Over the weekend at HackMIT 2025, Dedalus Labs joined a standout group of sponsors and sat down with Ankit Gupta, Y Combinator's newest General Partner.
In a conversation with Dedalus Labs CEO, Cathy Di, Ankit Gupta shares how he works with founders, what YC looks for in a founding team, and the startup trends he has observed.
Transcript
Cathy (00:00)
Hey everyone, this is Cathy here from Dedalus Labs, and we are currently at day two of HackMIT. I'm here with someone very exciting. Do you want to introduce yourself?
Ankit (00:08)
Hey, I'm Ankit. I'm a general partner at YC.
Cathy (00:10)
Okay, first question. Right off the bat, we won't name any names, but tell me your favorite project at HackMIT so far?
Ankit (00:17)
Oh man, there's so many good projects at HackMIT. I don't want to spoil the company that is the prize that YC's giving out later today, so I won't mention them specifically, but there's a bunch of companies that are working on interesting, really high-performance indexing of systems that can be used for AI. And so I saw a bunch of cool companies that are making Dropbox competitors, basically. And so, yeah, I'm kind of into that. I like the intense technical projects.
Cathy (00:45)
Interesting. Oh, on this note, I'm very curious: what do you think is the edge that some of these projects have over Dropbox, let's say if they do go on to build their own companies?
Ankit (00:54)
I think it's less the edge that they have over specifically Dropbox. It's more just broadly, I think the entire developer stack is being rewritten right now.
And the types of things you would optimize for 10 or 15 years ago when your primary user is a human, versus today where your primary user of your tool might actually not be a human—might be some AI system—means that you're going to optimize for different things.
And so it's less that another company couldn't do it and more that they probably hit product-market fit well before they needed to. And so it's quite challenging for them to necessarily work on something new.
Cathy (01:30)
Yeah, that makes a lot of sense. On this note, I've also been talking to a lot of these Dropbox competitors that are at HackMIT right now. And one thing I noticed was just that, like you said, because the audience is sometimes not even humans anymore—it's for maybe scrapers or AI chatbots—the workflow or the intended user experience is very different. I wonder if this is something you also see in the YC startups in the recent batches?
Ankit (01:56)
I mean, I think generally what we see these days is that the best way to find great ideas is often to just start playing with the newest AI tools that are available. So, for example, playing with voice AI systems or browser automation or other things on the kind of new dev stack and just noticing things that seem off or missing or unexpected.
And as a result, you often find that it's just people playing with the newest tools that find these new things. And so as a result, they're often building things for themselves. We see a lot of developer tools that are in YC these days where it's basically dev tools they're building because some app they were previously building was missing something.
And I think that ends up leading to lots of interesting new user experiences because they're building for a stack that didn't really exist before.
Cathy (02:42)
That is super interesting. I feel like something that I, as a YC founder, personally thought a lot about was just: how do you identify the problem and who should you be building the solution for?
You mentioned a lot of people are now building for themselves. I feel like that is not always the case. As a founder, oftentimes if you build for yourself, you may or may not be solving a non-problem that nobody is willing to pay for.
What's your take on this now?
Ankit (03:05)
I think it's interesting. I think there was a really long period of time in which the right way to solve a problem was to go work in another place for a long time and get a really niche understanding of that thing and then maybe generalize the thing you were doing to somewhere else.
You make an internal tool at Amazon and then you leave and now you sell that internal tool to everyone—that kind of thing. I think that's still a fine way to come up with good startup ideas. In many ways, becoming an expert in something first and then noticing unexpected things from that is a very reasonable way to find ideas.
It is also true that when you're in a transformative technological moment like we're in now—or at least I think we're in now—you can kind of just see gold everywhere. You just have to be out there panning for it.
And the way you go out there panning for it is basically: you start building. And if you do that, you're very likely to strike gold in every direction you turn.
Cathy (03:55)
Totally agree. I honestly think that's why hackathons are so great now.
It gives people just a very concentrated 24 to 48 hours to build something that they wouldn't otherwise build. On this note, I feel like a lot of people are curious about this. I don't know if you can speak more to it, but what are the things YC looks for in HackMIT projects?
Ankit (04:13)
I think the things we look for are the things we always look for. We look for really strong technical folks; those are often people that would come to hackathons, so that's usually a good sign. I think often what we look for more is what happens after the hackathon.
So often what happens with hackathons is you make a prototype of something. That is not in itself necessarily a startup yet, but two things often happen next. Either in the making of that prototype, you discover some things that are strange, or unexpected, or unusual, and those can become interesting new dev tool ideas or new ideas in themselves.
Or you take the thing you built and you go get a user. And often the feedback you get from that user can give you the next direction to take. And building a company is much more like a greedy search on user feedback than coming up with a preconceived idea and going.
So in many ways I think of what we look for as more like the process that someone's running. Often doing a hackathon and building something is a great way to start that process, and where people go from there is often what turns these things into lots of really great startup ideas.
Cathy (05:12)
Yeah, super interesting. Something I've noticed—because for context, we just hosted the first-ever overnight hackathon at the YC office—is that a lot of the teams, like so many teams, actually went on to keep building their projects. It often starts with a waitlist on their website and then 30 people sign up. They're like "Wait, people actually want this," and they keep going. A lot of them are now actually applying to YC and may or may not have gotten an interview. I've been keeping tabs, and I guess my question for you is—and this is probably something a lot of the hackathon attendees here are wondering—what is a good sign that I should keep building the project? How do I know if I should continue or if I should move on to something else?
Ankit (05:50)
I think the natural instinct of most computer scientists is to think about that question as something like: Is this really cool? Is the technology hard? Could someone else do this technology? Am I the first person to do this?
