for disruptions or new?

category leaders to emerge and why.

Astasia Myers (08:57.517)
It's interesting. I think the two layers that are most compelling right now is this data layer. Data has always been companies' crown jewels, but there's so many new aspects to the data today. We talked about Eventual's DAF data processing engine. How do you offer something that is really exceptional at multimodal data, which is image, video, voice, that is the key component for this new Gen.AI workload?

Also with thinking about a chalk, know, the faster you can get data into the model pipeline at inference time, the more performant your service can be. So we really think like data has always been a great category to invest in. This is just the next wave of opportunity. The other area that we think is ripe for disruption is observability broadly.

We think about observability today in two core buckets, traditional observability for services. So think about the re-imagining of Data Dog with an AI first lens where you have more intelligent models identifying anomalies, but then also helping with the root cause analysis to decrease mean time to resolution for issues and automate, help automate and augment, you know,

engineers time. We think that's very exciting and hasn't been possible before. And then also observability for the Gen.ai applications themselves. There was kind of a first wave of looking at the inputs and outputs of the model and maybe the degradation of the service. But because these products are now coming online, it also is important to understand how the end users are experiencing the products themselves. What are the issues? What are the areas?

that they're actually enjoying to tune the system, as well as the efficacy of the AI agents themselves. You can't apply the first gen LLM observability company to AI agents because they're dealing with multi-step workflows that are trajectories, they're calling tools. And so you really need a novel approach to understanding how the agent is performing.

Astasia Myers (11:21.025)
doing the debugging and then also collecting that trajectory data for once again, RL post-training to get a better service.

Prateek Joshi (11:31.459)
Let's move on to finding talent and founders. And when it comes to infra, it's a little bit different. So where do you actually find, go to find founders, infra founders before everyone else? And over time, obviously different channels work differently, but what methodology or what technique that you use has worked best for you in spotting these founders early?

Astasia Myers (12:01.045)
Yeah, we think that there are three archetypes of founders today. One archetype, has been enduring is the domain expert, usually from a hyper growth or publicly traded company who has built the product internally and now wants to transition and democratize it to everyone else. So we still look around and

try to identify really exceptional teams that are on the bleeding edge of infrastructure needs and build relationships with them. And once again, the infra innovators group that we have kind of helps with that as well. The second archetype that we're seeing is the applied researcher coming from academia or the labs themselves who have seen the pain points in the infrastructure.

and are now applying their research to make it better. So we spend a lot of time with MIT, Stanford, and Berkeley. I've been really humbled to partner with teams out of the Sky Computing Lab at Berkeley, which is the same lab that originated Databricks and AnySkill. Actually, the Leta team that I work with comes from there, and LM Arena as well.

And it's really amazing because they will take their research and the community they've developed around it to commercialize the efforts. The Leda team, actually, I got to know because I was reading their research. I spent a lot of time seeing what's coming next. And I think that the research papers on archive are incredibly useful. And so we do that on a weekly basis. And then the third archetype that we're seeing is kind of

the Gen Z founders, the hackers who just have incredible insight and are much more like AI native, right? They are not necessarily framing the opportunity in terms of like what has existed before, but coming with fresh eyes in terms of what they need to be builders. And one of the ways that we help build relationships and, you know, activate that network is

Astasia Myers (14:19.757)
through our Felices Fellows Program, which is each summer, we just did our second one this July, we bring together undergrad and graduate students who have been doing research or have an interest in AI for a one week session where we bring in founders and executives to tell them about what it's like to become an entrepreneur, what are the themes.

that people are building in and it culminates in a really awesome hackathon where they come and they present their idea for the company and there's really nice rewards for the top builders and we also sponsor their work if they're interested in pursuing it. So there are the three different archetypes. I would say in Infer in particular, this archetype of the Gen Z builder is newer and that's why we have

programs for that group of individuals because it's really exciting. know, they don't come from the world of I worked at Google and this is how things are supposed to be. Really have a fresh take.

