Joseph Ruscio of Heavybit joins Nick to discuss DeepSeek’s Impact on Other LLMs, AI’s Effect on Software Engineering, Nvidia’s Future, and the Next Wave of Developer-First Companies. In this episode we cover:
- Impact of AI on Venture Capital and Developer Tools
- Challenges and Opportunities in Scaling Developer-First Companies
- Utilizing GitHub and Other Tools for Market Insights
- Recommendations for Founders and Investors
Guest Links:
The host of The Full Ratchet is Nick Moran of New Stack Ventures, a venture capital firm committed to investing in founders outside of the Bay Area.
Want to keep up to date with The Full Ratchet? Follow us on social.
You can learn more about New Stack Ventures by visiting our LinkedIn and Twitter.
Are you a founder looking for your next investor? Visit our free tool VC-Rank and we’ll send a list of potential investors right to your inbox!
0:17
Joe Ruscio joins us today from Portland, Oregon. He’s a General Partner at Heavybit, an early-stage venture firm investing in developer-first startups. Heavybit has backed over 60 companies globally, with notable investments in unicorn companies Snyk, LaunchDarkly, CircleCI, PagerDuty, and Netlify. Before Heavybit, Joe uses his many years of experience in academia, startup management, and venture capital to help practitioner-first teams scale their go-to-market strategy and their organizations. Joe, welcome to the show!
0:48 Thanks. Super,
0:50
super excited to be here. Talk a bit
0:51
about your background, like, give us the two minute summary of your path to heavybit, yeah,
0:56
yeah, I think. And like, the trope, I think, is I took a non traditional path to venture, as
1:02
they say, which I haven’t heard that. You’ve probably never heard that. Yeah,
1:06
yeah. So I guess my my background, I started out as a technologist, software engineer. I got really excited about programming when I was a kid in high school. I mean, honestly, it was kind of like the only class I enjoyed. We had a programming class was doing AppleSoft basic, Apple two ease. So I went to school, yeah, just to be, I want to be a software engineer. I got a gig Motorola, which, at the time, was an industry behemoth, and working on their technology for the embedded software around the cellular base station. And then, basically, pretty quickly, became immediately disillusioned with being a professional programmer in a massive company, and just like what that actually meant, and how hard it was to have agency or get things done and and so then I decided this was in 2001 is right, literally, as a.com boom was crashing. Well, I guess I got to go to grad school if I, if I want to hack on new, weird stuff and the frontier technology, I guess I’ll off to go to grad school, and I, so I started working on a PhD in pure science at Virginia Tech, and just serendipitously, ended up there at the time their high performance computing lab was was coming together, and joined up there as a research assistant and was got to be part of the team that built system 10, which at the time was the third fastest supercomputer in the world, built out of Apple hardware, and using this tiny Israeli company called Mellanox, is interconnect technology, which is now owned by Nvidia, and is part of what powers their interconnects for big training runs. So I was doing that, and got then the, I guess, the startup bug, and decided my PhD advisor, I decided, like I was work we’re doing, because my research was in system software, like reliability software for these large scale, couple 1000 computers, and decided to try and start a startup and productize the research I was doing. Learned a ton of lessons along the way on that one, first and foremost, make sure the TAM is big enough. Turns out, high performance computing is a very interesting area to write, build products, but not an area with a ton of budget for software at least, at least back then, but really got hooked on on building and starting companies. So then that led to a series of the next company I worked on was in data center orchestration, automation in the mid to late aughts. I was doing that when cloud computing, I think, was like mid 2008 I was at a conference in Warner Vogels from Amazon was giving a talk about this EC two thing they had just was year old, maybe, and I just got obsessed, because as a software developer, and even when I was working with you on data center orchestration, was all like, how do we get hardware provisioned. How do we get things in place so that software can be deployed on it? And the whole DevOps movement was happening, and the Phoenix Project and these kind of that book, these great things, and I just get obsessed with, like cloud computing, if, if compute can be ephemeral and on demand, and developer, as a developer, if I can, just like, hydrate the compute I need with an API call on the band and scale it elastically like this, just completely changes the bottleneck and what it will take to get go from a business requirement to working software in production. And now the bottleneck is not going to be requisitioning and waiting on hardware to be drop shipped and provisioned. It’s going to be how fast can a developer move so that, like all of a sudden, developer velocity becomes the single most important thing to almost any organization. What does that mean? And so my last company i co founded was one of the very cloud native observability monitoring companies called the brato. We saw the need. If you were going to be going all in on cloud like you were, you would need to watch that and monitor it that were built for the new paradigm. Sold that company four years later to SolarWinds as part of their strategy to build a cloud native business unit. And I got to spend like, three and a half years there helping to build out that strategy. We did like, four or five other acquisitions and a roll up along that way. I guess the my connection and heavybit had started around the time I started my last company. We went through heavybits, go to market accelerator, my company and I had stayed in touch, and it was. Around that time that the original founders and partners at heavybit Were looking at expanding the scope of the firm and starting to raise actual capital to invest, as opposed to just like kind of operating as an accelerator, having at that point been part of venture backed startups for over a decade, I had a lot of opinions about how I the art and craft of venture for early stage companies like the ones I had been building and founding like could be that there could be a better class of investor for those founders, and that’s really how I ended up there, and that was just over almost eight years ago. And so I’ve been doing that ever since, which, you know, it’s still, I still feel like I’m new to venture, but it’s but, yeah, doing it for almost eight years every
5:38
day Joe feels like a new day, yeah, so talk about the thesis and the investment pro approach at heavy bet. Yeah,
5:47
what I like I love about heavy bet? Why I’m here and I work here every day, so we have this very deeply vertical focus on products used by technical practitioners to build, operate, maintain, secure software. And pithy your way we say developer first, because that’s like the primary technical practitioner, but it includes like stories, DevOps, specialists, data engineers, security engineers, designers. Are the one of the founding theses of the firm was that every single company over a certain scale, it ultimately is a software company. And like the example I always love to use, is one of the best, one of the world, a world class. One of the best software operations in the planet is Domino’s Pizza. And if you look at like what they’ve achieved with like the pizza tracker and the omni channel and mobile to watch to digital, they’re just a world class software operation employs 1000 software engineers deployed into hundreds of services and so, yeah, so it these products only serve the entire global economy. There’s a certain amount of revenue over which you’re making. I haven’t I haven’t dialed in exactly what this number is. If it’s 50 million or 100 million, but over a certain amount of revenue, I don’t care what your category is, you’re hiring some number of software engineers to build proprietary leverage through software in your enterprise. Awesome, perfect.
