Latest / The Learning Corner by Precursor / Episode #75: Domain Expertise in Vertical AI, The Changing Math of Seed Investing, A Generation of Cheaters

Transcript

Episode #75: Domain Expertise in Vertical AI, The Changing Math of Seed Investing, A Generation of Cheaters · The Learning Corner by Precursor

• 0:00
  In the, like, two things can be true at the same time category, just because OpenAI,
• 0:04
  Anthropic, and Databricks and SpaceX are these far outliers and that the ceiling is higher, it
• 0:11
  doesn't mean that the average outcome is higher.
• 0:13
  And I think sometimes I hear people conflate these things.
• 0:15
  Yes.
• 0:16
  The top 1% of venture outcomes are bigger.
• 0:18
  And if you look at that VENCAB chart, they get bigger every cycle.
• 0:22
  That doesn't mean that the average company also is gonna have a proportionate increase in
• 0:26
  scale.
• 0:27
  It's the week of April 13.
• 0:29
  Welcome to the Learning Corner, a Precursor Ventures podcast.
• 0:35
  Welcome to the Learning Corner by Precursor.
• 0:37
  I'm Mia Farnam.
• 0:38
  I'm joined by RGP and managing partner Charles Hudson, where we will be talking through the
• 0:42
  articles and Goodreads that caught our team's attention over the past week.
• 0:47
  Our first piece comes from the newsletter, The Verticalist, from Euclid Ventures titled, No
• 0:51
  Country for Old Founders.
• 0:52
  And it asks a question that is super timely in the current AI era.
• 0:56
  Does domain expertise still matter for founders, or has the rise of AI tools leveled
• 1:01
  the playing field enough that youth and technical chops are all you need?
• 1:05
  The piece pushes back on the trend YC has leaned into, dropping its median founder age
• 1:10
  from 30 years old in 2022 to 24 years old by the 2024, with the majority of founders now 25
• 1:17
  years old or younger.
• 1:19
  Euclid Team ran a comprehensive analysis of 673 vertical software and AI companies that raised
• 1:25
  $15,000,000 or more in 2025.
• 1:27
  And the findings show that two thirds of leading vertical startups had at least one
• 1:32
  founder with prior domain experience, and that number jumps to 71%, specifically in vertical
• 1:38
  AI.
• 1:38
  Founders with domain backgrounds raise deals more than twice the size of those without.
• 1:43
  And what makes this particularly interesting is a counterintuitive conclusion of it all.
• 1:47
  As the AI stack becomes more accessible and more legible to everyone, domain judgment is
• 1:52
  not becoming less important.
• 1:53
  It's becoming more so because the product in vertical AI is the work itself, and that
• 1:58
  requires knowing what to build, for whom, and why they will trust you to build it.
• 2:03
  The data in the piece shows that the more regulated and credentialed vertical is, legal,
• 2:08
  public sector, health care, the more domain experience dominates.
• 2:12
  Even with this data, deals are getting done at breakneck speed with founders who have little
• 2:17
  to no domain experience.
• 2:18
  Charles, what do you think investors are actually betting on and and want your thoughts
• 2:21
  on this altogether?
• 2:22
  I think there were a lot of really interesting elements to this article.
• 2:26
  One of which is I forgot how, quote, unquote, old the average YC founder was back in the day.
• 2:31
  I had the mental model that they were younger, but if I think back to those batches, the the
• 2:34
  data they cited is correct.
• 2:36
  I think the question is, does AI change this dynamic?
• 2:39
  We were talking about this in our team meeting.
• 2:41
  Whenever I look at our the companies that are truly elite in our portfolio in terms of
• 2:44
  performance, it's a very striking bifurcation.
• 2:46
  Half of them are built by people who have deep domain expertise in the domain where they
• 2:50
  operate, and the other half do not.
• 2:52
  And I read the article, and I was like, this is one of those kind of cheesy yes and VC posts.
• 2:57
  He's got a quote in there from Keith about, hey.
• 2:59
  Like, most of the people who've built these really amazing companies have no domain
• 3:02
  expertise.
