What Investors Look For in AI Startups: Builders with Taste

Hi guys, my name is Marco and I'm talking with Jesse Robbins from Heavybit here at Shift Miami 2025. Jesse, welcome. Oh, thanks. It's been awesome. Thank you. So as an investor, what are you looking for in dev tool startups in 2025 and beyond? You know, generally, we look for a couple things. Heavy Bit, all we do is specialize in developer tools, infrastructure, AI investing. We're all founder operator builders. And so what we start by looking for are people that are builders like we are. They have great taste. They have a problem that they're desperate to solve, that they kind of need to fix in the world. They do it no matter what. And that they're able to get early communities of developers excited about. For me personally, I like to say that I look for people building things that change the way that you think about the world or about a problem from the very first time you see it. A lot of code assistants are currently doing that. Right now, my own investment in Continue, literally the first time I saw it, I'm like, "Oh, wow. I'm never going back." I actually switched from vi to using VS Code and Continue, which is sort of a crazy shift for an old school systems engineer. But yeah, I'm looking for great taste from people that really are building something for themselves and their community, and that it changes you the moment you see it. The other things we look for in the broader sense — always want to see great open-source software that then has a sophisticated business model built around it, and tools that are pretty hard to build without a lot of specialized knowledge. So how do you basically evaluate — what are some of the key points that you're looking for? So when people come and talk to us, the first thing is usually everyone that is in the Heavybit ecosystem has participated in one of our events or knows somebody that has. They're a member of our existing very large community. They've heard our podcasts. And we find that the people we vibe best with are people who are already active community members. When I think about what I want and how we evaluate things, the first thing that matters — and hopefully I haven't said it too many times — is I really want to be surprised by the first time I see the thing. A recent company of ours, Tailscale — been involved with them for a very long time, actually from the very beginning. They just raised this enormous round. And I've built a lot of VPNs over my career. The very first time I used Tailscale, it made me mad. It was so easy. I'm like, I've done all of this work for so long, and I see this and I realize, nope, this is the future. This is going to be the new way we build. Continue, a more recent company, completely changed the way that I write software and now does that for hundreds of thousands of people. So the interesting thing there is we look for just real usage. We're not the type of investor that is looking for hype or the thing of the day. We usually invest creative concepts. And so we're always looking at that earliest stage. People don't need to have a smooth pitch — we'll talk to people for a while. Mostly we're just looking for authentic engagement, because when we evaluate a startup, we're really evaluating a founder or founding team that we're going to spend the next decade working together in order to build something extraordinary. Hype aside, what parts of developer workflow are yet to be disrupted by AI automation? Well, I think we're currently in this very narrow beginning stage. If you look at the entire software life cycle and you say, "Well, where are we right now?" Well, we're a little bit of the way into code generation, a little bit into some types of test automation — a lot of what I think of as low-hanging fruit. The really repetitive stuff. It's like, well, what is the work that none of us ever really want to do in the first place? That's what we're automating right now. We're automating toil. There's so much more value to be worked on after that point. We're going to see complete changes in observability, complete changes in testing, CI/CD workflows. We're going to see completely new approaches to UI/UX generation on the fly, customization, personalization — every single area in the developer tools ecosystem is getting changed. And we're at the very, very beginning of the cycle. We've really covered a very small area so far and we're just beginning to see all the new places. How do you see AI changing the startup life cycle? What will AI native founders do differently? There's no evidence to support that real companies, real startups are being built in shortened life cycles. All the work is still the same. What has changed is what we think of as table stakes work is shifting. I look at what's happening right now for startups to be a lot like what happened when we launched cloud computing with EC2 and S3 — which I was at Amazon during that time and got to contribute to some of that — and then subsequently with Chef. We shifted from you've got to do all this bootstrapping work to get rolling, to nope, now you already have your infrastructure in place and you have the first parts of automation in place. So for AI native startups, what is happening is you're getting some interesting early advances. You've got a founding team, early engineers that are potentially a lot more productive. They're able to work with larger volumes of code. They're able to build more, faster. Now, that comes with complexity. That comes with new types of tech debt that you're accumulating. But we're seeing people get to certain product stages a little faster. It doesn't change the overall trajectory though, because that's happening for everyone all at the same time. We don't believe that you need less money. We don't believe that you need less time. What we do believe is that you need less time spent on stuff that doesn't actually relate to building your startup — just like with cloud computing, we didn't have to go to the data center anymore. That wasn't an important skill, but it used to be super strategic. How do you see software engineering evolve in the next couple of years? The definition of software engineer is going to change, and I'm pretty excited about that. If you look at the history of software and computers in general, we went from hand-wiring circuits, then transistors, then transistorized computers — but that's all still being built by hand. Then we got assembly language as the first abstraction that allowed us to go from hand-assembled physical switches to a somewhat English-like language that translated into computer language. From there we started adding languages, and the languages over time have become more sophisticated, closer in grammar and syntax to human spoken language. And that's continuing. So now we've arrived at the point where we're getting to being able to express software concepts and have another layer of abstraction that generates code, just like every cycle prior to that point. All of the change that has brought about — each time, is it 10x or 100x? I don't know. But the work you had to do to do software in the '70s, '80s, '90s, 2000s, right to today, has basically been about humans expressing themselves more naturally in order to build more and more complex things. So we're now seeing entire industries get built where you're going to write your product requirements doc in an enterprise, or your plan, or your vibe coding, or whatever — that's going to let you build out first versions of things. And that's been a dream since the idea of computers actually started. I'm pretty excited because every time we add a few million more people to building things, well, they get more complicated. There's more of it to manage. And so there's more things for me to build and invest in and more money to make. And I really like that, because more people get involved and it changes more lives. Would you agree that agentic experience is the new developer experience? Well, we're here at your conference so I think you want me to say yes. The truth is that agents are just another type of developer. Things I would agree with — I think we should have really good documentation. Humans, agents, same — need good documentation. I think we should have really good APIs and we should have lots of affordances in those APIs so that it's really easy to use whether you're a human or an agent. I think that we should make things easy to program against. So we should have standards in those APIs and in those user experiences that make it really easy for humans to use, including humans that might need a screen reader. Blind users and Google for SEO on the web are the same. Google needed these affordances — this thing that we now call search engine optimization. And so now here we are with agents, which are a new type of user, and they need great documentation, clearly defined APIs. They need to be able to change and understand. All of that infrastructure that gets built out, all the hooks and loops and little touch points, are going to help everybody — whether it's developers, whether it's agents, whether it's users of all shapes. What I'm excited about is we're also going to get into a mode pretty soon where your user experience personally can be personalized. Here's all the stuff I don't want to see, all the clutter. I've never clicked that button — why isn't it regenerating automatically for me based on my use? And in order for that kind of thing to happen, we have to have this agentic experience. So it's definitely the new wave that is driving that. But does it mean that any of the work we were doing before goes away? No. Hopefully it just causes people to do more.