7:07
Well, we have to address deep seek right new circle of the week, or the past couple weeks, so they kind of stunned the AI world by launching a significantly cheaper, open source LLM trained at a fraction of the cost of its US counterparts. Do you think that they’ve, they’ve actually innovated on the architecture of llms, or is this just a distillation layer, giving the appearance of efficiency while, while leveraging other models, like, where do you shake out on that? Yeah,
7:34
I shake out. I think it’s a qualified I think it can be both, right? So I think clearly a number of things. I think there was a couple of technical achievements. One of the things and so on. My prior life, I was a systems programmer, like I said, HPC and telecom and then even data center. So I was always on the low level. And part that part of me really appreciates one chunk of what they did was they just took some people who have deep expertise in how computers work, how memory gets moved around. One of the challenges, I think, with modern computing is we leverage these abstraction layers to get tremendous productivity gains where it’s like, my day to day, if I’m just like, stitching some APIs together, or build an e commerce website, like, I don’t need to worry how the chip works. Or, like, how memory is being moved the bits are actually being moved around. When you get into the kind of things, like aI training runs, or, like, heavy amounts of inference, I think one of the things they did is they just looked all the way down through the abstraction layers and said, Hey, wait, where are we? Where are we wasting system resources? And they identified a handful of places where they could just dramatically cut the resource usage used right in terms of pressing a key value cast. I mean, just some set of data from 32 bit registers to eight bit registers and not suffering dramatic kind of accuracy challenges. So there’s that aspect. The reinforcement learning piece is fairly novel as well, and really interesting has already been replicated at some like Berkeley or Stanford, where they’re able to take a relatively straightforward model, generative an LLM, and through some of the same techniques, endow it with quote, unquote reasoning capabilities, right? And so I think that’s another kind of interesting novel. And look, that’s probably, I mean, a lot of speculation is like, well, this is exactly they were doing inside of open AI or over at anthropic, where they’ve already built these reasoning capabilities already. But what’s fascinating, I didn’t know how expensive that was or how hard that was and and it turns out, like, maybe not that expensive or not that hard. I also think it seems relatively likely and obvious that they did leverage distillation to, you know, quote, unquote cheat, if you will, on training the model. But there’s something to admire about that. I mean, I wouldn’t advise anyone to break their terms of service with whatever vendors they have and similar. There’s, there’s news that, apparently, Facebook torrented 80 petabytes, or something along those lines, of books for training, llama and like. So it. Clear, like, all around the spectrum, people are sort of pushing the boundaries in terms of, like, training these things. And I think we’re still figuring out as a civilization with this kind of, like new technology, like, what what should be the norms, what should be right, what should be wrong. I think the idea that you can use distillation, if you want to go back to an academic sense, I’d actually like to know for sure. Because if you can use distillation to the extent that people suspect they have that again, is something that is going to dramatically reduce the barrier and cost of building out newer and more powerful, powerful versions of these, which I do think would be long fundamentally good.