• 3:03
  And I get it because sometimes domain expertise can help you see all the limitations and the
• 3:08
  instincts it gives you are, well, this is why this thing won't work or these are the problems
• 3:12
  we'd have to solve.
• 3:13
  And if someone who doesn't have those constraints might just say, well, rather than
• 3:16
  try to solve that problem, how can I just avoid altogether?
• 3:19
  There are other cases in our portfolio where that deep domain expertise gives you insight on
• 3:22
  the nature of a problem, but it also gives you relationships.
• 3:25
  And I think the thing that when I read the post, was like, the other superpower that
• 3:28
  domain experts have.
• 3:29
  It's not just domain knowledge.
• 3:31
  It's relationships.
• 3:32
  It's the ability to get in front of a customer as a design partner and be credible, and to get
• 3:36
  them to tell you more about the problem they're facing than they would tell a stranger off the
• 3:41
  street.
• 3:42
  So I do think right now venture is maybe over romanticizing naivete, whether that's like
• 3:47
  youth, whether that's inexperienced in a domain.
• 3:50
  And I think a healthy counterweight is there will be great businesses built by people who
• 3:54
  are older and have domain expertise and know what they're doing.
• 3:57
  It's funny.
• 3:58
  I think of it as, a two by two.
• 3:59
  Is that a founder who is, like, more experienced in life?
• 4:02
  And is it a founder who's experienced in the domain?
• 4:04
  And I think right now, we have companies in all four quadrants.
• 4:08
  And I'll say, just being an inexperienced founder who's young with very little domain
• 4:12
  expertise is really hard.
• 4:14
  You have a lot to learn.
• 4:15
  I think it's a really interesting framing.
• 4:17
  Also, anybody who talks about consumer surplus and deflationary sovereign economics will
• 4:22
  always have my attention, but I thought it was a really good piece.
• 4:25
  There's some concern on younger founders being able to penetrate some of these more affordable
• 4:31
  sectors that have really dated ways of working.
• 4:35
  Not having maybe a bit of gray hair is needed in order to gain credibility.
• 4:40
  You mentioned relationship experience.
• 4:42
  I think related.
• 4:44
  There's also this belief, however, that if the product works and potential customers need it,
• 4:49
  then they shouldn't necessarily care who is selling it to them.
• 4:53
  Like, how do we think about that?
• 4:54
  And I think that's the other tale of this.
• 4:56
  When when I read the article, my immediate thought was, if you'd written this article five
• 4:59
  years ago, what would have been different?
• 5:01
  And I was, oh, AI, duh.
• 5:03
  The duh part is, I think previously, if you were gonna buy a system of record or a really
• 5:07
  expensive piece of technology, you probably actually wanted someone you thought was gonna
• 5:11
  be around, someone who'd lived your problem for a long time.
• 5:14
  It was kinda harder to break through.
• 5:16
  And in some ways, the rate of AI changes inverted that paradigm.
• 5:19
  And now a lot of people are like, no.
• 5:20
  No.
• 5:20
  I wanna buy from the upstart.
• 5:22
  I wanna buy from OpenAI or Anthropic.
• 5:23
  I don't wanna buy from some stodgy company we've been doing business with for ten years
• 5:27
  that's gonna, like, try to retrofit an AI solution.
• 5:30
  I wanna buy from the upper comer, and I think that is creating space.
• 5:33
  We have a portfolio company interface.
• 5:35
  It's two young founders that are working in sort of oil and gas safety, not exactly an area
• 5:41
  where startups have have thrived in the past, and they're getting a lot of entree and
• 5:46
  attention.
• 5:46
  And I think it's because that buyer wants to buy from startup with people who feel AI
• 5:51
  native.
• 5:52
  And so in some ways, it's creating new opportunities where maybe those would have been
• 5:54
  liabilities before.
• 5:55
  That's a really cool point because I think we always talk about it from the perspective of
• 5:59
  larger foundational LLM companies eating up all the competition and kind of edging everyone
• 6:03
  out.
• 6:03
  But it might just prime everyone to be open to new things that they weren't ready to before
• 6:07
  because it just didn't seem as powerful or important.
• 6:10
  The last question here on this was the piece that make a really interesting distinction
• 6:14
  between domain experience and then just being an industry operator.
