00:00 — AI Costs vs. Engineer Costs 00:55 — Intelligence Is Getting 10x Cheaper 01:42 — How to Build a Startup Ecosystem 02:18 — GoCardless and the London Origin Story 03:58 — Where Should AI Judgment Stop? 06:18 — What's the Hard Bit Now? 08:13 — When AI Becomes a Better Partner Than You 10:18 — Why Founder-Partners Are Irreplaceable 10:50 — Research vs. Selling: When to Switch 13:09 — Get Out of Your Comfort Zone 14:02 — US Export Restrictions on Frontier AI 15:08 — Solo Founder or Co-Founder? 17:47 — Pick a Co-Founder for This Reason 19:35 — Is the VC Funding Path Still Relevant? 21:58 — What Are the Real Moats Now? 24:13 — Will Anyone Care Which Model They Use? 25:05 — AI Wrapper vs. Real Wedge 27:11 — Deep Tech Founder vs. Vibe Coder 28:24 — European Builder, No Budget, No Visa 29:48 — Early Decision Applications 31:38 — Habits of Top YC Startups 34:38 — Metrics for AI-Native Companies 36:10 — How to Actually Build an AI-Native Loop 38:04 — The Last Thing You'd Ever Automate 39:05 — When and How to Pivot 42:28 — Is High Agency a Muscle? 44:27 — Standing Out When Everything Is Easy to Build
AI Generated·May contain errors
Video Summary
AI Generated
Y Combinator partners discuss navigating the AI era, emphasizing that while AI drastically lowers the cost of software engineering, it shifts the value toward solving \"hard\" problems, high-agency leadership, and human-centric support. They advocate for empirical market validation over theoretical research and the necessity of high-density startup hubs.
So you were talking about AI-native companies, and I do really understand the capabilities because
it's really shocking, but how do you look at the cost efficiency? Because nowadays
AI is getting more expensive, so perhaps you might be able to do what an engineer can do
with AI, but when do the economics come in? Because sometimes it can cost even more with
the tokens than the engineer that gets it first time right. The first answer is you should just
join YC and you'll get a million dollars of free tokens and then offers for many million dollars
more. The second answer is the cost of intelligence is coming down so rapidly that even if that is not
true today, it's going to be true in six or twelve months. Like it's a 10x cost reduction per year,
roughly speaking, for the same intelligence. It's something we tell our founders not to worry about
too much because it will just solve itself with time. I think an engineer today with a model is
1,000x better than an engineer without AI. But would you not use it? I think a year ago it might
have been reasonable to say, well, a given engineer for a particular task might be more
cost effective than a model. I think there is no chance that is true today if you're using the
absolute best intelligence. Yeah, I'd probably get that. I just don't think there's a single
programming task that is more effective for a human to do than a model to do. Hello, thanks for
this opportunity. A lot of countries, especially in Eastern Europe, have underdeveloped startup
stage and underdeveloped startup communities. Do you have any tips on building especially a
community of founders, like not being a founder of a startup, but being a founder of a community?
How to build from scratch? Thanks. I was going to say move to SF. If you want to optimize the
chance of your startup, you should probably join an existing community versus having to create it
from scratch. And if you cannot go to SF, of course that's the biggest place to build a startup,
but go to London or Paris, where there is at least a critical mass of startups so that you
can surround yourself with great founders that are going to make you better founders.
If you are the only one in a city, it's very difficult because you don't have any peers to
push you forward. And the people who don't understand your life are going to hold you back.
Gray and I sort of experienced this. So I started a company in 2011 called GoCardless. It did
Y Combinator and actually came back to London. Gray was like our number two hire or something
like that, number three hire. And GoCardless and a small number, but really like less than five
companies were like the kernel of the London startup ecosystem. But you can kind of trace
from GoCardless this like family tree to many, many, many dozens of other companies. And TransferWise
is in there, Monzo is in there, and it's kind of amazing how it spreads. But in those early days,
so the direct answer is find a really great startup that embodies the startup
culture and join that. But in the early days of any great startup, it feels like a cult.
Like it feels like you believe something that no one else believes. And if you said it out loud
outside the room, people would laugh at you almost. But you like have these collective
set of beliefs and behaviors and ways of working that make you extremely effective.
