Seed to Series C: What VCs actually want from AI startups

AI investments hit $110 billion in 2024, and the funding landscape in 2025 is more competitive than ever. For early-stage startups, that means more money in the market but also more pressure to stand out. At TechCrunch Sessions: AI, Rebecca Bellan sat down with three experienced investors: Jill Chase, Partner at CapitalG; Kanu Gulati, Partner at Khosla Ventures; and Sara Ittelson, Partner at Accel. They broke down what they are really looking for when evaluating AI startups from seed through Series C. Their message to founders? Forget the perfect pitch. Focus on building trust, surviving the hype cycle, and being ready for copycats the moment you find product-market fit. Listen to the full episode of Equity to hear about: Why VCs say founders are over-indexing on pitch decks instead of relationships What it takes to go up against big incumbents without getting crushed Why consumer focus (and speed) still win, even in B2B AI How agents and automation are already reshaping the startup playbook Equity will be back Friday with our weekly news roundup, so stay tuned. Equity is TechCrunch’s flagship podcast, produced by Theresa Loconsolo, and posts every Wednesday and Friday. Subscribe to us on Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod. For the full episode transcript, for those who prefer reading over listening, check out our full archive of episodes here. Credits: Equity is produced by Theresa Loconsolo with editing by Kell. We’d also like to thank TechCrunch’s audience development team. Thank you so much for listening, and we'll talk to you next time. Learn more about your ad choices. Visit megaphone.fm/adchoices

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Anthony Ha?

  This episode of Equity is brought to you by Schneider Electric. Schneider Electric, your energy technology partner.

• 0:17Speaker 01

  Hello and welcome back to Equity TechCrunch's flagship podcast about the business of startups. I'm Rebecca Bellan and this is the episode where we bring on industry experts to help us explore a trend in the tech world and dive deep. This week, I'm bringing you three experts from an onstage interview at TechCrunch's first AI sessions in Berkeley. I chatted with three VCs about what they want to see in AI startups, from seed to series C. My panelists were Jill Chase from Capital G, Kanu Galati from Cozla Ventures, and Sarah Idelson at Excel. And they brought a lively discussion with plenty of useful insights for founders playing in the AI space. We discussed the challenge of competing with incumbents, how agents will impact the workforce, the importance of consumer focus, and the need for startups to adapt quickly. Take a listen.

• 1:03Speaker 01

  You've mentioned before that you kind of back phenomenal founders, right? But for a lot of early-stage founders, maybe they're young, maybe they don't have a pedigree, maybe they didn't come from big tech, like, how do you know if they're phenomenal or not?

• 1:12Speaker 02

  I mean, probably my advice would be to talk to venture capitalists, or if you're thinking about founding, start building relationships earlier. And I think that that kind of goes counter to maybe what some folks think, like, because of that insecurity, they want to be super buttoned up, they want to be like totally dialed in on their idea, they want to have their pitch down pat. And if you do that, you're probably too late. You're kind of showing up and saying like, "I'm ready for my check." What's actually more valuable is to build a relationship and to get to know folks and have them see actually how you think and operate and how you build, how you dissect a problem space. Actually, spending time with pre-founders when they're in that messy stage is the best data about their grittiness and how they dissect a problem.

• 1:56Speaker 02

  And so you're actually denying the funding ecosystem of seeing what a great builder you are if you only show up when you're perfectly polished. We haven't seen the journey to get there. And so when we often do sort of a day zero Check. It's because we've actually spent time with them for the six months prior as they've weaved around different ideas. And it's that process, more than the ultimate idea, that's given us the confidence in them. Because this space is moving so quickly. Our founders who are most successful are the ones who are super fast, super decisive, see that the tide is shifting and are willing to burn the boats and go a new direction. It's not about that like moment in time, what is the idea, is everything precise and perfectly aligned?

• 2:39Speaker 02

  It's how does this person think and act and operate, digest new data, how do they engage with customers? And all of that is really on display during that early exploration.

• 2:49Rebecca Bellan?

  And Connor, you focus a lot on like more technical founders. I mean, you mentioned that you shop around in academia for people who are really building that tech.