Prateek Joshi (15:19.937)
Yeah.

Prateek Joshi (15:25.216)
Right.

Yeah, no, that's amazing. And I think it's consistent with my observations because in the past, if there's a networking company, then a big VP at some big Cisco Juniper, they come out and exactly. And so they come out and they have the GTM locked down. They signed million dollar contracts, but now more and more interest is becoming the...

Astasia Myers (15:40.808)
yeah, Cisco, I went to Cisco, totally.

Prateek Joshi (15:51.073)
It's bottom-up motion is becoming very, very prevalent because all the developers and users are also Gen Z and they want to use tools first. They want to try out and then that notion is becoming very interesting, which means we have to track all the signals leading up to that adoption. So maybe my next question is centered on that. You mentioned that you read papers to get familiar with what's getting built and then that's one signal to see where you want to spend time on. What else do you track?

to identify some of these early signals? Could it be open source tools, some developer tools, something that's kind of picking up momentum? What signals do you track?

Astasia Myers (16:31.725)
So we really like to have thesis-driven investing here. That's one of the reasons that we publish out our thoughts and also have a call for startups. So we do a lot of primary research, speaking with users and buyers about what they are enjoying in terms of new products that they're adopting, as well as pain points where they wish there was a vendor that was solving it. So we really try to have on-the-ground feedback for, you

more public external resources. I've been a long time investor in open source technology is having partnered with five plus companies that, that have that go to market motion. So we're very mindful of what's going on on GitHub and open source communities. And then speaking to kind of that new or builder, that Gen Z archetype, we really are looking for individuals through social media.

that are publishing out their work. think what's been amazing and a huge shift as an investor, particularly an infra investor over the past few years is LinkedIn and Twitter as a distribution mechanism for founders to raise awareness of what they're doing. You can often see them launching many different types of ideas on there, iterating as they go. And so we do spend a lot of time trying to see signal.

through social media.

Prateek Joshi (18:01.369)
Let's move on to your picking process, decision making process. So let's say you spot something, it's interesting. What's your path from there to investing at the seed stage? Meaning things aren't really baked in, there aren't that many metrics to look at. So how do you make a decision at the pre-seed or seed stage here?

Astasia Myers (18:24.865)
Yeah, so for us, first and foremost at the earliest stages, which is where I spend the vast majority of my time, often partnering with teams where nothing has been built yet, we really look at the teams themselves. Is there a domain, if they've been further in their crowd, is there like a domain expertise that they have or like unique insights that inform the product and go to market motion?

Is there some secret that the company is predicated on that will give them a long-term advantage? And we kind of assess their understanding of the problem space, but most importantly, the ability to have really fast learning velocity and a propensity to action. You know, I started my career 10 years ago in infrastructure and the time horizons for change back then were

were much longer, you know, it could be years before someone would consider a new technology or the offerings would change dramatically. In AI, it's faster than ever before. Sometimes it feels like it's month to month. And so the only thing in AI and infrastructure right now that's consistent is that it's going to change. And so if you're a founder building at the earliest stages, we really try to assess, you know,

Prateek Joshi (19:43.926)
All right.

Astasia Myers (19:49.538)
your willingness to learn and take feedback, but also listen. Like really amazing founders know how to listen and listen deeply and pull out what is most meaningful to inform the product and go to market strategy. And so we try to assess that when we're getting to know them.

Prateek Joshi (20:11.371)
And during diligence, what's a red flag you try to spot? Meaning like if one of these things is true, I'm not going to invest. Like what are those things that you look for?

Astasia Myers (20:27.277)
Well, you know, I would say a really good smell test for us is if we learn something new about the market space or there's a unique insight, but that's non obvious. think that's a really good smell test. We, you know, since we are domain experts, we do spend a lot of time in the categories and, you can hear things that are repetitive, but if some, founder has something that really spikes and is different.