• 6:19
  And the highest performing cohort that they called out by capital velocity was actually in
• 6:24
  the ex bankers, ex consultant background founders, not the ones who actually came from
• 6:29
  the industry itself.
• 6:31
  Does that match any pattern that you've seen in our portfolio or founders that you're drawn to?
• 6:35
  We have one person in our portfolio who fits that pattern, and she's fantastic.
• 6:39
  But again, that's kind of an anti pattern compared to the past.
• 6:42
  And so Yeah.
• 6:43
  I don't know.
• 6:43
  Every time I read articles like this, I'm always like, it's really great that there are
• 6:46
  multiple ways to win because this business would be really boring if there were only one
• 6:50
  archetype that could succeed.
• 6:52
  I do think, though, my hunch is that the consultants and bankers are maybe winning
• 6:56
  because a lot of the sell the work is basically AI powered business transformation at
• 7:01
  microscale.
• 7:02
  And I think those Yeah.
• 7:03
  And, you know, you've you've had that training.
• 7:05
  It does teach you how to break down a problem into processes and steps.
• 7:09
  And people whose minds work that way are pretty well primed to succeed in the era of AI.
• 7:13
  Even, like, more promptly, just to, like, founder skills or selling skills, consulting
• 7:17
  and banking, if they teach you anything at all, it's essentially to, like, be client facing
• 7:21
  Yeah.
• 7:21
  And really push things as hard as possible in a way that you know it will land on the other
• 7:25
  side of the table.
• 7:25
  So that piece isn't lost.
• 7:27
  Our next piece is a LinkedIn post from Micah Rosenblum from Founder Collective.
• 7:31
  And it's another really great breakdown of how seed round inflation has fundamentally changed
• 7:35
  the return map.
• 7:36
  For early stage investors, Micah in the post lays it out really plainly where a 500 k
• 7:40
  check-in an $8,000,000 post money round in 2012 bought you around 6.25% of the company, and a
• 7:47
  $500,000,000 exit on that check returned over 30 x after dilution.
• 7:52
  That same 500 k check today buys you roughly 2% at a $25,000,000 post.
• 7:57
  Yeah.
• 7:57
  And the same $500,000,000 exit now returns closer to 10 x.
• 8:02
  The check size didn't change.
• 8:03
  The exit didn't change, but the return got cut by more than two thirds.
• 8:07
  He's really careful to acknowledge that today's seed company is more mature product than twenty
• 8:11
  twelve's, closer to what a series a maybe looked like a decade ago, but his real question
• 8:15
  is whether the math still pencils out the same way and whether the biggest alpha is still
• 8:19
  hiding in the deals that are hardest to fund.
• 8:21
  I was thinking about this, and I think two things can be true at the same time.
• 8:24
  Valuations have gotten out of control and out of hand, and you can continue to do them
• 8:29
  because you believe that they are the strongest companies and they have the strongest outcomes.
• 8:33
  I know some of our best deals came at really modestly priced valuations, but how do you
• 8:37
  build the discipline and eye to know which ones are worth paying up for and which ones are
• 8:42
  worth staying uninvolved?
• 8:43
  We talked about this in our L PAC meeting earlier this week when I got together with our
• 8:47
  LPs.
• 8:48
  And I think there's a bunch of things.
• 8:51
  Mike has been on a roll lately with some good posts.
• 8:53
  Yeah.
• 8:53
  He really has.
• 8:54
  One thing I think of, everyone says, well, Seed is the new series a.
• 8:56
  I'm like, except we're not buying 20%.
• 8:58
  If seed really is the new series a, we're applying a seed ownership strategy to series a
• 9:03
  prices.
• 9:04
  And I think in most things in venture, when you have a price strategy mismatch, you should
• 9:07
  always be concerned that maybe you've unintentionally or intentionally broken a rule.
• 9:12
  I do think that in some ways, these more mature companies and they're more certain,
• 9:16
  historically, that's been a recipe for buying more.
• 9:19
  And maybe one answer is seed investors are actually not buying enough of these companies
• 9:23
  to make the math work.
• 9:25
  So maybe the right answer is you should still buy 10% at sea.