And so I just try and find that basically. And the reality is there is always going to be one
global hub for startups. And that will attract the most ambitious people and the companies that
are going after the biggest vision. And you probably want to be there. It doesn't stop
there being other smaller regional hubs. But for a lot of companies and for the most ambitious
founders, you definitely should be considering San Francisco at the very least. My question would
be related to the thing you've spoken about, the artifacts we should be pumping into AI. So we as
founders should pump everything we could possibly find to AI. Investors would be doing the same
thing. Do you think that we leave too much judgment to AI? And where is that gap where we
should stop and leave our judgment to us and not outsource it to AI? Thank you. Yeah. So I mean,
there are many ways to answer your question. I think we can't answer all of it for sure.
Like one thing I've been arguing with Nico during the batch is how much AI should write for us,
because I think there is a direct correlation between the amount of thinking you do and how
you write. So I've personally been writing most of the things I've been writing so that I don't
delegate too much to the AI. That's one thing. The other thing that we've been talking about
is I feel like at YC there is so much emphasis on how AI can give superpowers to partners and
therefore can be shared with founders. And that's been the most mind blowing thing I've experienced
at YC is the speed at which this organization built tools to empower all of the people who go
there. I think that something that's super important for YC as an organization is to think
about the thing that AI is not going to be able to replace. And there are a few things like founder
well-being is incredibly important and putting more emphasis on how can we help founders take
care of themselves. Community is also something where AI is not going to help. Knowledge sharing
is something where YC has a unique thing where you have 200 companies that are obsessing over how
to make the most out of AI. And there will never be content available on the latest thing you can
do, but you're surrounded with people that are obsessing over this. And so one of the things
we've been talking about is not obsessing too much over what the AI can do in order to empower all of
you, but also what is AI not able to do and how do you make sure that you're excellent at it?
Because sometimes I feel like you can be blindsided if you're only looking at it in one way.
In the next months, everything can be done internally. I mean, every tool can be built
internally and instantly. What is the purpose of a B2B product anymore? Like if you can build
everything by yourself, so then everything that matters is just critical infrastructure.
So what will happen with B2B products?
So I broadly agree with the premise of your question, but I would reframe it more as a spectrum
than a binary one or the other. And the reframe I would take is like, what's the hard bit?
Like your startup has to have a hard bit. And 10 years ago, writing a lot of software
was in and of itself the hard bit. That is not hard anymore. And so you have to have other things
to be hard. To have any kind of durability, you have to have something that's hard.
And that might, in some cases, just be like brutally hard B2B sales into some industries.
Others, it might be a regulatory barrier, like getting a banking license almost killed me.
That's pretty hard. Or figuring out the physics of like spaceship launch, pretty hard. Actually,
most hardware in and of itself is just like hard. The atoms are at the hard bit.
And so I'd just be thinking like, on my spectrum of like, this is very,
very easy versus very, very hard. I'd like, I just shy away from the easier end of the spectrum
and just like pick bits that are harder and more ambitious because it will make your startup more
durable. So I would just absolutely not start something like Calendly today or Partyful or
maybe Partyful actually as a network effect, but certainly Calendly or like DocuSign,
basically dead. Like I think really pure software is just not, it's just too easy to replicate now.
So just ask yourself, where's the hard bit? You're recording office hours, you're recording
interviews, right? It seems like you're trying to do like, what you're trying to do is to make
Y Combinator an AI native company, right? Yes. Yes. What happens when the AI is a better
YC general partner than you are, right? What are you going to do? Are you going to be selling
Birkin bags or what? As Mathilde said, we're going to be therapists. Okay. Yeah. So I think
there are like, first, what I said I really believe in, which is there are so many things
that the AI will not be able to do. Like just so you know, and I think Nico can tell the story as
there is a new feature that's being experimented right now with YC where you can book an office
hour with a virtual partner. So you can literally, Nico was able to do an office hour with himself.
Currently it was really weird. So I think there is a part of it, which is like, you know, therapists,
community builders, but there is also, Gary Tan wrote an excellent post a few weeks ago about the
power of witnessing, which is the fact that you have someone in the room with you that acknowledges
what you're going through is something that's normal is incredibly powerful. It's an incredibly
lonely job and a robot doing this is not as helpful. So it's not just like, you know, the
human community and wellbeing and all this. I also think that there is a power of witnessing.
And I also think one of probably a misconception, I have, you know, new eyes on this because I was
a recent partner is you might think that at the scale YC has, a lot of advice is advice they give
to everyone. The reality is I feel like partners go really deep in understanding specifically
your business, your market, et cetera, so that they can give tailored advice. And so even if
you have a database of 7,000 companies, there is still so much that we don't know about your
company specifically, and that partners need to do a good job at understanding.