• 2:57Speaker 04

  Yes, absolutely. So as a firm, Kostaventures, we are fairly generic. You know, we have to convince ourselves there's a big opportunity. And we are in resonance with the founders on how we should think about the risk retirement order. And we would look at that space and the opportunity. I, yes, have a technical background and I do land up spending a lot more time with specific professors in different universities, more for my own learning. You know, the signal to noise is quite low in AI. I want to myself understand what can be demonstrable in the next three years versus what is still hype. That's one reason I spend time. But then in return, I can work with graduate students, postdocs, undergrads, who are looking at their next big idea and what can be accomplished and what could be interesting to venture capital and give them early feedback. I think it's a win-win for all of us.

• 3:41Speaker 04

  So I do land spending a lot of my time and have invested in companies that have come directly out of students from some of these professors. Like Poly AI is one of my investments in a voice first customer service. This is with Professor Steve Young at University of Cambridge, et cetera.

• 3:54Rebecca Bellan?

  So. And now I wanted to ask you also about agents, right? Agents are huge right now. You mentioned to me before that you think that soon agents will be able to take on manager level roles. Right now they're more of like the intern level. If that future plays out, what does that mean for startups and investing today?

• 4:10Speaker 04

  So our framework internally, my framework built on top of it. You know, Microsoft became big by leveraging software and reduced the cost of production to zero. Google became big by leveraging internet and made the cost of distribution to really zero. And the next set of successes, the really large companies are going to bring the cost of labor to real zero. And we think it's going to start from the intern or the junior analyst for most roles. We're already seeing it at coding agents and customer service. And they are already getting smart enough to do more complicated tasks because of all the technology vectors happening in reasoning, in model inspection and so on. These agents can now start doing, I think in a few years, less than five, they will

• 4:52Speaker 04

  be mainstream to do a senior IC task, think more complicated things like chip design and thermal engineers. We think this is going to happen and it is going to be exciting for startups to reimagine how they can be a chip design company without needing the 10,000 people that is needed by an AMD today.

• 5:09Rebecca Bellan?

  I guess another way to ask my question, and any of you can answer it, is, you know, the CEO of Anthropic recently said that he thinks that 50% of white collar work will be gone in the next five to 10 years. Does this just create a smaller pool of companies for you to be investing in?

• 5:23Speaker 04

  I might jump in there. I think we are very innovative as humankind in general. Even if specific tasks are getting automated by agents and specific company outputs are getting automated by agents, I think we will have a new class of, think, search funds, early-stage, recent graduates, buying businesses and figuring out how to automate the tasks that can be automated by agents and then getting a better top-line revenue or a better bottom-line revenue. I think we will become more and more innovative in how we think of companies and how we think about their growth, expansions, et cetera. So I wouldn't say like there will be no startups and there will be no venture funding. I think we're just very innovative as humankind.

• 6:00Speaker 02

  I think it puts a lot of pressure on sort of interrogating what's going to provide long-term durability because it does as this like flattening of production and reduction of labor, as things get super easy, you know, and we see it even now, and the prototypes that come in are way more advanced than I think you would have previously expected because I can develop a prototype this afternoon if I want. I think that with that flattening, it means that you have to be able to really articulate over the long term, why this should become a big company that matters, you know? And I think that that's like the job that we have to take on in our seat. I don't think it'll be fewer, but I think it does sort of force different pressure on different dimensions, because there were some companies where getting out of head and building really quickly, and of course, that's still really important.

• 6:46Speaker 02

  But some pieces of that have gotten easier. And so being able to articulate, this is like a genuine network effect, or this is a genuine like distribution advantage. It's not just one thing, but being able to say what makes sense for your company, I think, is increasingly important.

• 6:59Speaker 05

  I was gonna say, I think a logical extension of all the dynamics we're talking about is that there may be a very long tail of extremely personalized, not massive businesses or not massive companies, but one person can very quickly build something that is highly relevant for a very small set of people. And maybe that's not an enormous business, but it drives incredible value. Conversely, there will be, and I do believe like this is a very possible state of the world. You have that long tail, and then you have eight to 10 massive categories where there are sort of subcomponents of each category where you can build this durable value. So an example of that would be like customer experience. Like, that's really hard to get right, even if the cost of software development is zero, because there's so much on the product side and what the customers need

• 7:42Speaker 05

  and complexity with how enterprises operate. And so I think leaning into that sort of like, hey, where are their big categories of both labor and software spend today? And how can I really deeply empathize with what the problem is, what the use case is, and build into that, leveraging all these tools at your disposal, but really orienting around the sort of customer problem? I think that's how you can build a big business.