• 9:28
  It's just gonna cost you 2 and a half million dollars instead of 500.
• 9:32
  Okay.
• 9:32
  If you're gonna do that, you're gonna need a much bigger fund.
• 9:36
  The other thing in the, like, two things can be true at the same time category is just because
• 9:41
  OpenAI and Proppant and Databricks and SpaceX are these far outliers and that the ceiling is
• 9:47
  higher, it doesn't mean that the average outcome is higher.
• 9:50
  And I think sometimes I hear people conflate these things.
• 9:53
  They're like, well, like, outcomes are bigger.
• 9:54
  Yes.
• 9:55
  The top 1% of venture outcomes are bigger.
• 9:58
  And if you look at that VENCAP chart, they get bigger every cycle.
• 10:01
  That doesn't mean that the average company also is gonna have a proportionate increase in
• 10:05
  scale.
• 10:06
  So I think the nightmare case is sort of what Micah outlines, which is that valuations grow
• 10:10
  faster than exits on average.
• 10:12
  And so the tricky thing is how do you know when to pay up?
• 10:15
  And I talked to one of our LPs who's a GP at a series a five.
• 10:20
  And he told me some interesting data.
• 10:21
  He's like, you know, there's, like, a weak correlation between entry price and outcome at
• 10:26
  seed.
• 10:26
  But he said at series a, many of the best companies became expensive and remained
• 10:31
  expensive.
• 10:31
  The TLDR, after trying to parse it with him for a while, was there are good deals at seed.
• 10:36
  There is some mispricing of assets at seed, and you will find good companies that are cheap.
• 10:41
  And his argument was by the time they get to the a, there are so many i's and there's so
• 10:45
  much money that there aren't that many series a bargains out there to the market, at least
• 10:49
  right now.
• 10:49
  So it gets corrected in some sort of fashion.
• 10:51
  Which kind of makes sense.
• 10:52
  Would think Got it.
• 10:53
  With more data and time, the market would have a better sense of the act.
• 10:56
  One thing I I wanted to peel back is when Micah was mentioning companies today are a little bit
• 11:01
  more mature in product than they than they looked a decade ago.
• 11:04
  My statement to that is, like, of course they are.
• 11:07
  Now we have more tools available for people to build much quicker and bring something to life
• 11:11
  with fewer hands around the table.
• 11:12
  So my question is, should we be rewarding people for that?
• 11:15
  You get brownie points for that, or should we be collectively not moving the starting line so
• 11:19
  far ahead and rewarding that when that almost is table stakes
• 11:22
  It's a great and there's two other things.
• 11:24
  Also, these companies have more time and more money.
• 11:26
  You So would think if you give a company better tools, more time, and more money.
• 11:29
  And this is, like, one of my long held beliefs is that venture is always about relative
• 11:35
  performance, not about absolute performance.
• 11:37
  And I think founders are always thinking, like, oh, well, all of these companies that raised,
• 11:42
  they were at a million ARR.
• 11:43
  So that must be some absolute bar.
• 11:45
  I'm like, no.
• 11:45
  In the era that those companies raised, that level of financial performance was considered
• 11:51
  top tier.
• 11:52
  And I think that's the game of investors to figure out what does elite performance look
• 11:55
  like in terms of either metrics or team composition and invest in the teams that you
• 11:59
  think are elite.
• 12:00
  And I think what elite looks like changes every six months.
• 12:04
  Maybe in this in this area, maybe it changes every three months in terms of Yeah.
• 12:07
  Growth rate, in terms of total level of revenue attainment.
• 12:10
  Nothing is static.
• 12:11
  And so I'm always asking myself, if we're gonna pay 50 posts for a company, it better be
• 12:16
  amazing.
• 12:16
  And the only rule I can use is I gotta believe we're gonna see 200 x price appreciation from
• 12:21
  this point.
• 12:22
  And sometimes I'm like, wow.
• 12:23
  This is a great team in a big market.
• 12:25
  I just closed something at 40 post because I think there's a reasonable chance it becomes a
• 12:30
  $10,000,000,000 company.
• 12:31
  Two things can be true.