That's always fun to get like a fresh eyes on our job. But maybe one last thing on that,
all the partners at YC are past successful founders themselves who actually did YC.
And there is that, maybe that's the witnessing, we know what you are going through
because we did it ourselves. And that's a huge difference with a lot of investors out there
who have never been founders before, who don't know what you are living.
I think that piece is never going to disappear and cannot be replaced by AI.
So thank you for your talks. My question was regarding a research in AI startups,
because in my personal case, I built a startup while I'm doing like a PhD, okay,
in which I research like things for improving the product and all this stuff. So how to know
in each moment if it's like better to focus on their own research for improving this product
or focusing more on business, like to scale, get more clients and all this stuff?
I think it's an abstract question that don't have like right answer. Like the reality is what you
want to do is always confront yourself to the market in the fastest possible way. I was sharing
this with one of the groups. The biggest regret that YC founders have after the batch is not
having launched soon enough. So I think just know that you'll always have a tendency to research
more, to build more before selling more. And this is a tendency you need to go against.
And the more you can actually talk to users, confront yourself to the market and learn,
I think the better. Like one thing that we repeated over and over is just optimize for
learning constantly. And usually that equates to talking to your market. And then if you feel like
people are not willing to use what you're doing and it's not something that they want, then go
back to building, go back to researching. The motto of YC has never changed, like build something
people want. And to do that, it's a cycle, build, talk to customers, build, talk to customers. And
one thing that has really changed, it's the pace at which it's going today. That's why today we
have like some companies who do incredibly in just a few months. Like never since that,
back in our times, like companies doing like a million ARR in just by the end of YC. It's kind
of like unheard of before. And now it's happening because of that fast-paced cycle. And I think it
also applies to research. You want to make sure that what you are building is useful to someone.
Some projects take more time. Some deep tech projects, space projects, all of these things,
of course, have different timelines. So the advice cannot apply to everyone. But at the end of the
day, you want to make sure you build something people want. It's often useful to inspect your
own biases on this one as well. A lot of founders are a lot more comfortable building than they are
talking to customers. And the idea of going out to the market and trying to figure out whether
or not they're actually building something that people want is terrifying, because it may be that
you go back to square one, that all the work that you've done so far, you're working on the wrong
thing. And there's a lot of loss aversion that will dissuade you from doing that. But if you
want to build something that people want, that's the thing that you actually need to do. So if you
feel very comfortable in research, anytime that you're asking yourself this question, you should
just be aware that you probably have a bias towards the build a research part, and you need
to dial up the amount that you're talking to customers. I gave this advice like 15 times today.
I think Grey put it more eloquently, but basically that. Get out of your comfort zone.
So what do you think happens next with the US export restrictions on frontier AI models like
Fable? Do you think we will end up in a world where only US citizens can access the US models?
We don't know more than you do. We see the news the same way you do. So I hope,
like, I mean, for those of us who have tried these models before they were
blocked, yeah, they're crazy, crazy good. Can't wait to be able to use them again.
It's a wake-up call for the rest of the world that it was more dependent on the United States
government than it realized. I think there were a lot of people that woke up to that news and
I had no idea that a decision in the Oval Office would affect my ability to access these models.
And that will create demand for competitors and sovereignty over AI and inference that wasn't
there before. That's definitely created a market. What happens specifically in the US over those
models and what time period that happens? Yeah, we have no extra insight on that.
My question is, Sam Altman, a year ago was talking about in the next few years we're
going to have a one-person billion-dollar company. And what's the sentiment NYC
about having a co-founder or going solo? In theory, will you be able to
create a billion-dollar company with one person? Yes, probably. But
the way I see adding people is like marginal cost-benefit. And the marginal cost of adding
one person will be, relatively speaking, low compared with the benefit you get from them.
And it just seems illogical that you would, unless you're trying to break the record of getting a
one-person billion-dollar startup, you would just be more successful by adding a second person,
I think. But that ratio between the marginal cost and benefit actually changes quite dramatically
as you add more people because you add coordination problems. So having a co-founder,
you talk to each other. And third, it's Metcalfe's law, right? The square of the nodes is all the
connections. You have to maintain all of these different connections. So by the time you get to
100 people, it's quite hard to keep everyone on the same page. At 2,000 people, it's really,
really hard. And so I think what AI is going to be good at is compressing companies to keep
them down below Dunbar's number, so 150 people, where you can maintain relationships with everyone
pretty successfully. But I just don't think there's any point in trying to run a single-person
company. And I've seen this. We keep funding them, and some do well, but statistically, they do worse.