That's a good smell test for us of like, well, you've really gone that extra step in terms of a red flag for me, I think one of the things that another thing that's changed in infrastructure investing is the real importance of marketing and community development, breaking through the noise to truly reach your audience and build authentic rapport with them.

And so if a founder is undervaluing that aspect of building a company, that's a red flag for me because you know, the best technology doesn't always win and no one will know you have a great product or technology unless you get it out there. And so I really want founders, any AI infrared dev tool founder listening to really take away with, know, start building your

community yesterday, be unabashed about your particular perspective on a market. It's even okay to be controversial. Sometimes that's even better because you will find people that deeply agree with you. But really, if you're not inclined to the go-to-market side of the house, that's a concern.

Prateek Joshi (22:20.011)
And when it comes to understanding the business of a given company, what references do you always run for infra? What I mean by that is, is it the buyer or is it maybe the platform owner? hey, I'm building on Sonnet, so Anthropic could kill your business. Hopefully they don't, but they might. Or maybe it's the user, maybe a reliability engineer. Like who?

What's the killer reference here? Like who do you go to to verify? Okay, this is a real thing this company's built.

Astasia Myers (22:52.941)
Yeah, I actually think that it's more than just one, of course. You know, the user and the buyer are key to our diligence process, really assessing what is the pain point this technology solves on a scale of one to 10, one being no pain, 10 being extreme pain. How much pain do you have today? Trying to assess urgency and appetite for a new product and the timing. Sometimes the timing is off.

right? We're just not there yet. And also the willingness to pay. As I said, I work with a lot of open source companies. Open source projects and packages are amazing, but sometimes there's not a monetization path for them. So we assess willingness to pay very actively. The other prong that we like to understand is the ecosystem in which these technologies operate and who are the most strategic partners to them.

This is useful because it not only enables us to understand if there are the defensibility of the technology, right? If your biggest strategic partners, we go talk to them and say, yep, that's on the roadmap or yep, like we can imagine going that direction. Then it could be much harder for you as a business as compared to the strategic partner saying, my gosh, we are looking for BD relationships here. We think it is distinct from what we're building. It's not a priority. And you're like, wow.

That's really important to know. It's a key layer in the stack. They don't want to go after it. And now you have a new customer acquisition channel as a startup. So we do try to look at both types of personas.

Prateek Joshi (24:35.649)
And when it comes to usage of the product, you mentioned you partnered with founders very early, even before they have a product. is there a minimum threshold for product proof that you look for? Obviously, there are special cases where the founder is super strong, so you team up with them. But let's say the average case where you need more proof. So what kind of usage metrics or usage signals do look

Astasia Myers (25:03.765)
Hmm. That's a good question. We like to anchor around teams running in production. And there's a lot of, you know, innovation teams are doing exploration or talking things out pre-prod. That's fantastic. But we really like to see this is so critical it can get in production. And even at the pre-seed or seed stage, if they

have an MVP, we'll go talk and it's only being kind of trialed. We'll go talk to the design partners and saying, when do you intend to get this into production? What is the level of urgency? Because value accrues in prod, right? So we try to tease that out as early in the process as

Prateek Joshi (25:54.969)
Yeah, and I think people jokingly refer to experimental ARR, which is like people are trying stuff and surely enough the churn is going to be here and it's going to look horrible. So I think it's very important to delineate experimental ARR from real production. Oh my God, we need this ARR. So it's interesting. So going to the next part of the convo, I want to talk about building an edge.

Astasia Myers (26:14.689)
Thank you.

Prateek Joshi (26:24.697)
as an investor. So you mentioned a few different topics. You run Infra-Innovators, you read papers, you try to get to founders and builders early. But if you had to summarize, where does your edge

Astasia Myers (26:40.237)
The examples you gave, I think, all feed into kind of a bigger topic, which is like domain expertise and passion for AI and infrastructure. You know, I've been an infrastructure investor for almost 10 years. I've been really humbled to partner with teams at the earliest stages, be it LaunchDarkly or Modal, SuperBase among others. And so

Prateek Joshi (26:52.761)
All

Astasia Myers (27:08.837)
For me, think my edge is having the appreciation of the technology, how it's different, why it matters, and how it can become a really big company. So I think the domain expertise feeds into the innovators network that we have with buyers and builders. It feeds into our talent network of founding engineers and go-to-market executives and advisors.

