"AI Startups" are over done (finally)
Duration
51:50
Captions
1
Language
EN
Published
Sep 18, 2025
Description
A lot of people think that these days investors will only ever invest in AI companies, but it's starting to look like that's not the case at all... Thank you Blacksmith for sponsoring! Check them out at: https://soydev.link/blacksmith SOURCE https://x.com/garrytan/status/1936559113414074557 Companies I mentioned: https://nautilus.co/ https://cocreate.so/ Want to sponsor a video? Learn more here: https://soydev.link/sponsor-me Check out my Twitch, Twitter, Discord more at https://t3.gg S/O Ph4se0n3 for the awesome edit 🙏
Captions (1)
As you guys probably know by now, I
invest in a decent number of early stage
startups that are building things in the
general space that we're all in. Which
means that a lot of these AI dev tools
that you guys see all the time are
things that when I'm able to, I reach
out to, try them out, and if I like
them, I throw some money their way so
that if they do succeed, I can make some
more money in the future. Which is why I
just invested in Nautilus, the car wash
platform that converts, captures, and
keeps members. What else did I invest in
in this batch? Um, co-create
professional first tooling for faster,
smarter, better video edits. These
aren't developer tools. What's going on?
Where did I even find these companies?
Well, uh, turns out I'm finding them
from Y Combinator. Because I was a Y
Combinator founder myself back in winter
of 22 before I was actually taking
YouTube very seriously. I went from one
of the biggest Y cominator haters to
somebody who saw a lot of the value in
it to slowly really really liking the
team, the vibes, and most of what goes
on over there. Joining YC is one of the
best things I ever did for myself, my
company, my YouTube channel, and now my
overall portfolio. It has been awesome
for me in almost every single measurable
way. So, why the hell am I investing in
a car wash company? I'm sure there was
lots of other dev tools and things in
this batch, right? Well, that's what
we're here to talk about today. There's
a lot of stereotypes about YC and also
about investors both like myself and
ones that are very different from me
about how we think about making new
companies. In particular, this idea that
AI is the future and all of these
businesses should be shoving AI into
everything if they want to make a lot of
money and raise a lot of money. But
things changed a lot with this most
recent batch. There has been a
significant vibe shift that I am very
excited to talk about with you guys
because a lot of your assumptions about
YC companies are correct. A lot aren't
correct, but most importantly, a lot of
the assumptions that both you and I have
have changed meaningfully, especially
with this most recent summer 2025 batch.
I can't wait to tell you more about what
I see is changing here and how it's
going to affect the hopefully entirety
of our industry, especially because I
want to make some money here. Well, if
you know anything about early stage
investments, you know they take 5 to 10
years to pay out. And I've only been
doing this for two. So, uh, my bank
account's running a little bit dry.
Quick break for a sponsor and then we'll
dive right in. You know what's worse
than spending all day waiting for your
code to build? Not knowing why it breaks
when it fails. I still cannot believe
just how bad GitHub actions have gotten.
But what if it could be better? What if
it was faster, cheaper, and actually
observable? Today's sponsor is here to
show you what that looks like because
Blacksmith is killing it. You make a
oneline change in your GitHub action and
all of a sudden life is better. Look at
all of the companies that have already
made the move from Clerk to Terso to
Superbase and more. They're faster
because they use gaming CPUs. That might
sound silly, but traditional server
processors have way more cores and way
lower clock speeds. A lot of the work we
do for compiling, even your fancy Rust
code bases, by the way, is very
singlethreaded. So having one fast
processor is actually quite a bit
better. Up to two times faster than
GitHub actions are traditionally. Cache
downloads are significantly faster and
Docker builds fly because they actually
cache the Docker layers on the NVME
drives that are in those same machines.
That would be enough by itself, but the
observability is more and more my
favorite part. Their dashboards are
great. You can see how big the failure
rates are on various things. You can
read through the logs with an actual
search and filter. Do you know how
useful this is if you have crazy builds
going on? It's at this point I just wish
GitHub would buy them because this is
how actions should work. If you're
running CI for your code, you deserve
better and your engineers do too. Check
them out today at sooyv.link/blacksmith.