And I ran this experiment. At this batch, we funded a very, very, very strong solo founder
who wanted to build basically self-improving autonomous agents. They'd all communicate with
each other. He's like, I don't need a co-founder. And what happened is he got sad, right? Like six
weeks into the batch, he's like, had a crisis of confidence. He's like, I'm not sure this is
working. Is this the right direction? I'm like, dude, I mean, I can be here for you, but like,
this is what a co-founder's for. It's for someone to like pick you out of the gutter when you're
feeling down or like when you're hyperactive and manic, they like pull you back down to earth and
they modulate you. That's certainly what my co-founders did for me. And so, sure, theoretically,
I think you could do a billion dollar startup with one person. I don't think it makes economic
sense. I think the utility you get from adding a second person will always outweigh the cost at
that level, unless you're trying to break the record for some dumb reason. And I think they're
just very, very, very valuable. But I do think it'll compress companies down from 2000 to, you
know, 100 or 50 or something. And maybe one thing to add on what AI is actually changing on this
topic is I feel like now it is possible to go from being an average engineer to being a great
engineer, not knowing how to sell to being good at selling. And I do feel like the complementarity
of skills is less important. So, as Tom said, I think the fact that you have a co-founder is just
as important for all the reasons you mentioned. But I wouldn't obsess over, let's making sure
that our skills are complementary. Instead, just make sure that when you're sad, you're very happy
talking to this person and that person can bring your morale up. And just to make sure it's clear,
we do fund solo founders too. It's not, we have nothing against them. We just found that it's
easier as founders to have co-founders. And so it's solo founders, the bar is higher.
Yeah. At risk of rambling on, I want to give like 30 seconds on how to pick a great co-founder.
As Mathilde said, like someone who, you know, can help make you happy when you're sad is a
great start. I would think about who is the smartest, most determined, hardest working person
you've ever worked or studied with, and then add in like a integrity layer. Like, do you trust them?
Do you have aligned values? And would you be, if you're in a competition, would you be scared to be
on the other team? Like that's the kind of person you want to go for. And all the complementarity
stuff, I actually don't think matters at all. Like the biggest mistake I've seen is technical
co-founders thinking you need a business co-founder. I think that's almost always a massive mistake.
If you're a technical person, I would find someone who is as deeply technical as you
and who you work great with and who's like determined and smart and a high integrity,
all those things. The business stuff is actually very easy to learn, I think.
Hi, I have a question about the future of VC funding. So that now new employees are less
necessary than they used to be. And also the opportunity cost of having investor meetings
has also risen a lot. So I'm wondering, does the old path of seed and then the various VC rounds,
is that still as important as it used to be? Or has there been a change where new paths are
possible? You mean you can beat the company without VC money? Why not? I mean, I don't think
there is a rule here. I think for a lot of companies that want to grow fast, having access
to capital can still help a lot. Even software companies, because when you are going to consume
these tokens, it costs money. And so if you can leverage the VC money to simply grow faster
and do more, it may be worth it. Now, if in your case, you found out that you have enough,
whatever, like tokens or access to AI to move faster with less money, that's great.
And there is some truth here. Like, I don't think founders need to raise as much today
as they did five years ago to achieve the same thing, definitely not to achieve the same thing.
But it's the same work for everyone. Yeah, I agree. And I'd go back to the hard thing.
Like, if you are just producing the same software that people produced five years ago,
you can absolutely do that for way less money now. But unfortunately, it is much,
much less durable and defensible. And so what I think what AI is doing is enabling founders to
attack harder problems. But as those hard problems get more and more ambitious, then you're like,
the capital needs to get there, commensurately go up. So it's like the guy building the nuclear
reactor. Did I mention this? We funded a guy building small nuclear reactors, and he's going
to have to raise something like $800 million next year. And it's just like, it's really hard to
build nuclear reactors without tons and tons of money. It's hard to launch regulated banks without
tons and tons of money. So I think ambitious startup founders will just try and do harder
and harder things. And that will require, in a lot of, not all cases, but in a lot of cases,
capital. And that's where venture capital will come in. And I actually think that's great for
humanity, because we will get solutions to harder problems like curing all disease and producing
infinite abundant energy. So I actually want to bounce back somewhat on that. You've mentioned
that pure software startups are limiting in impact. And I'm sure within YC, this is one of
the main questions is like, what are the moats that you're betting on? And what are the moats
that you disagree with internally? I mean, at the end of the day, I think YC funds people. I
think this is the biggest moat, like by a wide margin. I think it was the case before. It's
probably just even more true today. And I don't think anyone at YC disagrees with this. I don't
know what we disagree on. I think every, like, on the disagreement. Disagreement is good. That's
interesting. Like a bunch of companies we fund, like companies I fund, if Tom meets them, he's
a few, there is maybe, I don't know, maybe 30%, whatever number that I'm going to fund, Tom would
not fund them. And so that's disagreement. And that's cool. Like we all have like going to find
things we like better than others. And all of that is cool. And so that's cool. On what we fund,
honestly, it's all depends of you. Like what do you apply with? We don't dictate what you should
work on. You figure it out. We fund you because we think you have the right team. And about the
third of companies, like last batch, a third of our company, like in our pod, devoted. Change
their idea anyway. And so the best is always the founders. Yeah. I mean, we can give an example.