10:34
You know, Joe, it feels like historically, China has been really good at kind of reverse engineering and copying a lot of American technology. In this case, do you think a lot of the American or headquartered AI companies are going to copy the techniques that deep sea has used? I
10:50
don’t have a strong position on this. There’s a couple different ones, but it what is interesting is the extent to which they open sourced these capabilities. Right? Because one approach would have been to kind of make these claims and demonstrate but not how they got there, right? And by publishing things as as transparently as they did, and providing everything in open source or open weights, that has made it very easy, like I fully expect, that the the labs, the American labs, or European labs, will absolutely, very quickly adopt these techniques and be testing and experimenting and expanding upon them. And hard to say maybe they already knew about some of them or not. But I’m sure this is only going to that pace, and that, to me, is, again, it’s just as a pure technologist is good, right? Like, I Yeah, you know, generally speaking, we’re at the kind of just earliest people to use, like, the earliest innings. Like, we’re barely even into batting practice on on this stuff and figuring out where it’s gonna go. And the more we accelerate by sharing and learning these techniques, like, the better long term it’s gonna be for all of us. So I was really excited. I was really excited when meta llama was kind of the first real big step in this direction to show that, like, hey, if as long as somebody has, like, the kind of the capex appetite, you can come up and do this and provide an open weight model and yeah, and now deep seek is even that requirement for capac even though, again, the number they gave was kind of sensationalist in the sense that it was like maybe just the final training run cost. There was obviously a lot of but the next set of people don’t have to do all the experimentation to figure that stuff out. So there it’s just gonna the cost will go down and down and down. As a firm when, when chat GPT launched, we pretty quickly decided two things. One is that this was just going to be dramatically, dramatically like transformative across like all aspects of the companies we invest in and beyond their customers, us as people. And the second was that we were had a very strong belief that this was not ultimately going to be controlled by one lab or two labs, which I think at the time seemed maybe more likely, just because it would be the first time in the history of it, that something like this actually played out that way, right? And so what we have just seen repeatedly in a innovation cycle like this is that a value kind of accrues in this barbell shape at like the sort of very bottom layer, like your Intels and videos and at the top. And the kind of ISV application software, the customer requirements actually meet the software, right? Yep, and these kind of middle layers invariably get commoditized, and that’s happened with cloud computing. It happened with virtualization before that. It happened with PCs before that. It happened with micro computers. I mean, you can just go down the line. It has always played out that way. And there were certainly, I think, some kind of interesting questions about how it would this time. But yeah, we, long story short, we were very confident that we’re heading towards a world with lots and lots of different models, lots of them being open and applied to different use cases. And it would be, I expect we will see the typical enterprise to be what I would think of as multi model, where they’re even within one shop, deploying a mix of on premise, open models, fine tuned for different tasks in different places, and mixing that with API inference calls up to potentially even a mix of the different labs and providers.
14:30
I mean, Joe will will the primary division be kind of an open source versus closed source approach? And are you seeing a lot of developers move to open source in light of llama previously, now, now deep sea.
14:45
It’s super early. I think what you’re seeing well two things. One is one of our one of my beliefs, if I fast forward like five years from now, even today, but certainly in five years from now, as the models are advancing, both the closed source or hosted lab. As well as the open source ones as an enterprise. You see, one of the challenges with this right is, for me, really good value add of these models. I have to take my proprietary data and put it in queries and prompts, or maybe hand it to agents right and ask it, or particularly in agents, I have to ask it to log into other systems of mine. And so if I’m given a choice between two models, one of which is hosted on someone else’s metal cloud somewhere else, and I have to shuttle all my data to it. I have to give it remote access to my systems, and it is 10% better than another model that I can host on premise. I can tightly dial in a trade off. So first of all, like no data provenance problems, security problems, I can tightly tune my inference costs on a per use case basis, right? I can deploy different, more expensive inference for more important things. I can deploy cheaper inference for things where the accuracy is not as high, like I’m gonna have a very strong predilection for that latter solution, because it just, it solves a lot of problems. It gives me control on my cost curve. So the only reason I’m going to go it
16:08
prevents the data leakage, which is, right? I think a huge risk here, right?
16:13
A huge risk. There’s a parallel to this coming from my background in like, observability. So if you look at things like synthetic monitoring, or like core to cloud infrastructure, like what a Datadog does, or my company did. Companies are fairly I mean, there’s a cost curve concern in terms of data. It’s like, well, I’m sending up numbers about performance up to your service. I think it’s very illuminating that if you look specifically at log management, there was an entire generation of hosted log management companies. Suma logic managed to go public. The rest of them. We acquired two of them at SolarWinds logically and paper trail. There was several others that got acquired in but Splunk has really 20 some years in absolutely dominant in the space and was not disrupted to the extent in any other observability category. And a huge part of that I ascribe to when you’re talking about shipping raw application logs up to someone in the cloud, that is a whole different level, because it contains customer data. It contains company data and and it was, you see this split where, for all other kinds of monitoring data, people happy to send it to cloud, log data, raw log data. And I think, I think llms, and the kind of, I mean, llms is literally like, we’ll just take the company knowledge base, take the company wiki, take all our customer data, put it in, put in the prompt. I think there will still be a set of use cases where it’s like, hey, we have got to have the absolute bleeding edge, best model working on this. And so there’ll be affordances for but I think, I think the default will be on, like, you’ll have to, like, you’ll have to promote to that. You’ll have to make a conscious choice to be like, Yep, we’re gonna go outside the firewall, as it were, because the requirements demand it, right, not because that’s like, the default. How do