So before we can go too far, I want to
give a brief like what is YC cuz I'm
sure some people watching this aren't
familiar. They probably even think and
I've seen this enough times that YC and
VC are the same thing. VC equals venture
capital is one of the many ways people
can invest money in hopes of getting
back more money in the future. Venture
capital is different from traditional
investing because you have to put in a
lot more money. You're taking a lot more
risk. The potential upside is much
higher, but your ability to get the cash
back and your overall liquidity is much
lower. When you invest in a company that
is early stage, you are effectively
handing the money to either get equity
or for a promise of future equity that
you cannot do anything with until a
certain amount of time has passed and a
certain number of things have happened.
If I own half of some really successful
company and they haven't IPOed yet, I
can't do anything with that stock. I'm
just sitting on it. It doesn't matter
how valuable they hypothetically are if
I can't do anything with the stock.
Whereas, if I go invest a bunch of money
into Nvidia and they jump 70%, I can
just sell it and I have my cash back in
a few minutes. That's the big difference
between venture capital style company
investment versus traditional stock
purchasing is you have given up all of
your ability to sell. And if the company
starts going down instead of up, you
can't do anything about it. You're along
for the ride. Which is why there are
such strict rules about who can and
can't invest in early stage startups.
It's a method of protecting people
because if you don't make enough money,
you don't have enough assets, and you
invest a bunch of money in these
startups and they start failing. You
can't get your money back. It is
incredibly rare for any of these early
stage startups to give money back to an
investor outside of failing and emptying
their bank account as they go bankrupt.
very very uncommon for people to get
their money back. But then if it does go
well and you invested in a company when
they were worth 20 mil and it turns out
they're 20 bill, you're doing pretty
well. So that's what venture capital is.
So what the hell is YC? YC is Y
Combinator. Y Combinator is kind of like
a boot camp for early stage startups. It
was built by Paul Graham back in the day
and it is still run actively in San
Francisco. Their goal is to help
earlystage companies, founders and
people who want to build some business
get the help they need. peers to run
alongside and then investment both from
Y Combinator as well as from other
potential VCs to get them accelerated
and going so they are most likely to
succeed with their business going
forward. Most early stage startups fail
aggressively. And why combinator through
a combination of how picky they are
about who they let in and the quality of
their whole accelerator process when you
go there in person for 3 months to learn
all about startups and how to run them
and scale them and get customers and all
of that. That process combined with
their pickiness results in a really high
successful pick rate for these Y
combinator companies. Here's a post from
Gary Tan. I'm very thankful he posted
the screenshot during one of the demo
day pitches because I would not have
permission to share this type of
screenshot. But since he has on Twitter,
you can tell it's a screenshot cuz it
has the iOS bar on the bottom and it's
also blurry as hell from Zoom. Now I'm
actually allowed to talk about this,
which is very, very fun. The returns for
investors who invested in early stage
companies in 2017, since then, the top
10% have had around a 3 1/2x return over
that time. So from 2017 to now, they
3.5x their money. The bottom 25% have
only added about 30% of their money
through their investments. 2017 was a
bit of a bull run with some really
successful companies. So a lot of
investors, even the worst ones, saw a
return there. But 2018 onwards, you'll
notice some things. Bottom 25%, so only
at about 6% improvement. 2019, 1%. 2020,
they're actually down 7% because they're
taking losses. 2021 they're down 13%.
Because they've had things go to zero
because all a startup can do it's like
it's not going to 2x. Most go to zero
and a few 10 to 100x. So your portfolio
is primarily composed of companies that
failed with the occasional one that wins
that boost it all up. So how does this
compare to YC's numbers for investors
that are focusing primarily on investing
through YC companies and two YC
companies? investors who have backed at
least three Y Combinator companies in
each demo day batch from 2018 through
2020.
The top 10% have 15xed their money. The
bottom 25% have still 3.3x their money.
The bottom 25% are doing as well as the
top 10% of nonc focused investors. YC
leads the industry. It is the best asset
class in venture. These are very, very
crazy numbers. If you're not familiar
with the term unicorn, it it means the
company reaches a billion plus dollar
valuation. And historically, each YC
batch has an average of a $6.5% unicorn
rate. Each investment either becomes a
success, in this case, unicorn, or it
doesn't, and it goes to zero. The
average valuation at demo day is around
$20 million post money save. So, it's
like the the value you're investing in.