There are so many examples of this. The Brex founders, Brex just sold for $5 billion. The
Brex founders applied when they were 19 years old and they were brilliant. They, as 15-year-olds,
I think 15 or 16-year-olds in Brazil, they'd started a payment network and sold it for
something like $50 million before they were 18. Like technically brilliant, amazing distribution
marketing business. And the idea they applied with was virtual reality sunglasses made out of
cardboard that you could build at home. And it was like, you guys can work on whatever you want.
Like those guys could literally have applied with any idea and we would fund them. So if the
founding team is strong enough, like it just doesn't matter. I think because we bet on great people
to figure it out. And if they figure out they need durable moats, great. And if they decide
that there's a really high impact software business that they should pursue, that's also
great. Ultimately, we invest in people. Do you think people will be thinking about which models
they want to use? Or do you think it will kind of just be automatic behind like this interface
with the AI? Like it would just be interacting with it, not caring which model they're using
underneath. Very quickly, it will be automatically routed for you without any perception. Your
spills will go down. The big labs will do it. And open source routing models will do it. And
it's like, and the timeframe on that, like 12 months probably is my guess. You'll just type
a query and it will decide what level of intelligence to route it to internally, or it
will like have a smart model and then spawn a bunch of sub-agents, which are like cheaper,
more effective models. That's not really the case today. Most of the time, except of course,
OpenAI is not going to use the models of Anthropic. So if you are using Chattipy directly,
of course it's OpenAI models. But if you are like a business, your customers don't care.
When YC selects companies, how do you evaluate in B2B AI a real wedge from an AI wrapper?
I mean, AI wrapper is this like derogatory term that is kind of meaningless. Like
everything is powered by AI or like most ideas that we are funding, because it is the first time
anyone has been able to attack that problem, have AI at their core. And like, whether or not they
are adding a lot of value on top of the core model, like, yeah, that matters. And so like,
if it's really like that thin layer, we're going to be looking at the team. We're going to be
at like how that is going to evolve, whether or not they're a team that's going to figure that
out, whether or not this problem space where there is enough depth to go really deep. It doesn't
matter how deep they go. Some people are still going to call it an AI wrapper and snicker to
themselves, right? They're so clever. Some of the biggest companies today are AI wrapper.
100%. Before them, they were SQL wrappers or S3 wrappers.
As you mentioned before, given the state of AI, the sort of cost of taking an existing solution
and implementing it into a product has gone down. So in terms of founder adding value to a product,
the way I see it, there's two ways of adding a value since the cost of implementation has
gone down to essentially zero. First one is actually moving state of the art in one area
and building a specific technical improvement that actually helps people and gives you that margin.
And the other way is essentially selling the product better and has this increased
capability of models and helping founders ship products change the ratio of, for example,
super technical founder working on a very specific niche technical issue in the batches and actually
people trying to market their existing solution better.
I'm going to summarize. It's distribution. Having a deep technical wedge can be really
important or distribution can be really important. I agree on both of those points.
We have founders in both camps. We have this guy with a MIT PhD in nuclear physics building
nuclear reactors. We have two guys who worked at a satellite company building re-entry vehicles
so you can do space manufacturing. We have guys doing radio frequency test equipment because they
were the radio engineers on latest SpaceX launches and on and on and on. Deep technical expertise.
And we have people who are self-taught vibe coders, really honestly. So I think that second
group is like, it's made us re-evaluate how deeply technical do we require founders to be.
And we don't know the answer yet. We were pretty biased towards founding teams with at least one
quite deep software engineer on. And we're testing whether that's still true. I think you still need
to be very smart and determined and like a systems thinker. But it's possible that with AI you don't
have, not all companies need someone who's deeply technical, possibly. Having said that, we'll fund
the full spectrum of the MIT nuclear PhD to the people who learn to vibe code in their bedroom.