17:53
you think about segmenting the opportunities for developer first, you know, AI investments, the
18:01
highest level segment, for me is augmentation versus automation. And so I do think there is an ongoing discussion, which I think is very valid, although I think there’s a wide range of answers to varying degrees of accuracy, about like, how many software engineers is there going to be in 10 years, right? Yeah. And you can find absolutely rational, intelligent people right now to argue that the number will effectively be zero,
18:30
right? And if there are some, what are they going to be doing? Right? What are they
18:34
going to be doing? So? And that, to me, that’s the Ottoman automation side of the argument, right, which is just like, Oh, these things are these models are gonna be so good at generating code or debugging code or running code or whatever that, like, like, literally, there’s not gonna be humans involved. And, you know, it’s gonna be this kind of nirvana. And, you know, I think historically, it’s interesting to me, because I think historically the business and sort of always view as a former engineering leader right running or in an engineering organ interfacing with the rest of the business. A lot of the other business leaders have always kind of viewed engineering departments as this, like black box, that they don’t really understand how it works, and it generates running code, and it costs a lot, and it complains about free soda a lot, and it’s really annoying. And so this, like, this notion that there’s a new kind of black box which doesn’t require any free soda and doesn’t actually complain that could generate code and you don’t really know how it works, is, like, very, very appealing. And I get that, you know, for me, I’m I lean much more heavily towards the augmentation side in that all the most senior kind of, like, like 10x engineers to use, take an overused term that I know or are clearly all leveraging AI daily in, you know, it’s not just, it’s part of their it’s what they it’s part of what they do, right? It’s just, like breathing like it’s just part of their routine, and it’s making them dramatic. More productive in a number of ways, but outside of some very basic use cases, I think it’s falling well, well short of like, Oh, we’re going to fully replace these people, right? And and so I’m very much focused in that distinction with products. Tell me how you’re going to give human superpowers. Tell me how you’re going to enable humans to do with one person what it used to take 10. And what’s super important there, if you, if you’re a student of mythical man month, right? Like one of the single biggest challenges, this is why Amazon has two pizza teams. You know, this is part of how, like Elon Musk, runs his organizations. Communication paths are like the killer engineering orgs. And so it’s not just that, like, oh, one person can do the work of 10. So I’m paying one salary instead of 10 for every n. It’s also, I’ve quadratically reduced the number of communication paths that have to take place to get things done. And historically, that’s why enterprises are so slow. It’s not because they want to be, or they’re not full of smart people. It’s the communication burden. And so I think that’s gonna be really fascinating. And is a benefit to this augmentation that a lot of people aren’t really kind of grasping yet, is if I can, if I have 10 people, it’s not just cheaper. So we can literally move faster than the 100 person, or can, if we can output the same amount of code as 100% or right? So, I
21:24
mean, related to that in your earlier point at the top, about the early in your career, shifting from provisioning to developer efficiency. Do you think that continues in this LLM world? Is the bottleneck still developer
21:37
efficiency? Yeah, I think it is. And I think, I think llms are just one of the most impactful things in terms of increasing developer velocity, although I want to be careful, you know, an interesting thing is that I think historically, if you look at the amount of time, particularly like anything a mid level or above, like junior engineers, are a separate thing, and that’s also like, kind of a separate interesting question about what that role looks like, or exists like, or anything, and how you have a pipeline to have mid to senior level engineers in the future, but looking today, if you look at your everything from mid up to senior and beyond engineers, which is where really most of the work gets done, they don’t spend a tremendous amount of their time, on Average, hands on keyboard, generating code, right? Like generating the code, you know, there’s all kinds of activities downstream, which I bulk group in the developer velocity, right? Like, I’ve got to, I have to understand the business requirements, right? I have to have a notion of what we have to create. I have to write the code. The code has to get it and verified, right? It has to get deployed out in production, it has to get, like, monitored and run. There’s these systems have entropy. Like, I can have I can just say, hey, the software is done and not want to ever touch it again. And I’m still going to have to be kind of continuously, like, upgrading systems. You know, new vulnerabilities have come up, so there’s a lot, and I think llms can help with a lot of those spots. I think generating code tremendously helped there, which is exciting, but, yeah, I do think of it as, like a dramatically increased the velocity of developers and any of these places where they and I think it’s also super fascinating in terms of activation energy, is another place I really like helping where, oftentimes, particularly in a larger organization, people will see something like out of place or or they’ll see something that they could do a little better. And just like the getting to the point where, like, Oh yeah, I’m gonna, I’m gonna sit down and do it. It can be challenging. And so I think llms dramatically, or even just prototyping a new idea, right? Maybe you’re like, oh, I have this idea, but I don’t have time for this week. I think llms, like, dramatically reduce that activation energy required, because so easy just to get going on something. So I think that’s exciting, too, Joe, there was
23:51
a big dip for Nvidia stock this past week. Are you bullish or bearish on Nvidia? By the
23:57