This implies a 50x multiple needed to
achieve a1 billion plus dollar outcome.
point I'm trying to make here in
particular is that YC companies have a
very high chance of success. More
importantly, if you know about the
numbers YC takes, because YC gets 7% of
your company and a little bit extra on
top through an MFN, we don't need to go
in the details of the fundraising, but
they get between 7 and 11% of your
company. That's a huge amount of take,
especially for a less than great amount
of money. They only do 500K and 125K of
that is for the 7%. So, that's kind of
insane. But if you look at the value of
the average company that goes through YC
versus the ones that don't, you see some
crazy things. YC's round sizes are 23%
larger. Their valuations are 65.7%
higher. And their ARR for when these
companies are getting invested. So their
actual revenue is 63% lower. So YC
average round size for a YC company 3
million, average valuation 25 million,
average AR 100K. So 3 mil over 25 you're
getting about 10% dilution a little bit
more than that versus a nonc company 2.5
million raised 15 mil cap 3x the ARR so
you have to make three times more money
to raise slightly less money at a
significantly lower cap you're giving up
almost double the percentage of your
company by not going through YC from
your first round alone that 7% ends up
being worth it YC is a surprisingly good
value for what it is. And that is why Y
Combinator has stayed as relevant as
they have and also why they've made many
changes throughout. When I did YC, they
were doing two batches a year. Every
winter and every summer, they had a
3-month window where a bunch of
companies flew to San Francisco, moved
there, and then went through the batch,
learned as much as they could about
startups, got thrown in front of a ton
of investors at the end on a day called
demo day. And now they get to pitch to
these investors what they were building,
how much money they're making, and what
they need to succeed so they can
hopefully get investments in auction
style to increase the value of each of
those companies so they can go off and
hopefully make millions of dollars and
become a billion dollar plus company. So
previously we had those two batches, one
every 6 months. That was a three-month
window and they had 3 months off. Those
batches had increasingly large numbers
of companies. Like when I was doing it,
if I recall, it was in the 3 to 400
company range. Now they've knocked it
down a bunch to around 120 to 140
companies. But they also doubled the
number of batches. So they do one every
three months now. Winter, spring,
summer, and fall. The summer batch just
finished and the fall batch is just
starting. Now that introduced a very
interesting change. The number of
companies being like let's say
hypothetically
50,000 companies apply. So we have 50k
companies. If we're accepting the top
400 or so, like they did for my batch,
I'll nug it to 40k. So, it's the top
400. So, they take the top 400 of the
40k. That's about a 1% pick rate. YC's
pick rates are historically much lower
than even the hardest to get into
colleges. So, this has resulted in
problems where a lot of the people who
are hunting for the prestige of getting
into Harvard end up going ham trying to
get into Y Combinator because they've
just had their brains rotted with like
prestige hunting. Very weirdly common.
Not my thing at all. I don't give a [ __ ]
about prestige or credentials or any of
that. I just want to see what you're
doing. So, that part's not interesting
to me. But the fact that this pick rate
is 1%. Means that they are already
cutting out the majority of the options
that they had, the majority of the
companies that could have went through.
And I'm not saying that there aren't
good companies in the other 39,600
that applied. In fact, some of the best
companies out of the whole 40,000 are
going to be rejected. That's a cost they
choose to eat because they want to have
a good enough filter to increase the
likelihood that this 400 is at least one
standard deviation better than the
overall pile of ones they did not pick.
That was a 1% acceptance rate. But
things have changed. One, more batches.
So they're doing multiple batches a year
now. Four instead of two. Point two,
this one's really important. The number
of companies they're accepting has went
down, not up. They're only letting in
120 to 140. But most importantly, one
that you might suspect, the number of
applications has gone up significantly.
There are way more companies applying
than ever. I've heard numbers as crazy
as like a 100,000 companies applying per
batch. It's kind of insane how high the
application rates have gotten,
especially when you consider how low the
acceptance rates are. So went from 400
over I made up 40k. It's probably like
60, but you went from like a little
under 1%.
140 companies over 100,000 apps. Now
we're at like.14%.
We're in an entirely different world. I
don't know about you guys, but if
there's any quality to their process of
picking, which we've seen the numbers,
there is. If I have the choice to pick
between the top 400 of 40,000 or the top
140 of 100,000, there's a very good
chance that if you're picking from this
set, you're going to do very well. And
also, there's a good chance you're
picking from this set, you're still
going to do pretty damn well. This is
fun for a handful of reasons. One of
which is to go back to this post. These
numbers are based on 2018 to 2020 where
the pick rate was much higher. more
companies were getting in against a
smaller set of applications. If I was to
be frank, if the Theo that applied to YC
in 2021 that got in for Winter22, the
Theo that had recently quit his job at
Twitch had not worked much at other
companies beyond a little bit of AWS
stuff and Microsoft stuff. Didn't have a
YouTube or a Twitter that was relevant,
was just building tools for creators.