• 8:03Rebecca Bellan?

  I think that's the future that we all kind of want to see with this. But at the same time, we see dev-focused AIs heating up. We also see enterprise search heating up. For example, OpenAI yesterday announced new tools for enterprise, like Dropbox and Google integration. I think last week, Reuters reported that Microsoft GitHub Copilot had 500-plus million in revenue. So these larger incumbents are kind of eating what a lot of the smaller startups are trying to do. How do you place a bet knowing that the incumbents are scaling so fast?

• 8:32Speaker 05

  Yeah, I mean, I think that's always been a challenge in investing as long as we've all been doing it. I think it's even harder today because if you were to compare this to the cloud transition as an example, you know, it's quite hard for incumbents to completely rewrite everything and become cloud native. Like, that was a very long period of time, and so it sort of gave startups the time to really go out and build, develop customers, get distribution and compete. Here, it's a lot easier, especially as the models and the brains themselves get more powerful, for incumbents to just layer in these capabilities. And honestly, they can use some of the early-stage venture environment and startup ecosystem as sort of like free product-market-fit testing and say like, okay, this company is taking off. This is kind of the approach they're taking. I'm going to layer it in and sort of do the exact same thing. That's the skeptical view of it.

• 9:15Speaker 05

  I think the positive view, and we've seen this play out, is like cogeneration is a good example. Like, obviously, you know, what GitHub has been able to accomplish from a revenue perspective and a distribution perspective is unbelievable. It's awesome. And you have Cursor at 300 million plus in revenue scaling incredibly quickly. And I think this is where, even as a growth-stage investor, Sarah, a lot of what you were saying about like, what are the things you value in a founder that really matter? Historically, in growth-stage investing, it used to be like, hey, can they manage? Can they hire super well? Do they understand new economics? Like, that style of thinking. Now, it's that, but really even more of the like, hey, can they burn the bridges and adapt? Because things are changing so quickly and, you know, Cursor might be scaling so dramatically, but then GitHub does this, and what do you have to do next?

• 9:59Speaker 05

  And so I think sort of the extension of that skill set into the growth stages and IPO and beyond is even more critical than it had been historically.

• 10:07Speaker 04

  Yeah, I would like to add on to that. I think the challenge for larger, more established incumbents is exactly that. It's harder for them to rethink their entire technical stack, their data models, and how their schemas work such that they can be ready for an agentic world. It is so much easier for a startup, starting from scratch, to plan that stack, set the customer expectations right. They don't have to retract an existing contract and say what my SLAs were. They can set that expectation that for this class of problems, I can solve it to 80, 90% accuracy and go in with that right step to a customer. So I think the founders in the early-stage companies today have a different set of advantages that they can use to then overcome the distribution advantages that the incumbents have.

• 10:45Speaker 02

  I do think it's a mental exercise that everyone should go through if you're setting out to start a company is like, it's not like OpenAI is coming for you. They are just going to like incidentally steamroll you if you haven't like thought it through, you know?

• 10:57Rebecca Bellan?

  Or acquire you, right? It's like, is anyone gonna IPO or are they gonna take it to the bank or is the exit strategy always M and A now?

• 11:04Speaker 02

  Well, that's another good question. But I think on this idea of, you know, fighting with incumbents is you sort of have to play forward. If they could just roll it out at the next big announcement, then maybe you haven't come up with the right idea. Or like the audience that is distinctive enough, like an interesting ICP that has more than just a last mile, but a last marathon of interesting, unique needs, workflows, whatever it is, that the big guys will never get to because they have, their strength is also their disadvantage, which is they have to be pretty like vanilla to serve the entirety of the world, you know? And as a startup, you can say, they're never gonna

• 11:45Speaker 02

  do this extra distance that we know this like very valuable pocket of the market will really appreciate this extra distance. We can do that. They will never do that. They can never do that because it would be like another tab that would confuse another 5 billion users. If you think that through, you can find something where you can operate the laser focus on what you're building and not live in fear that the next announcement is gonna be your demise.