And it feeds into the BD relationships that we have with the hyper scalers and, and model companies to really give the early stage startups that extra leverage. So I think it's kind of comprehensive. I know it's amazing to be a generalist and work across different categories, but for me, I just have such a deep passion for infrastructure. I love it. I think it's the foundation of everything else. But I think that that resonates with the founders that I work.

Prateek Joshi (28:01.945)
Yeah, one of my favorite go-to lines in a Fafner dinner party, like the world runs on infra. That's what I open with. it's usually deliberately a little push people, but I agree, the world runs on infra. when you invest in an infra company, post investment, across all your investments,

Where did they end up needing the most help from you?

Astasia Myers (28:37.887)
I think it's two prongs, since we are such early state investors, we really do try to give them that 10x leverage in terms of customer discovery and insights. The number one recommendation I would have for any founder across any domain is talk to as many users and buyers as possible as early in the journey. Set a benchmark of, hey, I want to talk to...

20 customers a month just to really inform what you're building, the why behind it and assessing willingness to pay. And so we really try to amp that up for our portfolio companies through additional customer introductions. And the second thing that we help with, because often these are very amazing technical founders, applied researchers is being a thought partner.

an advisor on the marketing community and sales work, thinking through everything from the copy on the website and relating their technology to business value, to thinking through contracting of how to set up your pilots and even introducing them to more hands-on advisors who, once again, have lived this experience of some of the hyper growth and for companies to pass on that wisdom.

So I would say those are the two big buckets that we focus on.

Prateek Joshi (30:05.235)
Recently on X, the platform, there was a big debate on consensus versus non-consensus. And so when it comes to when it comes to infra, what's consensus right now and what corners are currently contrarian?

Astasia Myers (30:13.398)
Now, you can.

Astasia Myers (30:24.641)
Yeah. think that, you know, consensus is that demands on compute and data are growing and that even smaller businesses who are have an AI aspect of their product are going to have more demands on compute and data. I think that's fairly consensus. It's actually funny. Even yesterday, I was talking to a seed stage startup that's

building agent technology and they're just like, my God, our data volumes. I was like, yes, yes, you're seed stage, we understand. I think contrarian is that, infra founders today are the grittiest founders out there. And I say that because it's really hard to find categories in infrastructure that are big enough where you can go build a really con...

competitive company where there isn't already an incumbent that's executing incredibly well. know, a lot of the the growth stage and for businesses still have really high NPS. People really like the product they haven't given up. They've built a multi product portfolio. They're hyper aggressive and often the founders are still at the helm. So I think that if you are an infra founder today.

You really got a lot of courage, honestly, go after this because you have the growth stage companies, which could be the infra players. You also have the model companies that are expanding scope in terms of their, dev tool offerings. And so, I give a lot of credit where credit is due for infra founders right now.

Prateek Joshi (31:49.767)
Right. Right.

Prateek Joshi (32:09.241)
All right. And yeah, think Infra founders are some of the most hardcore people that I end up meeting, is, I agree because it's really, it's tough, but also the value creation is spectacular if it works out. Now, when it comes to investing, every sector, every stage has tourists. Once in a while, just, the wave comes in and they do their thing and they're out. So when it comes to Infra,

what's a tourist trap and also the areas that you think are tourist traps and everyone should avoid.

Astasia Myers (32:48.397)
I think the trap is this belief that if you have a really successful open source project, it's inevitable that you can go hold a competitive business around it. You have the distribution, you have a million devs using it. They all love it. my God, this is the best thing since sliced bread. And I think that's a tourist trap for people that come in and out of infra is

Prateek Joshi (33:03.171)
Bye.