way, this is not stock picking advice, but I do think it was an overcorrection. I think was an over correction, and I feel like it swung back pretty quick, but for a number of reasons. One is, I think people again, I think the actual number provided was not it was sensationalistic. It was click Beatty. I think it was directionally correct, and it achieved its goal. But like, it just actually wasn’t that cheap to do what they did. And then the other thing is, like, I think there’s training costs and inference cost. And like, the more widespread these models, like, I think the appetite for meaning, like, actually, like, I’ve typed something in a prompt, the model runs it and gives a result, right? I think the appetite for inference is almost limitless, and that happens on these chips, right? Like, I think so the cheaper we make the models, the better we make the models, the more inference we want to do. So I think the market was looking at like, oh, there’s going to be less $50 billion capex layouts for training, right? Funds. But I think there’s a whole lot of people who don’t own GPUs today who will own them in the future for inference purposes. I
25:06
mean, they’re going to be embedded in just about everything, right? Yeah. And
25:09
if I want to run an open source model, like deep sea, like, I have to have a GPU to run it on, right? And, like, can I rent that from someone? Sure, but at some level, I’m probably going to, I’m probably going to want to own it, or whatever. My rental demand is gonna drive demand for the people providing that access to me to buy it. So I Yeah, I’m sure some people bought that dip will be very, very happy. I mean,
25:30
a lot of people are reacting strongly to the whole deep seek news. There’s articles on streetcary from Ben Thompson. And I got an email from an LP the other day that was asking about, is this a mass extinction event for VCs, invested a lot of money llms And yeah,
25:47
that might be a different question than Nvidia. But yeah, I think, Oh,
25:50
I agree. Yeah, I agree. But yeah, I would. I mean, where do you net out on that? How do you think this affects, you know, our asset class? Yeah,
26:00
well, I think for the asset class as a whole, I think there’s, there are a set of orthogonal questions about the asset class, right? I do think the asset class is in a very interesting point right now, in terms of, like, I mean, one thing has been fascinating. I was part of my first venture fundraise from as from the operator side in 2004 so, like, I’ve basically been in the kind of venture ecosystem for just over 20 years now, you know, as a 12, as a raising money, and now, like I said, the last eight, kind of on the investor side. And it’s staggering how much the industry has evolved changed in that time frame. And multiple evolutions, right? Like it’s actually for asset class, for an incredibly illiquid asset class with incredibly long time feedback cycles. It the evolution. The evolutionary cycle is actually very quick, and I feel like we’re at another one of those moments right now where there’s just a number of things. I mean, AI is absolutely one of them, the kind of capital flows in from allocators, and how they’re sort of centralizing in the mega funds, both dropping kind of dramatically from peaks a couple years ago, interest rate being out of a zero interest rate environment for the first time in like 15 years. So there’s a number of things going on. So I think what does all this mean for ventures and asset class is a very interesting question. I think there’s probably less kind of certainty. I mean, I have my own ideas. I think generally speaking, there’s less certainty about where things will be in a few years from now than probably there has been for a while in the asset class. But all that being said, and AI specifically, I think, is like, volatility can be a very powerful thing for asset manager, if you if you play it correctly, I guess is what I would say. I think we’re in creative disruption, both in terms of AI as a fundamental technology wave. I think there’s some creative destruction both in like, what is it going to mean to be a good VC in a few years? And so I think there’s an opportunity for new managers, or even semi established managers to adapt and be in a much stronger place than they were. Like, paradoxically, I’m not sure what your opinion was, but I always tell people 2021, was my absolute least favorite year to be a VC. And like, the system was just like a wash in capital, and I didn’t feel like there was any real art or craft. I mean, there’s some art and craft to like, how do you navigate tsunami of capital? But in terms of, like, most managers were like, well, you just deploy it as fast as you can. Like, yeah, with with kind of no discernment and so, and that, to me, is not fun. I mean, every bit we’ve always been a high conviction, kind of concentrated in that sense, kind of a classic model where we want to work with like, 810, maybe 12 companies a year tops. We want to have strong positions in them. We want to be really aligned with the founders, and we want to work really closely and deeply with them. You know, almost all ex operators. We want to work really closely and deeply with them on like finding product market fit, finding go to market fit. We’re not here to play the casino, right? And, yeah, so I think, sorry, that’s probably, I mean,
29:03
just on that you’re advising companies. Often, these are often developer first companies. So how do you advise a company, a founding team, you know, going from like a grassroots community developer model in gom into more of a large enterprise motion, one of
29:21
the things that found fascinates me and always has and I love about our focus at heavybit, as a former developer, and then as a former founder of companies that sells to developers and technologists, I have a deep appreciation for both how important and valuable those products are, as well as how challenging it is to sell to that market, right? You know, I, when I talk with founders, kind of from more like broad spectrum, or if they’re like, Oh, we just like, we just like, call, or we do, just do this in the email, we contact the people like the traditional kind of top down sales, your typical developer, adopter just is. Like, hates phone isn’t gonna respond to your emails or your LinkedIn. And so how do you, how do you get to these people? And starts with, first of all, I mean, I think we look for founders who have some kind of clear, crisp, like, it actually starts back from the product market fit, right? So there are examples of deploying a lot of money just brute force into selling a product top down. And if you get the messaging right, and the products like, sort of sufficient, like, you can make that work, and you can drive, like, to a real business. That’s, I think, less interesting than to us. To me, we focus on finding founders who have some and they’re typically technologists. They typically have some kind of, certainly, engineering or at least a product background, and they have this very, like crisp, clear view of some better future that should exist, and that’s either through often creating some new kind of category or redefining an existing category in, like, a fundamentally new way. And so like first is like, Okay, do you have something like that? Do you have, as a founder, like, an incredibly strong point of view, right? Like, it’s not enough for for me or us to to say, like, Oh, of course, you know, it’s like, if someone’s