• 12:12Rebecca Bellan?

  Right. Okay. That's good thought to have. So, I mean, again, just not like digging into the incumbents winning or anything, but they've got data, they've got distribution, they've got compute. What other moats are important to you?

• 12:21Speaker 05

  I think so much of it these days is really customer love and being able to prove that you're meeting the customers where they're at. And I think exactly what they don't have or can't really have is the incredible focus that startups have. And it's just a reality. Like they have all of the advantages that you just described, but they can't be everything to everyone. And so if you find a big enough pocket of an ICP or customer base that really can drive a lot of value, either because they already spend a lot on software or there's a lot of labor that they sort of represent today, find that ICP and build for them. Constantly and iterate with them constantly because it's just unlikely that the big companies are going to spend the time doing that.

• 13:02Speaker 05

  And I think we're all living in a state of the world where like if something is not functional for me, I am so frustrated so quickly. Which is pretty funny because you know, 20 seconds ago, we didn't have chat GPT or any of these things. And now when it doesn't immediately give me the right answer, I'm like, God, this is so frustrating. And so I feel like a little bit of that finding sort of like getting to the right customer and really building for them can drive a lot of value and can drive a unique advantage.

• 13:25Speaker 04

  Now going down specific verticals also is another way to think about it. It's rare that OpenAI or one of the larger model providers will go down a specific industry, a specific vertical and only solve for them. They only want to go after multi, multi billion dollar opportunities at that scale. But for a startup, having that focus allows you to go to that last mile much faster, much easier. And my suggestion would be for startups to build in a way that you can leverage all the advancements that's coming out of the industry. There's so much happening in AI and the newer capabilities that's coming up. I don't think it is now smart to go and say, I'm going to start with building my own foundation models from scratch. You should build a company where you have access to a unique data set or a unique data model

• 14:06Speaker 04

  or a unique insight on what the customer here really needs and the service experience and the SLAs that you can actually beat. But build and leverage all of the advantages that are happening in AI.

• 14:22Anthony Ha?

  What moves our world forward? Energy. History has been shaped by it, revolutions won with it and progress dependent on it. For 190 years, Schneider Electric has been a leading force in this story, creating the technologies that harness energy and transform the competitive landscape. And today, in the dawning era of energy and industrial intelligence, Schneider is helping customers build more efficient, resilient and sustainable operations across every sector. From data centers to homes, from buildings to industry. Schneider Electric, your energy technology partner. Discover more at se.com.

• 15:00Rebecca Bellan?

  Yeah, let's talk a little bit about industry-specific things. Sarah, you mentioned that you're on the board of Assembly AI. How are you thinking about voice infrastructure space now that a few leaders have emerged, some have dropped off?

• 15:11Speaker 02

  Yeah, I mean, it actually speaks to Jill's point. Our expectations are so high, and I actually think that voice is a perfect example of this. And it's part of why it has exploded so much in the recent moments because it has crossed into the zone where it can really, really deliver. But there's so much nuance in how you build to make it super, super performant. I would add speed as another moat as we're talking about that. And I think that is super critical within voice and speech. Like, Assembly AI's latest model came out, just an example of how, like, I think great founders are thinking about it. It does sort of a predictive score throughout the entirety of the speech. It's predicting, is that the end of their statement?

• 15:55Speaker 02

  As opposed to saying, like, when have I paused long enough? And you can just imagine the impact that that has for a downstream developer on their ability to, like, tune the dials of whether or not it feels like I'm just the engagement model with that sort of agent on the other side. So something that's like, oh, did she pause? Okay, she paused. Like, I can jump in, you know? Is clearly going to be worse than something where at the end of every single word that I'm saying, it's providing a numerical score predicting whether I'm done talking, and you can choose when to jump in with your agent. That is like the level of like, in the grittiness of what does it mean to really deliver an agent through voice that can be performant in a degree that I'm not sitting there, like, pressing zero or asking for, you know, to be transferred?

• 16:40Speaker 02

  And actually, when you do that last mile, really, really important work, it's a better experience, and that's why it's exploding. It's like you now have something that isn't a crummy voice exchange and knows all the right answers and is, you know, 100% consistent and, and, and, available 24/7. I don't have to call during business hours to get an answer. There's just 1,000 things that are better if you can get over this hurdle of is the model and is the technology strong enough? And I think that's a good point.