Astasia Myers (33:18.017)
Dev love on open source does not necessarily convert to an enduring large scale business. I think another tourist trap in infra is thinking that developer productivity improvements are a strong business need. We'll buy your dev tool. It's kind of a soft reason.

Prateek Joshi (33:38.017)
Right. Right. Right.

Astasia Myers (33:42.83)
Right. It's like, yeah, that's, that's great. You know, you increase developer velocity by 20 % and developers are paid, you know, 150 plus a year. And so this is how we justify the ACV of our dev tool. And of course everyone will pick it up. You know, I think that that's not necessarily true. Like that's could be a component of the business value you're bringing, but

Are you decreasing costs? Are you part of the deployment pipeline? Are you doing something that has never been done before for the team? Are you increasing revenue somehow? So I think you need firmer business metrics that you're moving. And as a founder, you should be assessing that in your customer discovery, right? Even before you start building. It's like, if I had this, what are the business metrics that I can improve upon?

because dev productivity is, we've seen time and time again, isn't sufficient.

Prateek Joshi (34:43.097)
That's amazing. And the developer productivity is such a huge thing. every week, I feel like I end up reminding someone, you want them to care, but just people just don't. Because there are other burning needs that they're spending time on, that increasing the productivity by 23.9 % is just a noble intent. It just doesn't work. All right. Let's move on to the market structure and GTM.

for infra. Now, as we sit here in 2025, how do you think the go-to-market motion is playing out for infra startups? And also, assuming it's bottom-up right now, when should they layer in enterprise sales? What's the rule of thumb to figure out, OK, now we've got to get in like real sales people?

Astasia Myers (35:39.744)
Mm-hmm. Yeah, I think and for I mean, there's two buckets of in for a company right now. There are the bottoms up go to market motion companies that can be distributed through open source or community led efforts that you tack on sales assisted once you know, you're seeing that these companies are

using you across the organization and you try to have like an upsell motion. And then you have infra businesses that have a smaller group of targets that they can sell into kind of, you know, probably somewhere under 5,000 customers where it's much more of a targeted account-based marketing sale where you do founder for sales first and then you build out a sales team. So we're actually seeing

both types of infra companies today. Of course, in this latter bucket, you're going for 200,000 plus land deals to justify bringing on AEs earlier in the company's life cycle. And then from the bottoms up, we're seeing pricing around 30 to 75 land deals. But it is interesting. I didn't think we would see the revival

of the sales led motion and infrastructure. But we are seeing some of it with these RL as a service companies. And it's quite interesting. And it's really a function of who is your ICP and how do they procure software and how do you get in the door as fast as possible.

Prateek Joshi (37:25.645)
Right. And let's talk about the pricing mechanics in infra. I'm a huge pricing nerd. I don't know why it excites me, but it just, find it fascinating because there's a very interesting, nuanced engine that people have to build. And in the early days, it doesn't matter. It just, it's fine. But I think it becomes a huge lever as you, as you grow up. So when it comes to pricing in infra, how do you think about it? And how do you advise companies to

Astasia Myers (37:47.437)
Thank you.

Prateek Joshi (37:55.523)
help them figure out pricing.

Astasia Myers (37:58.001)
Mm-hmm. Pricing is one of the hardest things to do. Oh my God. Especially as an early stage startup, right? Pricing is a moving object. I think there's three things, which is one, as a founder, expect it to change, right? Two is try to, when you're doing your pricing discovery, anchor really high, right? To see kind of the upper bound and then

Prateek Joshi (38:01.322)
Look.

Yeah.

Prateek Joshi (38:08.322)
Right.