like, oh, well, AI is super interesting and big and AI people probably need this. So I want to do that that’s not interesting, right? Like, you have to come in and kind of explain to me what you understand that no one else understands right, and what you’re incredibly opinionated about and correct about, because that we talk about all of our write your manifesto right like you should be able to write down in semi abstract terms, like a movement like the future of software development should look like this, and it should be something that you can rally kind of community and developers around, and they should all agree and say, Yes, that’s right, like this, this part of building software should would be better. Like that. It would be and then it falls out of it that, like, if you do that, if you can, kind of, like, create this movement, this community around this better way of doing things, well, clearly your product is going to be the best one for me to go use by default, right? And we’ve had a number of companies, you know, sneaked in, kind of famously launched darkly with with feature flagging, replicated with enterprise ready software. So we’ve had a number of companies kind of follow this, once you’ve done that, right? So I think having done that, then the question, I guess, zooming out, the other thing that super, super important, and this is where, this is the very first thing that many of our founders, especially the first time, want, particularly first time founders don’t actually realize, at scale, all these businesses look the same from a revenue perspective, and that there’s a power law in that 8080, to 90 plus percent of your revenue is going to come from your from your top 10 to 20% of your customers logos. And you can go look at the quarterly the earnings reports for like Datadog or hashic. Hashicorps just get acquired. But prior to that, you know, PagerDuty. And if you dig in there, one of the things they always report is like, oh, we have 100 logos over a million or, you know, or 200 logos over 10 million, whatever, and that’s important because it’s the velocity and that that’s where all the revenues, that’s the revenue engine. And so growing that number is super, super important to the health of the business, but particularly in developer first companies, it’s so if and those the top 10, 20% of those customers, they’re on multi year contracts. They’re negotiated by humans like you have an enterprise sales team that runs those like it very much looks like the traditional top down enterprise sales. But what’s interesting is like, where do those leads come from? And if you have like a developer, first bottom up go to market, the bulk of them come from your your long tail, your your bottom 80 to 90% of your customers, who are only 10 to 20% of your revenue. Percent of your revenue. They really don’t move the needle from a revenue perspective, but they do like air cover, awareness, social proof. There’s some amount of farming there, where you get some startup right out of YC, or something is signs up and uses your thing, because it’s the easy thing to self serve and just get going for 50 bucks a month. Fast forward a couple you’re raising a couple 100 million or whatever, and signing up for your enterprise plan. And so if you can pull this off, where you span this this range, and there’s a ton of art and science to how you do that, just in terms of your packaging and pricing, you have to meet each kind of tier of customer, where they are. You have to like like, you want them to feel pulled up into the higher levels of pricing, not forced by your sales team. And so there’s a whole bunch that goes into that. But if you can pull it off, it’s like, this is part of the reason Datadog is just such a monster, is because they’ve, they’ve done a really, really good job of this, and it’s incredibly hard. It takes all the oxygen out of the room for some upstart like, there really has to be, you know, there’s no way to disrupt their go to market, because they they own the whole go to market. It makes it dramatically love, drops your CAC. It improves retention expansion, like there’s just a whole bunch it really and it really aligns, you ultimately, with the end user. Because at those lowest levels, let alone the free plan, you have to make. Happy, right? It’s not the CIO said I had to use this. It’s that, like I chose to use it. So we have a really strong predilection and kind of thesis around around those kind of go to markets. And for I think it’s important caveat, not every category can be won that way, and we have some companies and categories that can’t be won that way, and that’s fine, but if you’re we do believe that your category can be won that way. It ultimately will be so it should be you, right, and not someone who comes along after right? How
35:30
do you advise companies or founders on making their first sales hires? Yeah,
35:35
well, going back to at the end of the day, the whole point of this is to close enterprise deals with humans. And I think one of the most fascinating things there, particularly as you’re going from zero to one, like your first few sales hires, the ideal person for your first few sales hires is dramatically different than the ideal person for sales hire. Like, I don’t know, 20 through 30. Like, once you’ve got the machine running and oiled, once you’re once you’re, like, scaling. And there’s a great, there’s great article. So one of my seed investors at my last company, angel investor, is this guy, Mark Leslie, who is the co founder, CEO of Veritas software back in the original.com boom, and then spent like, 20 years at this Graduate School of Business. So he co wrote this article, I think, almost 20 years ago, in 2006 called the sales learning curve. And there’s a bunch of great stuff in there, like people should read it if you’re remotely concerned about starting a sales team. But one of the key kind of insights observations for me, this just stuck with me. And I work with all my companies on when you’re going from zero to one, first and foremost, and it fits into the sales learning curve, right? Like nobody actually knows how to sell the product. Like you’re by definition, doing something that’s, if it’s interesting, it’s it’s a product that’s never been sold before, or a kind of products never been sold before, if it’s really interesting. And so nobody knows how to sell it. And so the very, very first thing the founders. Because occasionally, especially if your founders are technologists, they historically do not like salespeople or the idea of selling. It’s like, Look, you are going to you the founders. Only the founders can close the first few deals, and that’s because only the founders actually can speak to the kind of objections that come up. The founders are the only ones that have all the context in their heads, in those discussions, and they have, I refer to as like magic founder, Pixie Dust, like they just have an ability to navigate those conversations. They’re good founders, great founders, can navigate those conversations. They can on the fly. When the customer raises