It's not that surprising I snuck in
here. Twitch as an XYC company, me as a
person who can speak well in interviews
and write decently well. Fight it all
you want of the types of people who
applied to YC at the time, I was within
the 1% you would think is most likely to
succeed. If you know the history of the
companies I worked at and you liked the
way I talked, it makes sense that people
would pick me, especially because our
revenue was going pretty fast at the
time. I understand why I would end up in
that top like 2,000 that get interviews
and the top 400 who get in. I would not
get into YC right now if I was the
person I was then. Theo from 2021 would
not get into YC now because it is
significantly harder to do so and I was
lucky to sneak in in the first place.
The bar is much higher. That is a good
thing if you like investing. That said,
there were problems. There's a lot of
stereotypes about VCs and also Y
Combinator in particular. Like if we're
being real, the vast majority of these
companies are basically just chatbt
rappers. Not that I would know anything
about a chat GPT rapper, but there is
some depth to this statement that is
important to understand. The capability
of AI is really cool, especially when
applied the right way. But there are
also problems. If you don't understand
the thing that you're trying to add AI
to well, you're going to do some stupid
things. Let's take our perspective here
as developers. I don't know about y'all,
but for me, AI first really felt useful
when I tried Co-Pilot for the first
time. I have a whole video of my
reaction to, "Oh my god, Co-Pilot's
actually really good. What the hell?" I
I had no idea it was going to be that
good, especially that early on back in
what, I think it was 2022.
Never would have expected what happened
there. C-Pilot was a really good thing
to have introduced us to AI because it
wasn't taking anything from us. Copilot
would be introduced to the editor that
many of us were already using as a thing
that just happens that tab completes as
you're writing. It's not taking away the
planning process. It's not taking away
the reading of the code. It's not taking
away the actual working in the codebase.
It's just autocompleting a little bit of
the typing as you write your code to
solve your problems. You couldn't even
tell it what to do beyond writing a
comment in your code and hoping it would
use that comment to do the right thing.
That was as far as you could go. But by
doing it that way, it never felt
threatening. It never felt like the AI
was trying to replace the things that we
do or the things that we like. And we
all got introduced in a way that felt
like it it matched what we wanted and
what we were doing. It made the boring
parts easier to do. It didn't get in the
way of the fun parts. But most
importantly, and I hope this isn't
controversial, Copilot was built by
developers for developers. Duh, right?
Like obviously devs built the thing.
Devs build things and obviously devs use
it. It's a dev tool. There's an
important thing here. And yes, it's
obvious. The arrows shouldn't be
necessary. Copilot was built by
developers for developers. Let's make up
a company. Let's say there's a company
that wants to make AI for flower shops.
We'll call it flower flowi. Flowy. We
have flowy flowers but we we make it
easier for flower shops to manage their
customers with AI. Now imagine this
flowi was built by devs for flower
shops. This type of thing was very very
common for us to see because co-pilot
went so well. It seems like adding AI to
things that are hard to do is inherently
pretty valuable. And as such, one of the
most common formats for startups that we
were seeing at Y Combinator was co-pilot
for X. And no, I do not mean X.com.
Please stop. I mean X as in a variable
for some random thing. Copilot for thing
was the whole brand of I think two or
three different Y combinator patches.
The problem was that a lot of the people
building this were developers, recent
college grads, people who love AI and
know a bunch about the cool stuff going
on with LLMs, people who know a lot
about infra and AWS and those types of
things, but they didn't necessarily know
about the thing that they were building
for like flower shops. Another
interesting one was a company that
founders were recent college grads, had
never worked real jobs in the industry
and they were trying to do co-pilot for
product management. So obviously they're
product managers and they saw some
problems they wanted to solve with AI,
right? Nope. They had never had a
product manager. They had never worked
at a company with a product manager.
They barely understood what one was.
They thought they had identified a hole
in the market and that they would fill
it. And this is a very common mistake
that Y Combinator companies make. But
more importantly, startups in general
make this all of the time. They see a
solution that works for them in a space
they care about and they try to apply it
in spaces that they do not understand at
all. Very common. The equivalent of this
would be imagine if the first time that
developer tooling was using AI instead
of it integrating into our tool, some
non-dev person, I don't know, some like
CEO of Salesforce or something that
doesn't know anything about code comes
in and says, "We're going to replace all
of your developers with AI. We built an
agentic tool that knows [ __ ] and Python
really well and is going to replace all
of your code bases with our new magical
AI. How many devs are going to try out
that [ __ ] thing? The answer is none.