Astasia Myers (38:26.925)
customer kind of rationalizing where they would land. Three, really tying it to the business value that you're delivering, which is beyond developer productivity, but cost savings, time to market, revenue lift. And the fourth thing that we tend to think about is, is there a product that's adjacent but related?

to your offering that you can use as an anchor point to say, if you're paying $100 a year for X product, how many relative dollars would you be spending with us just to give a sense of willingness to pay? often infrastructure is priced around usage, networking, storage, compute costs. But what's been interesting with

this new wave of kind of these dev tools that are AI first. You can see that historically pricing has been around gross margins as well. And what you want to see from a gross margin profile to become a publicly traded company. And now that is less of a concern as compared to distribution usage with the belief that AI model costs will continue to go down, which we have seen 10X year over year.

and think we'll continue to do so. But I would say that that shift in winning the hearts and minds of users has influenced some of the pricing, which has been a huge change over the past two years.

Prateek Joshi (40:09.539)
Amazing. And also for platform companies, how early do you think they should push the ecosystem like SDKs and integrations and partner motion versus deepening the wedge that they already have? They entered the market with the wedge, it's strong, and should they push forward in that direction or should they spend time on these other things?

Astasia Myers (40:39.105)
Some of it is a function of what they're seeing in their core business and where they're being pulled. As a small team, you only have so many hours in the day. And if you are seeing incredibly strong pull for the main product, it's really useful to try to get to a few million in error around that core service before expanding scope with

an ecosystem play or a second product line. What's really important is every exceptional infra business has become a platform with multiple different products. And if you are building an infrastructure service and you don't think that is possible out of the gate, you're probably not going to go the distance. So you should be looking for those opportunities of how to build an ecosystem. What else you can tie into what else you can own over time.

and really focus on the core for the first few million. But if you want to become huge, you do have to expand scope. Like I think that's been a huge learning and infrastructure over the past five years is, you know, in the networking times, you could have the best switch. You could do incredibly well. You could go get acquired for a few hundred million to a billion, or you could build like one core database. but now you have to have multiple products. We've seen this with Databricks. We've seen this with Versailles.

You want to go to the distance and become 500 million plus ARR and be a candidate to IPO. You can't just stick to your core.

Prateek Joshi (42:15.329)
Right. Yeah. Let's move to the lessons learned from your own investments. And you work with great companies. I'm sure you have companies that didn't pan out. So I would love to understand across all the companies, spanning databases, open source, ETL, all these companies, what repeated pattern has separated outlier companies from

the just merely the good ones.

Astasia Myers (42:46.379)
Yeah. What I've observed across my portfolio is that the outliers consistently don't just ride a wave, but create a category. They really start to either create the category out of nothing or to become so prominent that they redefine how the category is thought about. And so they do this through the product experience.

data flywheels and networks effects and this platform expansion we just talked about. But we really own a subject matter and are considered the experts in that space. The other thing that is I've noticed as a pattern is, you know, it's really critical that the people behind the companies have extreme learning velocity. We highlighted this earlier in the conversation, but they

have to have an incredible ability to take feedback from customers, their VC partners, their go-to-market partners, and their employees to make smarter and smarter product and strategy decisions. The founders who have done good, not exceptionally well, don't have the same feedback loop velocity. And so they get

They can't execute as fast and the market sometimes moves beyond them.

Prateek Joshi (44:17.689)
And now that for all the infra founders and would be founders listening, what can you name three infra wedges that you would like to fund today?

Astasia Myers (44:30.497)
Yeah. So as I said, we're really excited about observability, both the re-imagining of traditional observability, like data dog reinvention. Two is thinking about AI agent evals and observability. I this is the next phase of how we create really amazing production systems. And three, we think that it's quite interesting to think through the rise of

voice technologies, both model, middle layer, and I'll put it out there, I know we're talking about intro, but app layer. anything in the voice stack, we published a piece on it recently, we're quite excited about voice.

Prateek Joshi (45:14.689)
And what's an infra category that looks commoditized but still has room because there's just so much work to do?

Astasia Myers (45:23.743)
that's a great question.