So my question was going to be about more like cost and stuff. As a person that actually wants
to initiate in the startup culture and has the motivation to like go all in, basically move to
San Francisco, et cetera, but is from Europe and cannot do that because of visa and budget reasons.
Wait, who's from Europe on the stage? For example, we talked about like open
rate models and stuff. And how can like a Brock student compete with people that have like API
money to actually pay for these models versus like having a lower intelligence model, et cetera.
What would be your like... Are you a student? Yes. Okay. So out of this event, you'll get 25k
plus over 25k of credit. Oh, I didn't even know that. Congratulations. Everyone here, if you are
a student, I think with a valid student email... I don't know how we're going to do that. Probably
you're getting $25,000 of credits to use each. So that should help.
So then what would you recommend as like a European builder? Then that's my question.
Go build something cool. We've now solved your biggest problem. You've got to do the rest, buddy.
My question is regarding early decision. Like you officially state that early decision applicants
are treated the same as standard ones, but as a general partner or when you're looking through
that application or that interview, do you have a sort of extra checklist or do you see
something extra from them or is there anything that you expect from them? Because for someone
that's going to join the badge a year from now and he's applying right now, do you want to see
an additional drive or additional motivation for that? Like what is sort of that personal factor
that you want to touch on in an interview when you're interviewing someone of that sort?
Thank you. I don't know. I guess all of us are different. A year from now seems long,
but you know we do four batches a year. Each of us only work alternative batches, so I could look
now at funds for four because that's the next batch I'm going to do. Now the question I'm going
to ask myself is that why early decision? Like why would I not wait for the next application cycle
and would I fund you today? So that's kind of like that bar and there may be many answers to that.
Maybe if we don't fund you now, maybe you'll find a job and because you cannot afford not having a
job and you have to decide now, I don't know what that could be. But if it's a year from now,
I don't think I would fund you or maybe the bar is way higher. We want to understand why you need
early decision, right? We have that question on the application and we read that answer, right?
And for some people it'll be a very legitimate reason and for others it's like yeah I'm really
excited to do a startup but not so excited that I want to drop out. I'm saying I'm really committed
but not committed enough that I don't want to do another year and a half of school and that's
going to make us question whether or not you're going to have what it takes.
Thank you so much for the amazing day. Two questions about what happens during the batch.
What are the common traits and habits of the top-ranking startups? And second, you mentioned
you've updated the user manual. What are the non-obvious stuff except face that have changed?
The most successful founders launch early and often. That is the biggest thing by far. The
founders, for whatever reason, are too theoretical, they're too in their heads,
they're too afraid to put something out into the world and face rejection for possibly the
first time in their life ever. And so they convince themselves that just a few more weeks
of research or a few more weeks of building or whatever is a smart thing to do. And it's just
not a smart thing to do. There's a great tweet from the founder of Cursor showing all the times
he launched on Hacker News and it's like launch, zero upvotes, launch, two upvotes, launch,
two upvotes, launch, sell to XAI for $60 billion. And it's just like kind of insane that someone
like him is just is willing to put himself out there again and again and again. So just launch
early and make contact with the real world and then get data to like update your plan. Startups,
the problem that very smart board, like you got your handpicked to be here. Like we had
many thousands of applications and we picked you to be here because of all of the impressive
things you've done in your background. The problem though is many of you have, are in environments
where like sitting in a library and thinking really hard about a problem is how you get the
top grades. And that translates extremely poorly to startup success. It turns out humans are really
bad at predicting the future. And instead I think startups are like an extremely empirical exercise
where you need to come up with a hypothesis very, very quickly and then run a test with the real
world to see if the data you get back meets your hypothesis or not. And if not, you change approach
and running that cycle is like the key activity of an early stage startup. It's not sitting in
a library thinking really, really hard. So basically launch early and talk to users.
Snappy, so we made a catchphrase out of it. And set really ambitious goals and like keep
reassessing them every two weeks and hold yourself accountable to them. And keep overwhelming
the biggest actual problem that your startup faces. Like the best founders, every two weeks
we meet them, it's a new problem because they've overwhelmed the last bottleneck. And like, you
know, we chat through and we agree like, okay, like, yeah, that's the biggest next thing that
you need to go and focus on. And they go and like throw everything at that and they overwhelm it
and they make progress and come back and now it's another bottleneck. So like a tight timeline where
you are constantly like reassessing what the most important thing that you need to be working on is
and throwing as much of your focus at that rather than being distracted with the hundred other
smaller fires that you could be thinking about. Hi, you've mentioned the journey on becoming an
AI native company and with some processes such as rewriting the rulebook with better advice.