Because nobody is trying to replace
themselves at their job. They're trying
to make the boring parts less boring,
the hard parts less hard, and the fun
parts more relevant in their day-to-day
lives. And that was what Copilot did
well. So obviously why combinator
companies had to adjust because too many
of them were making these types of
mistakes. And they did. They moved from
co-pilot for thing to a much much better
cursor for thing. That solves all the
problems, right? Because cursor is
definitely not the same thing or
possibly even worse because it takes
over more of the work and leaves you
less of the fun parts to do. That
wouldn't be that couldn't happen. The
place where I saw this pattern the most,
at least the place that affected me and
I cared about the most was in video. I
cannot tell you how many video AI
companies I have talked to that went
through my combinator that were people
who either didn't make content or didn't
like or care about content that decided
everyone would be a YouTuber or an
influencer. If only it was easier to
make videos. Editing is too hard. That's
the reason nobody is out here making
videos. If we made it easier, everyone
would be a creator.
Uh, the reason nobody's a software
engineer is because typing is too hard.
It's way too hard to use a keyboard. If
only we made keyboards easier, everyone
would be a developer, right? Do you see
how stupid that sounds? You know how
often I would hear pitches that sound
like that? I won't say it was the
majority of companies, but it definitely
felt like the bottom 30 to 40 percentile
was stuff like that. I cannot tell you
how many times I had the co-pilot for
thing, cursor for thing, AI assistant
for thing conversations where they
understood the AI part, the co-pilot
part really well, but they didn't
understand jack [ __ ] about the other
side. This is the thing that changed in
this batch. It was a change that was
already starting to brew due to the
nature of how much pickier they're
getting with their acceptance rates, but
also because I'm pretty sure YC noticed
the same trend. They were investing in
these companies that were doing AI for a
thing, but they didn't really understand
that thing very well. I understand why
seeing this trend, you would assume that
Y Combinator encourages this. They
actually do the exact opposite. Here is
Gary Tan's thoughts on this from before.
the the most important part of design, I
think, is actually the empathy for the
user. Like, you sort of have to be like
an ethnographer. Um, you're going in and
seeing, you know, what is this person
like? What are their goals? What are
their motivations? If you're selling
enterprise software, the most important
question is how do they get promoted?
Mhm.
>> And so if you can actually do all of
those things, load it into your head, uh
you know, whether it's having actually
lived that life or being able to be
around people all the time who uh live
that life, only then can you design
something that um is appreciably 10x
better than the thing that they were
doing before.
>> Kind of to that, I know IC's been at
least public about advising founders or
would be founders to like go work in a
call center, go, you know, do some of
this
Exactly. Go undercover to like develop
that empathy. Um, are folks taking you
up on that?
>> Yeah, totally. I mean, there was a
company recently, I mean, this is one of
the cool things about uh open-source
large language models. We had a medical
billing company where one of the
co-founders went and got a job as a
medical biller on Zoom
>> and uh, you know, they were able to
write software and write prompts and
like deeply understand what is it that a
medical biller does. they did it in a
way that um you know didn't violate
anything because none of the data it was
it's sort of like
>> uh they were just doing the job and
instead of you know them using their
wetwware to do the job they were writing
software locally using uh you know I
think the latest version of meta's llama
and so I think that that was an
incredible example obviously the best
way is to deeply understand someone's
problem or a space and go in and um you
know totally above board, understand
things. But sometimes you might not need
to do that. You could go undercover and
just do that job directly and then
deeply understand it.
>> I love this. It's really hard to improve
a job that you do not understand.
>> Building software for users that you do
not get, that you do not empathize with,
that are not you is really, really hard.
Y Combinator's whole motto has always
been build something people want. That
is much easier to do if you want the
thing. But if you don't, you don't
understand it should at least go get a
job in the space so that you can
understand how others do it. And they've
been encouraging this for a while
because they're actually very tired of
this of co-pilot for thing. Meaning they
don't understand thing. They only
understand AI. And this is how the shift
has happened. If we were to rank AI
startups in Y Combinator by how well
they know AI versus how well they know
thing and we were to do this batch for
batch recent batches have overall looked
kind of like this. They know AI pretty
damn well but thing which I would argue
is the important part the thing they're
actually building for they don't
understand that well and I hate this. I
go out of my way to not invest in
companies that look like this, where
they really deeply understand AI. They
can go toe-to-toe with people like me
about every detail about the new models,
what they do, what they can't do, all
the cool tooling and things that happen
in the AI world. If this is what your
spread looks like when you're building
AI for thing, and you know thing too
poorly and AI really well, I'm out.