Astasia Myers (45:32.622)
So AI inference has been an incredible space over the past few years. I was humbled to partner with Modal when I was at my last firm. You have other vendors in that space as well. They're doing incredible things. We can imagine in the span of time as there were more companies that have their own specialized models.

or third-party model services, they'll want really tuned inference stacks for their models that the more broad-ranging platforms can't offer. So that's one area that could be very interesting is a super customized inference stack for your model in the cloud or also on the edge where you have different compute resources and are quite constrained.

that could be a really interesting, underappreciated market opportunity. Kind of see it little bit with fall, right, as this kind of second wave, but think about it like a third wave, or you get that for everyone.

Prateek Joshi (46:46.979)
All right. I have one final question before we go to the rapid fire round. AI is moving so rapidly and infra is becoming AI native. So what AI advancements are the most exciting to you today?

Astasia Myers (47:06.733)
I think it's all this stuff with RL. I think this is kind of the rocket fuel for AI and AI agents. We're incredibly excited about that space. What else is really? As a completely frontier opportunity, we think that energy management is very excited for AI. If we continue to have the breadth of models,

We're going to really have to think through how we manage energy and alternative sources. We work with a company called Crusoe that is really trying to help be a more green Neo GPU cloud. And then the third thing we're pretty excited about with AI is the app layer. I know we didn't talk about it today because I'm much more in the infrared space, but voice.

AI applications across healthcare, insurance and financial services, field service teams, the ROI for using voice AI agents to automate a historically manual process is quite exciting.

Prateek Joshi (48:21.977)
Amazing. With that, we're at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready? All right, question number one. What's your favorite?

Astasia Myers (48:34.903)
God's Bankers by Gerald Posner. Probably not what you expected.

Prateek Joshi (48:39.897)
Amazing. Amazing. I'll check it out. I'm a huge reader, so it's great. All right, next question. Which historical figure do you admire the most?

Astasia Myers (48:50.381)
Steve Jobs and because Apple.

Prateek Joshi (48:54.869)
Right. Next question. What's the one thing about infra investing that most people don't get?

Astasia Myers (49:04.641)
best technology doesn't always win.

Prateek Joshi (49:07.161)
I think we talked about it earlier but what separates great infra products from the merely good ones?

Astasia Myers (49:16.813)
ability to become a multi-product company.

Prateek Joshi (49:20.597)
What's the most underrated in Founder Autocode?

Astasia Myers (49:25.993)
Right now, I would say it's the Gen Z builders.

Prateek Joshi (49:31.881)
What's a signal that makes you want to actively pursue a company?

Astasia Myers (49:37.818)
exceptionalism.

Prateek Joshi (49:41.118)
What have you changed your mind on recently?

Astasia Myers (49:44.299)
Domain expertise trumps velocity.

Prateek Joshi (49:48.513)
Interesting. Maybe part B, 10 seconds. Like, why? That's interesting. Why?

Astasia Myers (49:54.91)
you can be the smartest person in the room, but if you're not moving fast enough, someone will catch up and they'll have the customer base and they'll get that data flywheel and insights from them that they can surpass you.

Prateek Joshi (50:08.978)
What's your wildest infra prediction for the next 12 months?

Astasia Myers (50:15.553)
Gosh, well, let's put it out there that Anthropic buys a company for over a billion.

Prateek Joshi (50:24.04)
okay. All right, final question. What's your number one advice to founders starting out today?

Astasia Myers (50:30.711)
Talk to as many potential users and customers as possible even before you start building anything.

Prateek Joshi (50:38.937)
Amazing. This has been a brilliant, brilliant discussion. I loved your depth of thought here. Clearly you spend day and night only thinking about infra. So it's always nice. It's always fun to chat with people. So thank you so much for coming onto the show and sharing your thoughts.

Astasia Myers (50:47.917)
working track.

Astasia Myers (50:54.658)
Thank you so much for having me at a wonderful time.