• 12:32
  I can believe that maybe it's a little bit over what the price that I would feel comfortable
• 12:36
  with ten years ago.
• 12:36
  I can still believe that it's a good deal to come into relative to the outcome that I think
• 12:40
  it's had to do.
• 12:40
  The real piece of it is, like, there is really no functional difference losing money on the
• 12:44
  company at five and 50.
• 12:45
  You're still gonna lose your money.
• 12:46
  Well, our last piece is from Kyla Scanlon from The Substack, The Argument titled A Generation
• 12:52
  of Cheaters, making the argument that AI has dropped the cost of cheating to essentially
• 12:56
  zero.
• 12:57
  And when the cost of a behavior hits zero, it stops being a deviation from the norm and
• 13:02
  becomes the norm.
• 13:03
  Over half of American teenagers have already used AI for schoolwork, and the police draw the
• 13:07
  straight line from that reality to a much bigger legitimacy crisis.
• 13:11
  The social contract that once made honesty worth the effort, work hard, get a degree, get
• 13:15
  a good job, is fracturing.
• 13:17
  Students and, honestly, a lot of parents are doing the math and concluding that the rules of
• 13:21
  a game no longer believe in are not worth playing by.
• 13:24
  And Pilot closes with a real question underneath all of this, which is not how do we
• 13:29
  catch the cheaters, but whether we can rebuild something worth being honest for.
• 13:34
  Charles, you're at a Starbucks right now on Stanford's campus.
• 13:39
  You're also not just the parent, but you're a parent living in SF proper.
• 13:42
  So I'm curious to know how your perspective has maybe evolved at all about this concept of what
• 13:47
  is considered cheating.
• 13:48
  What's a good use of your time as a young person attempting to be successful and strive
• 13:52
  towards a great career while the ground constantly moves beneath you?
• 13:55
  It's so hard because I think if I think about what cheating looked like when I was growing
• 13:59
  up, it was buying a little yellow and black book that summarized the novel you were
• 14:04
  supposed to read for a class, or it was a calculator that could do graphing and algebra
• 14:08
  for you.
• 14:09
  You still kinda had to know how to manipulate these and you still had to perform.
• 14:13
  And I think what's so seductive about the LLMs is you don't have to do any of that.
• 14:16
  The sort of the returns to cheating are so high.
• 14:19
  And on the one hand, we talk about it as cheating, but we're giving people this very
• 14:22
  mixed message.
• 14:23
  We tell them, Mia, you must become very good at these tools if you ever wanna have any hope of
• 14:28
  getting a job, and you absolutely must not use them in education.
• 14:30
  Which begs the question, well, then when are you supposed to learn how to use these tools?
• 14:34
  And I remember when the Internet first came, I was like, oh my god.
• 14:38
  People are just gonna look up answers, and they're gonna copy those answers.
• 14:41
  And I was like, yeah.
• 14:42
  That's we figured out how to deal with that.
• 14:44
  And I just worry that with AI, the returns are so high, and the skill level required is so
• 14:51
  much lower.
• 14:51
  And there's one other thing that I I've become particularly attuned to.
• 14:55
  You know, I teach at the at the business school, and I have a lot of friends who teach.
• 14:58
  In many cases, different classes have different standards for acceptable use.
• 15:02
  I don't envy the students who are like, well, in this class, I'm not allowed to use AI at
• 15:06
  all.
• 15:06
  It's considered cheating.
• 15:07
  In this class, I'm allowed to use it, but I have to disclose how I use it.
• 15:11
  And in this class, it's required.
• 15:12
  So how do you cognitively keep different standards for different use cases and different
• 15:17
  classes in your own head that seems like an impossible challenge?
• 15:22
  Also, think the temptation of having a chatbot where you could type something in and it can
• 15:27
  give you an answer so fast that you could get cold called in class.
• 15:30
  And maybe if you're good at typing out of the table, could maybe even get an answer while you
• 15:34
  stall or vamp for time.
• 15:36
  And this is just a new paradigm, and it feels like cheating today right now because it's a
• 15:40
  shortcut.
• 15:40
  I don't know if we'll still think of this as cheating in twelve months.
• 15:43
  And I do go back to even my own experience.