How would you say you define key success metrics that you implement for AI? Because normally for
an LLM, it's mostly about reading success and error about, for example, previous batches of
advice you've given. And also maybe taking this question more generally, what would be a key
advice you would give in us when we use AI as an advisor and how it could assist us successfully
and define some guidelines for AI to assist us in our journey as to become a founder?
I think the one of the biggest mistake you could do is just to use AI and just look at the result
and move on. Either you use it or you are not satisfied with the result and you move on.
I think a big thing, you know, what Tom presented was that cycle, self-learning cycle. And I think
that's one thing people, like when people hear like, you know, everyone says like AI is building
so much slope, it's true. Like the vast majority of AI output is bad. But once you start having that
self, like that loop, that feedback loop, once you start having an harness that actually
memorizes everything you tell it, that changes, that updates the skills. Once you start investing
in that, in building that, the results are completely incredible. Like in a matter of weeks,
it can become game-changing for your company. Hey Tom, this question is for you. You did an
amazing presentation, so that was great. I have a question that I think most people in this room
have about the AI-native company. I fully understand and I think everyone understood
because of your presentation. However, how can you actually practically start in building the
structure to build that AI-native company? Like do you go on Claude's code and tell him
exactly everything that you told us? Because we understand the idea, the concept, but
in terms of how do you actually do it? So thank you. Thanks. I'd pick a single loop, basically.
So presumably your company is small if you started a company. Yeah, I have a company and I want to
apply what you said and that's why I'm asking this. Cool. Yeah, I'd pick one part. I wouldn't
try to do it all. So first of all, try to make all the information legible, readable, right? So like
it's your emails, it's your Google Drive or whatever, your customer support tickets. I'd
try to make that like queryable to the agent and you can use something like Obsidian or Gbrain or
there are a bunch of other tools you can use. And then I would basically pick a single narrow
process and that might be like after every sales call, I should send a follow-up email to the
customer with the key points that we discussed and the follow-up. That's like a dumb example, right?
A more sophisticated example might be like during every sales call, I want it transcribed
and any product ideas, I want an agent in the background to build me a prototype of that thing
so that by the end of the call, I can demo it to my customer. So I just basically pick something
really, really narrow, try to implement it, see if you like it, see if it's good and if not, just
improve it. Just pick a narrowest possible implementation where you can see value. That
was perfect, thank you. Hi guys, let's say you guys are restarting your startup from scratch.
So looking back in the days and restarting an AI native company, what is the last thing you guys
would automate or what would you never automate at all? Talking to customers. I think that is what
kept me focused on what was important and it's the last thing that I would trust. It was the thing
that meant I was the person that was closing the loop and had all the context on like what it is
that we should build and I never wanted to stop having that information and so like delegating
that's the last thing for me. Even if you are like self-serve PLG, you talk to customers,
not all of them, but you need that direct exposure. I also think that you don't want to delegate
talking to your co-founder, like this is the main reason why companies fail during the batch,
co-founder issues and so just have real conversations with them. Thank you, my question
is about pivoting. I think we've consumed advice from YC that you should pivot, go do something
that's directly home. How would you update that advice now in the AI era that now that pivoting
is way cheaper etc etc and more specifically even if you have revenue and users and you are
profitable etc? Writing software is cheaper. Pivoting means throwing away a good chunk of
like what you've built which may be the hard stuff as well. I'm not sure that's got particularly
cheaper. It's really hard to give a like general answer to this one. It's like one of the like
primary things we would always say hey come and speak to us directly, to YC companies. I think
about this from a like framework perspective about like having a hypothesis. I have a hypothesis
that like this is the right business to build. The world is going to look fundamentally different
in five years time and then I have a load of sub hypotheses that are like this is exactly the right
product to be building and all the information that I'm gathering from the world when somebody
says no I don't want your shitty thing is like extra data testing those hypotheses and it's only
when I've got real evidence that my hypothesis that like fundamental of what I was trying to
build was in fact the wrong thing that you should pivot because you have genuinely exhausted
all of the options. Now normally that's not what happens. What happens is you run out of
enthusiasm. Like you're working on something the like kernel of a good idea is there. It's not
exactly going to sell itself because it's a long way from product market fit. You suck at sales
and you keep getting rejected and you get sad and as a result you think oh like maybe this is the
moment I should pivot. Maybe it would be so much easier if I went and worked on this like totally
other idea. The grass might genuinely be greener over there and that generally isn't a good way to
pivot. The grass will not be greener over there particularly if you're pivoting from something
that you are deep in into something that you are super shallow in because what you will find is