• 17:09Speaker 02

  I think that's just recently happened and now we're seeing this incredible explosion of all the possibilities of where it can be applied. But I just use that as an example of it's not just, oh, speech to text models. When you're really in it, which is why I think even in the infrastructure stack, there's like so much opportunity because when you're really in it, there is like so much to build and so much to get precise to really deliver a great experience. And I think the founders that have that obsessive North Star are proving really successful.

• 17:37Rebecca Bellan?

  Yeah, I mean, I do want to ask about that a little bit more because as you say, there's a lot more than voice to text to speech. There's also, you know, voice cloning. And when we talk about media generation tools, there's deep fakes, there's all these tools. So what ethical lines do you want founders to draw and how early?

• 17:53Speaker 02

  I feel quite strongly about this and I've made investments in this category. I think that in sort of the social media transition, we were naive about like the public square and like the kumbaya community that would occur in the public square. I think we have to be very sober to the fact that with incredible technology, there are a bunch of people who are going to try and do wonderful things with it. And there will always be people and actors that try and do bad things with it. We know too much to sort of put our head in the sand. So from that perspective, in this moment, I think the answer is you should be building with safety and trust and assumption that people will try and do bad things with your product from the very early days.

• 18:36Speaker 02

  What has historically happened is that you had to get to a really large scale to become a target. And so it was okay that only the like Twitters and the Meadows and whatever had sort of an apparatus around these functions. The reality is that especially with AI and given what it can do in terms of deep fakes and cloning and all the rest, you have to assume that if you're working with this technology, you have to build those checks in way earlier. I think the big difference that's happening now is it's actually becoming part of the infrastructure layer. It's no longer something that is slowing down production. It's actually an accelerant. So if you build your stack, assuming that you are going to be constantly pushing new surface area, you have to assume that that new surface area will introduce new risk vectors.

• 19:21Speaker 02

  And so your whole production cycle has to be able to train on that. So I sit on the board of Cinder AI. They're publicly listed as a data subprocessor for OpenAI, a bunch of other large AI companies, specifically around this content moderation, GPT moderation. And so you just have to assume that this very powerful technology, whether it's T-SAM or non-consensual imagery, there's just so much surface area where people can do crummy things.

• 19:49Speaker 05

  Yeah.

• 19:49Speaker 02

  And we have to build, not with our head in the sand. We have to build assuming it. But I think the big distinction is it's not about slowing things down. It's actually about speeding it up. It's about if you put the apparatus on.

• 20:01Rebecca Bellan?

  So that you don't have to have to pay back and like change it all later when regulation, if regulation catches up.

• 20:07Speaker 04

  I might think about that a little differently. Large companies are actively working on it. Like Anthropic has a whole group working on model inspection. So does OpenAI, so does DeepMind. Like they will come up with ways to solve model explainability that instead of each startup having to solve that at their own use case, they should just go the fastest to produce the capability that they want to show, find the SLA, the customer experience that they want to provide. And then just bet that, yes, you know, two years from today, the cost of running these models is going to go down. And the fact that I can query the model and steer the model and inspect the model will also happen. It's too early for founders at this stage, like at the seed stage, early worrying about safety and ethics, where everybody's definition is so different. I don't think for every use case that should be needed.

• 20:48Rebecca Bellan?

  Before we get too into this topic, because we only got five minutes left.

• 20:51Speaker 02

  I was excited for this debate. I don't know if you guys, I feel like we should let it keep going. Yeah, what do you think?

• 20:55Rebecca Bellan?

  I mean, we can, but it's a touchy topic. Maybe we can, maybe I'll have you guys on Equity Pod and we can, we can really duke it out then. But as you said, like you mentioned that seed stage is too early to know about ethics and all these things. I personally disagree, but I think that these companies are growing really fast, it feels like, or they're able to. Like Jill, you're mostly focused on growth stage. Are AI companies reaching growth stage faster than past tech cycles, do you think?