You've lost my interest. I'm not doing
it. And this was very common for a
while. But a shift has happened. This
batch felt fundamentally different. It
now feels like this. Multiple companies
in this batch admitted that they were
actually a bit shy about AI. They felt
like they were behind and they didn't
really understand it. They'd only
started getting into it for real a few
months before. That sounds like a red
flag until you realize that I also
wasn't into AI until late December of
last year. And by January of this year,
I had one of the most successful AI apps
in the space. And by March, my research
was being cited in various things. I
could spend a much longer time ranting
about how most of the AI influencers and
sources of information are absolute hot
[ __ ] garbage. I'm not exceptionally
talented for figuring it out in 2 to 3
months. Everybody else just kind of
sucked. So being early to AI and being
really deep on AI is often a
disadvantage because things change
really fast. Things aren't necessarily
relevant to your space. And most
importantly, as the AI tools get better,
your need to understand the AI well
actually goes down. But your need to
understand the business you're building
in only ever goes up. And that's how we
end up on Nautilus, one of my favorite
companies from the recent batch that I
just can't really shut up about because
I find what they're doing so genuinely
cool and interesting. Nautilus is a car
wash platform. They're kind of like
QuickBooks. They're kind of like Square
Pay. They're kind of like all the things
you need to run a small store anywhere
in the world, but specific for car
washes. That sounds really dumb, right?
Like, why wouldn't they just use the
existing solutions that are generic and
probably cheaper and work well for
everyone? The reason is because car
washes have specific needs and also
specific opportunities to make more
money. Things like memberships. How many
people in my Twitch chat have a
membership to a car wash? One if you do
and two if you don't. That's a lot of
people who drive that don't have
subscriptions to their car wash. There's
been about three so far that do.
That's a huge opportunity. That's a way
for these car washes to go from just
barely profitable to potentially quite
profitable. And if this is software that
can be sold to a car wash, so that they
need fewer employees, they need fewer
subscriptions, and they have a
membership program that will
automatically send a text out saying,
"Hey, you're overdue a car wash." And
things like that to make it more likely
that the car wash can make more money.
This is worth a ton of money to all of
those car washes.
So, how did they make this? How do they
know all of this? The founder has been
working at car washes since he was 16
years old. He integrated all of these
things as the cashier at a car wash when
he was 17 years old. He ran into so many
problems with the software and saw so
many opportunities in the space that he
started building Nautilus to make that
car wash and others nearby even easier
to manage. And he succeeded. And then YC
let in, as far as I know, their first
ever car wash company. And this kid
shows up in a [ __ ] sailor outfit and
hat and pitches his car wash in a sea of
AI dev tools. It was magical. It was
powerful. It was exciting. And then he
comes up to me at the end and says, "Oh
my god, Theo, your videos helped me so
much when I was trying to level up and
like build the thing I built." And then
I heard what they were doing. And then I
heard how much money they're making. And
I invested immediately without any
further questions because these guys
know car washes really well. They didn't
know the AI and developer side that
well, but they found the resources to
figure it out, people like me. And they
combined their their knowledge of car
washes. And I was chatting with them at
one point with like other people. And
somebody asked like, "How do you get all
these car washes to sign up?" Because
they have like a few thousand, I think,
that are already using their stuff. and
a ton of like big brands are moving all
of their like actual shops over. They're
doing very very well with it. So,
somebody asked, "How do you actually get
these companies to adopt your car wash
software?" They said that one of their
biggest leads is their regular
appearances on wash talk on carwash.com
the podcast as well as the thing that
they had on their website, the webinars
that they do, as well as the car wash
events and conferences they go to. There
is a whole world of car wash content, of
car wash events, of car wash CRM and
webinars and all of this [ __ ] that none
of us know a [ __ ] thing about. We are
talking about a different world here.
And if you look at my chat's reaction to
all of this, OMG
or OMFG, uh, that's so cool. On today's
episode of the Car Wash Podcast,
you get it. It's kind of nuts, but
that's why these guys are going to win.