Is there competition in it? And those are actually, counterintuitively, the wrong questions to ask for the most part. They're okay questions to ask, maybe in the back of your mind. The thing in the front of your mind should be: Does anyone want this?
And ideally, does anyone want to pay for this? And often you can find a paying user for a new AI tool especially very rapidly. I think that's one of the most shocking aspects of the current moment we're in: people are adopting, with actual dollars, the AI tools that people are building faster than literally ever before.
More software is being sold faster than ever before. And so that is the primary thing that can drive you to just keep going—or it should be the primary thing—is you basically see: okay, do people want this?
And if they do, I'm probably onto something. And if you find that literally no one is willing to pay anything—you know, even $1—for the thing you're coming up with, it's probably a sign that you should think about that. That feedback is really valuable.
Cathy (06:54)
Yeah, makes a lot of sense. On this note, I've been looking at a lot of the hackathon projects here and something that I do notice in these university-led hackathons, as opposed to the SF hackathons that companies host, is just that there are a lot more visionary projects, a lot more projects on accessibility or social impact.
And it's "How do I change the world?" as opposed to "How do I make money with this product?" But then again, we face the problem of: will users actually be willing to pay for the solution you found that actually solves a real problem?
So my question for you is, do you feel like the types of things people are willing to pay for have been changing, now that there are so many more types of solutions or ways to solve a problem?
Ankit (07:33)
I mean, yeah, I think what's interesting right now is that more industries are buying software that didn't historically buy that much software than ever before.
Like some of the fastest-growing companies I can think of off the top of my head are selling to industries like dentists, where suddenly dentists are buying way more software than they ever have before.
And you know, if you told me that five years ago, I'd be like, "Yeah, there's no way dentists are in the top quartile of YC batch startups selling software for dentists," but you see that now.
And so I think generally making an incredible visionary product with great social impact is actually not mutually exclusive with making something that people want or want to pay for, especially these days when there's a massive number of industries where you can make a huge amount of impact.
And so yeah, I think chasing after the visionary things is great. And I think the surface of things you can do that for is really big these days.
Cathy (08:23)
Yeah, I 100% agree. I feel like that makes me happy.
Cathy (08:28)
Okay, let's wrap up this interview. But one last question for you is just: what should people be building? I feel like I talked to a lot of teams here and they were just honestly confused. Like, how do I find a problem?
How do I find something that I'm willing to work on for maybe the next 10 years?
Ankit (08:48)
I think it's actually very hard to a priori know that you're going to be willing to work on a problem for the next 10 years.
It is often much easier to know that you want to work with a particular team for the next 10 years. So I would actually say, counterintuitively, don't worry about the problem.
Maybe have an idea space you're excited about—like maybe you're generally interested in dev tools or you're generally interested in healthcare or whatever. And find a person or a team of people, ideally one other person or two other people who you're really excited to work with.
And we find that teams of two and three that have a lot of experience working together—these are usually best friends or siblings or people who have taken a lot of classes together or people that have worked together on side projects a bunch—is a great place to start.
And then what I would do in terms of converting that team into a high-probability early idea is to basically just follow your interests along the Pareto frontier of what's possible. Basically, if you just take the newest tools—like the types of dev tools coming out of YC companies from the last two years—and you basically try to just build whatever is interesting for you using those tools, and you're kind of at this Pareto frontier of knowledge with the team you like working with, and you just keep your eyes open for anything that feels unusual or unexpected and find users for what you're building, you're very likely to come up with great startup ideas just using that simple formula.
And it didn't actually really used to be that way like 10 or 20 years ago, except for the great team part. But we're in like an amazing time to find startup ideas like that.
Cathy (10:14)
Yeah. Okay, last question from my personal interest, but I see a lot of agent-related startups in both HackMIT and also at YC. Like I feel like 80 to 90%—stats off the top of my head, these are not official stats—like 80 to 90% of my batch is something agent-related. I wonder if you see any trends in how people are building this. And, for example, I think there are two types of agent startups out there.
Ones that replace humans altogether: your AI coworker, your AI executive assistant, etc. And there are other types that maybe augment or enhance a part of your workflow. I wonder what's your thought here?
What patterns is YC seeing, and where do you think the future is trending towards?
Ankit (10:55)
I guess people often assume that YC has preconceived assumptions for industry trends or stuff like that.
Honestly, we don't think about that stuff that much. We think more just: what are the things that the best founders are looking for, like the best teams of really technical people. And I think roughly the way I think about the state of these AI things these days is: if you could roughly partition knowledge work into acquiring information from the web and summarizing it and organizing it and doing deep research, and then taking actions into the world like making a phone call or clicking something or filling out a form or writing a document—I think these systems have gotten extremely good at the former over the last few years and they are very nascently getting good at the latter. And I think a lot of the best founders are drawn to spaces in which you can experiment with both sides of this.
And I think that's why you see a lot of agent things. It's not because people are inherently drawn to AI agents or whatever; it's because you just are at this frontier of knowledge and what's possible.
And so you can take something that just barely works and make it really work if you're a clever technologist. And I think there's something quite fun just in that possibility space. And so yeah, I think we're going to see a lot of people trying to push the edge of that.
Cathy (12:07)
Yeah, no, 100% agree. Last plug for Dedalus before we end the interview. We are exactly positioned between finding interesting things and making sure that AI agents are able to very reliably take action in a non-deterministic, non-linear way, which is very nascent.
And I think people are just about getting used to this concept. But yeah, you should check out our open-source SDK and our platform which deploys MCP servers in three clicks.
Ankit (12:35)
Cool.
Cathy (12:37)
Thank you so much, Ankit. It was an honor interviewing you.
Ankit (12:39)
Absolutely. Thanks for having me.
Cathy (12:40)
Catch you guys on the next hackathon!