• 21:19Speaker 05

  Yeah, it's interesting because it sort of depends how you define it. You could either define growth stage as sort of the valuation range or the revenue range, or for like, how far along is the company? Like, what have they de-risked? What have they showed? How big are they? How established are they? I think that in the former two categories, like from a revenue perspective and evaluation perspective, it's kind of the same stage as it always has been. It's just that so many of these companies have scaled so quickly that it's really interesting investing at the growth category because you're like, okay, I'm looking at a $50 million ARR startup and I'm paying, you know, X billion dollar valuation for it, which I'm used to. But typically, when I look at this size of company, they have a lot more infrastructure

• 22:00Speaker 05

  set up on the safety, on, you know, hiring, they have an executive team. And now you sort of look at this company, you're like, wow, this company was started 12 months ago in some cases, and so they have none of those things. Which in sort of, on one hand, like, that's really exciting. It represents, obviously, this brand-new trend of extremely fast growth, which is awesome. On the other hand, it's a little bit scary because you're like, okay, I'm gonna pay, you know, X billion dollars and valued at X billion dollars for this company when they didn't exist 12 months ago, and things are changing so quickly. So who knows who is in a garage somewhere, maybe in this audience somewhere, starting a company that in 12 months will be a lot better than this one I'm investing in, you know, that's a $50 million ARR today. And so I think it's made growth investing a little confusing, which I think, again,

• 22:43Speaker 05

  gets to this point Sarah started with, which is like, really what you have to feel good about is the category and the ability of the founder to like, very quickly adapt and see around corners and be able, you know, I think Cursor is a really good example where you're like, they jumped on the exact right use case for AI code generation that was like, available and possible, given the technology at the time. There will be, by the end of this year, AI software engineers, there absolutely will be. In that scenario, what Cursor has today is going to be a little bit less relevant. It is incumbent on the Cursor team to see that future and to think, okay, how do I start building my product so that when those models come out and are much more powerful, the product surface represents those and I can very quickly plug those in and sort of shift

• 23:27Speaker 05

  into that state of the world and that state of code generation? So I think it's sort of focusing on those types of qualities that's quite important.

• 23:33Rebecca Bellan?

  Okay, how do you avoid, or do you not avoid, the hype, right? When we're seeing VCs frothing at the mouth to get on certain cap tables? It's a little intense, but you don't want to miss out, so what's your thought on that?

• 23:45Speaker 05

  Yeah, it's a really hard question. We at Capital G tend to be extremely thematic. And so we have subsectors. So again, I lead a lot of our AI stuff and someone else leads fintech security, et cetera. And so I actually don't like, we would never enter a process that's like, oh, you have one week to decide on this deal and everybody is jumping and submitting a term sheet. That is not for us. We spend a lot of time, like I spend six to 12 months learning about a category, meeting every company in that category and then investing, which I think just allows us to either feel really good about leaning into the hype and paying what you have to pay to get into the companies because you've spent so much time working on it and you feel confident about the end result, or saying like, you know what, this hypey company is not for me because I don't know enough about it and I'm not willing to pay that price because I'm not confident in it. So that's our approach.

• 24:28Rebecca Bellan?

  Yeah. Any thoughts?

• 24:29Speaker 04

  We tend to be by and large thesis-driven investors. So think about a category, figure out where does the opportunity, given the advances in technology, where can you intercept and build something big and capture enough value such that when you meet founders who are already in that space or getting to that space, we can react very quickly. So we do not have such rules like, okay, if it's only a week, we won't jump in. We would, we would do our best, but still try to be very disciplined, have very strong conviction on why, even at a high price, this is worth entering and where do we see the 10x or 100x going forward. But yeah, we will try to have a response even for the fast-moving opportunities.

• 25:02Speaker 05

  I think you have to do it earlier. Like it's easier for me because I have a known set of companies at the growth stage. For you guys, you have to be able to do that, I think.

• 25:08Rebecca Bellan?

  Great. Okay. Well, that's our time, unfortunately, although I do think we'll have some fun chats backstage. But yeah, thank you. Thank you all for being here. Thank you.

• 25:17Speaker 05

  Thank you.

• 25:21Rebecca Bellan?

  Equity is produced by Teresa Locan Solo with editing by Kel. We'd also like to thank TechCrunch's audience development team. Thank you so much for listening and we'll talk to you next time.