a completely legitimate objection, they can figure out a way to paper over it, and you should, and they can give the customer confidence to like, Okay, I have no business signing up for the startup this early. But like, I believe in you ultimately that the founder right? So the founders, they have to prove that the product can be sold. Like, if they can’t sell it, if the founder can’t sell it, nobody can sell it, right? So then the next inflection, and this where the sales learning curve really comes in. It’s like, okay, well, the next Well, the next thing we have to do, once the founders found out five deals, a million dollars, whatever you want to call it, there’s it varies by company. How do we hand this off? Because now as a founder, I’m actually not learning as much in every new conversation, like they’re starting to be repetitive and it’s and there is a lot of work to close a deal, right? And so probably we need to start scaling this. So I got to bring someone else in. And the sales learning curve talks about two kind of sales people. So there’s Renaissance sales people and coin operated sales people, right? And so if you think of a typical company at scale that has a sales org, they have enablement materials. They have clear messaging. They have decks. The coin operated sales person is the person who is wearing the expensive Rolex. Has the most expensive car in the parking lot. You hand those materials in a lead list to them, and they print cash, right? And from a cash con perspective, they’re the highest paid employee in the company, and it’s great. Everyone’s happy like they just have these super powers to just go in get deals closed, you know, they’ll kill a tiger to get a deal done. Yeah, yeah. But in the early days, when you’re handing off, none of that stuff exists, right? Like, sales enablement, well, we have a wiki page, right? Frequently Asked Questions, well, when are we doing our kind of weekly review? What’s a weekly review, right? Like, and so you’re out there by yourself. And so a renaissance salesperson is a salesperson who can still close, right, who’s good at selling, can close, but has more of like, almost like, a product management kind of mentality, mindset. And it’s just someone who’s just more fundamentally curious about what the customer’s doing if something’s not matching up, like, why it’s not matching up? Like a coin operated sales person, their job is to get to know as quickly as possible so they can get to the next Yes, and like, somebody in sales ops is going to sort out why we’re getting the no’s Right, yeah. The Renaissance salesperson walks into your office and it’s like, hey, founder, you. I had a couple of these conversations go and I think it’s this, I think it’s that and like, and those people, generally speaking, don’t want to be in that coin operated like, kind of kill or be killed, you know? And so I think, as
40:13
a technical they enjoy the idea maze of it. They enjoy the idea. They enjoy the intellectual
40:17
challenge. They like closing too, and they’re good at it, but they enjoy this other piece, and particularly technical founders, I think we, we tend to like paint sales people in just this like one lens. And so it’s really important the article does a much better job of laying this out more. But how to think about where you are in that sales learning curve, and where the transition point is, and yeah. So particularly, if you say yeah, if you not been in a sales org, which most of our founders haven’t. I think it’s just like, super good to look at and think about,
40:44
make sure to link to it in the show notes. Yeah, we can. Reminds me a little bit of many years ago, I had a gentleman taking on the on the show, and he talked about three different sales profiles, and they’re for different stages of a company. But in the early days, you have like the Davy Crockett. And as you starting to mature, you have the Joan of Arc. And then as you get really mature, then you’ve got, like, the Dwight D and it’s like, different competencies and different ways of thinking about sales and executing. But I don’t know I like the the metaphors, yeah,
41:15
yeah, that one sounds probably like the sales leader. Kind of like, what’s the right sales leader, person that’s
41:21
right, yeah, yeah, good. You know, last question on this, you mentioned that accessing developers is a challenge, right? Like they’re not hanging out on LinkedIn. They’re in a type that you just call up, but they are often pretty active on GitHub. Are you actively monitoring GitHub activity? You know, do you use it to identify major trends across developers. Yeah, I think, I think
41:43
how we use it, you know, it’s kind of interesting, as you might imagine. I mean, I don’t know, I created my GitHub account in like, 2009 think, I think I have a five digit GitHub idea. I forget exactly, but so I’ve been, I’ve been big fans since the early days. You had that had the fortune serendipity. I used to work with a couple of their founders in a coffee shop in San Francisco. I lived in SF back then, and, you know, it’s just like three or four people with this weird code hosting website. It’s become, obviously this, this just fundamental fabric of software development. So it certainly is. It is interesting. What happens on there is interesting. I think for us, it’s more of a validating signal. There are a lot of people, I think, if you are coming at, if you’re trying to come at this from a more generalist perspective, right? Like, I’m a generalist investor, so, but I know developer tools are valuable, so we should have a thesis there, and we should invest there as a generalist. You’re like, how do I get more signal? How do I, like, get more informed? And as a place that you can just kind of go scrape some data in an automated fashion, like, it’s very attractive. You know, for us, I think it’s one of those things. I forget the name of the theory, but I mean basically in Finance, any arbitrage position, just by definition, only exists for a certain amount of time, because the market figures it out. Everybody starts doing it, and the alpha goes to, like, zero, right? And so, you know, it almost any kind of venture firm with any kind of sizable budget and team is tracking that. And then it gets worse. Because, if the, if you want a hot take, like, you know, everybody’s aware of that GitHub star count, right? There was a period of time, and there’s all these stars on GitHub. You know, there’s a spectrum of founders, and some are less scrupulous than others. And like, you absolutely like, there is a, I’ve caught a black market, but there is a gray black market for purchasing GitHub stars, right where you can go pay a flat fee, I assume, in some kind of cryptocurrency, and a bunch of stars are gonna show up on your project. And, I mean, to the extent there’s, there’s kind of a white hat product that I’m aware of, that a boutique product that’s being sold, or at least to venture firms, where they try to do Fake Star detection using a bunch of different like security type techniques, behavior analysis. No
43:57
kidding, yeah, how?