there are a whole load of other problems that you just weren't aware of and it's also going to be
really hard building something over there. There is another good version of pivoting which is that
through working on something that turns out to be really hard you are exposed to an even better
idea. You're working this kind of you know what happened with go cardless right like you're
working on like collecting payments for sports teams and sports teams turns out don't really
want this. The problem is not acute enough but the core of what you're building about getting access
to collecting payments you get a bunch of users that say well I kind of would like this for my
business just not for my sports team and there's a much bigger idea that you've been exposed to
and like that's the one that you want to run after but those pivots where your customers
pull the pivot out of you tend to be easier ones to make the decision about making. Normally we're
talking to founders that are getting sad and the question is like does this founder need a pep talk
or have they genuinely exhausted either themselves or all of the options in terms of like turning
this into a successful company no pressure. So this is a question for Tom and I suppose that the
other people on stage might share a similar view as well. It's about how in the future all the
gains from AI will accrue to a small number of high agency people. Do you feel like do you think
high agency is a trait you can really build some muscle you can build or is it just you're born
with it or not? I think high agency is basically a belief that your actions are going to produce
output like you actually are going to have impact right like and if I were trying to develop that in
myself or try to teach like a child to do that I think I would point them towards projects where
they could like that seemed hard enough to push them but like tractable enough that they could
actually like see some output and then just like continually make that harder and harder and
I just like pick something that seems quite difficult and try it and that might be a side
project and to keep yourself going I would probably try and find a co-founder and at some
point this looked like doing a startup right like find a friend who thinks really really smart and
determined and together see what come up with an idea of a side project to build that seems like
quite difficult but probably not totally totally impossible and then do that and see how it went
and then just like if that succeeds do another one and another one and eventually you're like
you end up building rocket ships to mars and that's sort of how it goes like no one actually
realistically starts by building rocket ships to mars right they build zip2 and then they build
paypal and then they build tesla and then eventually they build a really big thing
so I think that's in the abstract how I would approach that problem yeah I was going to say
I think you I don't think you can teach it but I think you can get better at it that's pretty much
what you described if you have the kernel there new some small core from which you can expand
you can train yourself yes uh so I've seen this strange phenomenon in sf recently that
every billboard and every plane and every balloon is a tech company nowadays and since building
everything became so easy there's this funny thing that companies just try to gather traction
regardless of what they do so I'm just curious how do you think about standing out in the crowd
when you build companies because way more companies are being built at least at the
early stages nowadays we had a nice talk from uh the brex ceo where he was saying that so brex in
sf you've been in sf uh I don't know six years ago or something sf was covered with brex ads
and it was basically saying the reason at the point they just bought all the billboards in
san francisco and the way it was justifying it was no one was doing it so just want to stand
apart and in order to stand apart gonna buy billboards and clearly now billboards are another
thing that can make you stand apart and so the job of a founder every time is to think about how you
can differentiate and if the answer was obvious then everyone would do it and so I you know this
what you need to obsess over and you always have an answer that's better for your company
specifically um I think it's an issue that every billboard in sf is about ai during my time all
the billboards were about tech already it's always always been the case in sf it's industry town you
go to la every single billboard is about a tv show a movie just the us has industry towns in a way
that europe doesn't so much but i really like matilda's answer which is like you have distribution
matters more than ever like that's clearly true and they're the way to get distribution like the
meta game is evolving so quickly you've just got to figure something out for yourself I mean
the clearly founder found a way to get reaching distribution right like I do agree that
distribution is more and more and more important than it ever was because it's one of the hard
things like software no longer is the hard thing about building the company distribution can be
one of the hard things but remember what mini game you're playing now if you're at the like
very early stage you are playing the mini game of like getting towards product market fit right
like finding some first customers and figuring out from them what they need and building that
and then getting to the point where you need to like put fuel on that fire and like scale it and
so when you're looking at those billboards and how companies that have product market fit are
getting their customers remember they're playing a completely different game the way that you stand
out from them to an individual customer is by giving them like real like white glove service by
like turning up in person spending an inordinate amount of time for what is like a small contract
and building something that exactly fits their needs that's the like unscalable thing that allows
you to compete with somebody that's putting up a billboard on the highway in san francisco thanks
everyone thank you