They're not going to win because they're
the industry leading experts in AI.
They're not going to win because they're
first. They're actually quite late to
the space. They're not going to win by
making a solution that literally every
single company in the world can use.
They're going to win by making it so a
car wash can make two times more money.
Being the best solution for them by far
and the only one they even know about,
much less consider. And now they can
charge 10 times more than QuickBooks
does because it doubles their revenue.
And despite having 100th the customers
that QuickBooks has, they'll be able to
have one their [ __ ] revenue. That is
crazy. People underestimate how many
weird niche subfields of sub fields
there are out there. Absolutely. But
there was a problem with this before.
The problem before was that a sub field
would need the same amount of
engineering effort as the greater higher
level field. Didn't make sense to build
something like Verscell before AWS. It
didn't make sense for a while after
either because the number of engineers
it would take to build something like
AWS was too high. things like AWS,
things like Verscell, things like AI,
things like React. One of the effects of
all of these tools is they allow for a
smaller number of people to build a
similar piece of software. The number of
engineers it takes to build and maintain
big applications goes down every single
year. The number of devs these companies
have go up because the number of things
they're doing goes up. But if you want
to take a product that exists today and
rebuild it next year, it'll be easier.
If you take a product that was built 10
years ago and rebuild it now, it'll take
you 150th the engineers to do a pretty
good job. And that's the bet that a lot
of these startups are making. This is
actually a good question. I think I got
a bit disconnected. A quick Google
search says the car wash industry is 24
billion. They may know car washes really
well, but it's a billion dollar company.
Absolutely. You don't need to make a
billion a year to make be a billion
dollar company. You need to have good
profit on like 150 mil a year. So 150
mil a year divided by 24 billion. They
only need to capture 6% of the total
spend in the industry in order to be a
billion dollar company. And if they
actually can increase the amount of
money the industry is worth because all
of their customers are able to double
their average revenue per customer, this
might go from a $ 24 billion industry to
a $40 billion industry and they can
capture a significant portion of that
revenue. Why such a high multiple?
because they are probably printing
profit off that. Like what are the costs
for a business like what they are doing?
They don't have very many employees.
They don't have very much
infrastructure. They don't have very
high costs. If a customer pays them 500
bucks a month, they're probably spending
$15 in that month. I don't know any of
these numbers. I haven't talked with any
of them about any of these things. It's
just my understanding of how businesses
like this operate. Their cost margins
are going to be crazy. Their profit's
going to be insane. Their ability to win
a significant portion of the industry is
absurd. And if they decide to, they can
increase the amount they charge and
nobody can say no because they're the
best solution by far. They are in an
incredibly powerful position right now.
Another similar example is co-create. As
I mentioned before, there have been a
lot of companies doing AI video tools to
make it easier for people who don't know
a thing about video to make a crappy Tik
Tok that gets five views. Those products
have no place in this world and they
will all die aggressively and quickly.
They piss me off a lot. I cannot tell
you how many times I've had to sit in
the YC office with a random video
company trying to get me to invest in
them where I had to explain, "You guys
do not know a single [ __ ] thing about
video. Your business is fundamentally
doomed." The big thing I had to explain
is that the things that make editing
slow have very little to do with opening
the editor, moving files around, and
cutting. When Faze, my editor, gets
access to our assets, he can edit an
hourong video in like 40 to 50 minutes.
Hi, FaZe. Faze, if somebody was to tell
you that they would replace Final Cut
with an AI that auto chops the video for
you, what would your honest reaction be?
How much fixing am I doing after?
Probably a lot. And now you have to use
their jank Electron app that doesn't
have any of the things you need that
crashes when you put 4K videos in.
You're in for a bad time. Now imagine
somebody tells you, "Hey, we're building
a solution because we were video editors
for years and we found that managing our
assets and synchronizing them was
obnoxious."
Start to make a little bit more sense.
This is the first video company I have
talked to in the startup world, period,
that mentions things like time code on
their website. They have string outs.
They'll assemble a timeline based on the
content in natural language, creating a
string out in daily for the daily
recording session you did to see what's
worth taking or not taking to trivialize
the process. Export it to your
professional editing software of choice.