43:58
And so, yeah. And so for me, I think first of all, by the time the star count is, like, skyrocketing or something, it’s sort of like, it’s probably too late. And so I more, we more use that kind of activity. I like more certainly, activity like, Okay, how many people are actually contributing? How viable is a community? But as once I’m interested in something, let’s go take a look at this,
44:20
Joe, if we could feature anyone here on the show, who do you think we should interview and what topic would you like to hear them speak about? That
44:26
is a great question from a VC perspective. I mean, I work a lot with Reed Christensen over CRV. Yeah. I think he’s great. Yeah, he’s great. I don’t know if you’ve had him on the show or not, but
44:36
I have. It’s been many years, though. Yeah. Well,
44:38
I mean, you know, CRV is playing in kind of a larger stage, and so I think it’s going to the asset class and where it’s going. I think it’s interesting from a founder perspective. I think, I mean, we already talked about Avery, I think he’s not just a kind of bit about Gong, but I think tailscale is doing a really interesting job of mapping just a very broad, bottom up, huge community of even. Obvious users to kind of large enterprise contracts. And I think it’s succeeding at that model to an extent. And we have a lot of companies succeed at that model, but I think doing really well. A lot of insights. There. Perfect
45:10
Joe, what book, article or video would you recommend to listeners? I know you gave us the carve out before on the was it sales learning curve?
45:17
Sales learning curve by Mark Leslie is great article, you know, book. I’m sure you’ve heard this before, but I think for my founders is important. High output management is just like an incredible book. I think for the founders I work with, who tend to be technologists, the fact that Andy Grove, like, part of what I think makes that books work is Andy brought like, a level of like an engineering mindset and rigor to a human problem, but didn’t forget the humanity, and he balanced those two. Usually it’s one or the other. Either get people trying to manage on vibes, or you get the kind of, like, impersonal, like, well, I can just reduce these people to spreadsheet boxes. And I think, yeah, there’s just, there’s still, like, just a ton of really valuable stuff. And it’s, I find myself coming back to quoting the founders. I’m like, Oh, you mean what you what you need? You need the, you know, the complimenting metric. And they’re like, what’s that? I’m like, Okay, what’s three pages in high output management, you have to go read but we have to understand the North Star metric. But we need the metric that tells us we’re not cheating, and it’s just, like, full of stuff like that. Love it. Love
46:16
it. Joe, do you have any habits, tactics or behaviors that are a force multiplier? Yes, besides Deepak,
46:21
yeah, yeah, yeah. One is always be looking to up level. Like, don’t let yourself fall in a rut, right? Like, just always be open to trying new things. They don’t work out. Stop doing them, right? But you won’t know until you try. I think, just as a general, you know, founding a company, leading a company, being an investor, we manage a lot of human to human politics and conflict. I think it’s just super important to run towards the conflict, like, if you’ve identified like inside your organization as a founder, or if as an investor, you kind of sense that like a founder’s maybe kind of withdrawing. Or like, if you sense conflict avoidance, like, force the issue, right and politely and with respect, but like, just, it does nobody good to let things, like, kind of fester in length. Things just get worse over time. So I think, like, if you find if you’re even remotely conflict avoidant, like, you just got to train that muscle to just go at it. I think the other thing too, it’s really important is, and it’s easier for some people than others, but just like, stay calm. I mean, in early in my career, I pretty honestly hot headed, but like, I certainly like, I feel things deeply, and I was just fortunate to work with some co founders, and like I said, like Mark, and I think I just internalizes. It just doesn’t do anyone no matter how frustrated, or it just, it doesn’t do anyone good, any good flat to handle, right? Like, all we can do is, like, assess the ground truth, be real, realistic about where we are and what possible paths forward are, and just put the company in the best possible strategy that we can based on that. And, you know, I’ve had some founders in tight spots be like, Oh, I thought you’d be upset. I’m just like, Well, I’m not happy. You know, we’re all adults. Like, yelling isn’t gonna do anything, so let’s, let’s just like, let’s just fix this. So perfect.
48:09
And then finally, here, Joe, what’s the best way for listeners to connect with you and follow along with heavybit?
48:14
Yeah. I mean, if you go heavybit.com, I think one of the things our background is an accelerator. We’ve always had a huge educational component to our mission that goes on to this day. And we’ve, we’ve built like, this massive kind of content library with, think, over 1500 different like talks and seminars, etc. And you can find like transcripts and high production videos of all that. So if you’re doing, and we’ve got even just like syllabi, so if you want to, if you want to do more about pricing, we’ve got your top 10 pieces you got to hit if you want to do more about hiring your first salesperson, if you want to talk more about product market fit, anything you’re worried about founding one of these companies, we’ve probably got some like, tailor made content for you. And then, yeah, I mean, I’m, you know, I’m on LinkedIn, GitHub, X blue sky. It’s usually my name, Joseph Russo, all one word. You can find me there. Always love to talk to people.
49:05
And yeah, he’s a VC. He has to be everywhere at all times. Everywhere at
49:10
all times. I know people are making their choices, and I’m like, Well, it’s my job, so I’m there. Well, he
49:15
is Joe rucio. The firm is heavy bit Joe. Thanks so much for the insights on AI, on developers, and, of course, on go to market, super helpful.
49:21
Yeah, I love the podcast, and super excited to be a guest today. Thank you, sir.
49:33
All right, that’ll wrap up today’s interview. If you enjoyed the episode or a previous one, let the guest know about it, share your thoughts on social or shoot them an email. Let them know what particularly resonated with you. I can’t tell you how much I appreciate that some of the smartest folks in venture are willing to take the time and share their insights with us. If you feel the same, a compliment goes a long way. Okay, that’s a wrap for today. Until next time, remember to over, prepare, choose carefully and invest confidently. Thanks so much for listening.