• 15:46
  I think of my AP in US history class in high school, and my teacher was one of the best ones
• 15:52
  in our school, and he did this exercise called the gauntlet.
• 15:55
  It was just four desks in the center of the room, and you had all week to prepare for
• 16:00
  basically a a group debate.
• 16:02
  Mhmm.
• 16:02
  And then he would call on whoever and you had to do it in front of everyone in the center of
• 16:06
  the room.
• 16:06
  Yeah.
• 16:06
  You could bring your notes if you would like, but obviously no one's bringing laptops.
• 16:10
  You can't really fake that.
• 16:11
  I actually think like that would have been a really good use of melding together
• 16:14
  foundational LLMs prep my work and prep my thought process.
• 16:17
  I don't have that laptop in front of me when somebody across the table for me brings
• 16:20
  together an amazing point, I have to rebut that in the moment.
• 16:23
  That's right.
• 16:23
  That's a test of my critical thinking Yep.
• 16:25
  And my delivery, which are skills.
• 16:27
  Like, I think Yoni from Slow released that whole piece that was like, hot people are
• 16:31
  always gonna be in the market.
• 16:32
  Yep.
• 16:32
  Hot can mean a bunch of different things.
• 16:34
  Yes.
• 16:34
  Hot meaning you're able to corral a lot of attention and excitement, and you can orate
• 16:37
  things in a really digestible way.
• 16:39
  And I remember, like, that APUS history teacher, he would run multiple gauntlets Yeah.
• 16:44
  Over the course of one day.
• 16:45
  And people would try to cheat the system by asking, like, my gauntlets after yours.
• 16:49
  What were the most insightful things that your gauntlet brought up?
• 16:51
  I can repeat them.
• 16:52
  But, like, he knew people were talking and cheating, and they just don't land better the
• 16:57
  second time.
• 16:57
  So that piece, I think, can still live on.
• 17:00
  One thing did feel like was relevant for where we sit today, how do we still instill a sense
• 17:07
  of grit and critical thinking, and how to make sure that founders can also screen for these
• 17:13
  things when they're hiring because they're gonna continue to hire younger talent.
• 17:17
  And how can you actually test if foundational work ethic gets set into them before they join
• 17:22
  your team if the rules are being shifted?
• 17:24
  In some ways, I think younger people have it easier because they're gonna have to navigate
• 17:29
  all of this stuff, having it present from the beginning and figure out, like, what does it
• 17:33
  mean?
• 17:33
  Versus, I think, people who are maybe older in life or further on in their career, they're
• 17:37
  gonna have to figure out how do I retrofit the existence of AI to my existing paradigms of
• 17:41
  what fairness is.
• 17:42
  Mostly the adults that are doing the embarrassing things, like I look at these
• 17:45
  lawyers who go to court and cite court cases that never happened, academics citing papers
• 17:51
  that were never written, people citing arguments that were never made.
• 17:55
  I do think at some point, it's gonna become professionally embarrassing to simply throw
• 18:00
  something into an LLN and just ship whatever comes out without any review.
• 18:04
  I kinda feel like we're already there.
• 18:06
  The question is, like, where will the bar be for laziness?
• 18:09
  When I read the post, I also thought there's a difference between sort of, like, cheating and
• 18:13
  laziness.
• 18:14
  Yeah.
• 18:14
  And I do think, to some extent, part of the appeal of LLMs is the ability to be lazy, is
• 18:20
  the ability to just say, I don't wanna think about this.
• 18:22
  I don't wanna read this article.
• 18:23
  You read it and tell me what it's We don't know I need to know.
• 18:25
  I don't wanna do this problem.
• 18:26
  You solve it.
• 18:27
  You think about it.
• 18:27
  And I don't I don't think that's the right way to utilize these tools.
• 18:30
  Well, we are out of time.
• 18:31
  Thank you, Charles, for jumping on a train early to make the recording session from Palo
• 18:36
  Alto.
• 18:37
  Thank you to our listeners and subscribers for jumping in.
• 18:39
  Please like and subscribe and share with friends, and we will talk with you all next
• 18:42
  week.
• 18:42
  Thank you.
• 18:43
  Bye.