Running all of this locally so none of
your content is being sent to a random
cloud somewhere with AVYNC again with
time code and things like that all built
in automated organization and management
for these giant recording processes that
have thousands of hours of footage. And
as my editor has said, this is great for
AI grouping and stuff. Sounds like
awesome for pros, for people in real
productions. Absolutely. We had to go
the opposite direction on my channel
where those types of things were so
annoying that I built my whole process
with my markers, my external system, my
breaking things up via CSVs and then
dumping them on my Dropbox so that my
editor can access them. We built that
whole system because all of the
orchestration of your content is the
annoying part. takes us more time to
make sure the right files are tagged in
the right place than it takes for us to
actually edit the video. And you
wouldn't know this if you weren't a
video editor and you didn't spend a lot
of time talking to actual video
creators. This is so common. And it was
so nice to sit with a YC company that
was going through the batch doing video
product that literally had just gotten
back from New York because they flew out
and sat on the floor of a big edit
warehouse with like 15 full-timers just
doing pre-editing, orchestration, like
assistant editors just making sure the
right assets were in the right place and
making all of their jobs 2x easier. This
is a real video company. This isn't an
aspiring creator company. I am so tired
of the aspiring businesses. We want to
make it so everyone can program because
programming is too hard. We want to make
it so everyone can be an influencer
because editing is too hard for
influencers to do. We want to make it so
anyone can make a viral post because
coming up with posts is too hard. All of
these things are stupid and none of
these things will work. And the only way
you can think this way about any
industry is if you are here, if you are
done in Krugerging yourself because you
tried to be an influencer, you tried to
be a developer. You found it too hard,
so you gave up and then decided you
could make a business selling the
solution to other people who gave up.
The harsh reality is the people who give
up are not good customers. The people
who give up are not good to build
around. If your business's average
customer is an aspiring thing, not the
thing, you're doomed. And this is what I
realized when I built my Y Combinator
startup, Ping. Even though we were
focused on building for professionals,
the ping.gg GG video product very
quickly became used primarily by small
hobbyists. People who aspired to be a
successful creator. They couldn't figure
out why nobody cared about their stuff.
So their solution to solve this problem
was to use professional expensive tools
like ping in hopes that if their quality
of video went up, they'd go from five
viewers to 50. Aspiring spaces are evil.
They are full of what we call tarpets.
All of these things sound like a good
idea until you understand the space at
all or you think about it a little bit
more deeply. And this trend has existed
in most startup spaces, but in
particular, I have seen this a lot in YC
startups.
Not in this batch. Many companies looked
like this. One more fun example that you
devs can probably relate to. We've all
seen the AI vibe coding apps. Most of
them are built on top of React Native
because React is something that's easy
for AIs to write. So, React Native with
enough rappers means you can vibe code a
mobile app relatively easily. There's a
bunch of these. Disclosure, I'm invested
in most of them. So, if I put a name
down, assume I am at this point for the
vibe coding stuff. There's ro.app,
there's a zero.app,
there's vibecoder,
there's others I'm forgetting for mobile
apps specifically, but these were and I
think still are the big three. All of
these companies were built by people who
were pretty nerdy about mobile, like A
in particular. Those devs were two
friends in high school that made dozens
of apps together and they built a zero
to make it easier to build apps. But now
there's an interesting one. Bit rig.
Build apps for your phone on your phone.
You might notice how basic this website
is. The reason is because unlike all of
the other vibe app companies, these guys
do not understand web for [ __ ] And
that's okay because they're not building
a web company. They're not building an
AI company. They're building an app
builder company. And ready for some very
fun facts about Bitrigg. Bitriig is the
only one of these AI app companies that
I know of that isn't building apps with
React Native. Bit Rig is using Swift UI
because they think that Vivecoded Apps
should meet the native bar that they
care a lot about. But what do they know
about Swift UI? What do they know about
native apps? Why are they coming in?
Because this is just AI for that, right?
Ready for the fun spoiler? The creators,
I [ __ ] you not, are the creators of
Swift UI. The people who made SwiftUI
left Apple and built a vibe coding
platform to make Swift UI apps. Imagine
competing with the creators of Swift UI
in vibe coding, building native apps.
They admitted to me when we chatted that
they did not really feel like they
understood marketing or AI stuff very
well going in and they were doing their
best to learn it. They asked awesome
questions in particular about branding
and marketing stuff. I immediately
invested as much as they would take from
me. And now they're getting a big moment
as I mentioned them here. If you want to
build native mobile apps and you're
interested in real native apps by people
who understand the platform better than
anyone else in the godamn world, Bitrig
are those people. It's worth a shot. And
this is also fun for me because I
personally honestly think React Native
is more than good enough. But if I am
wrong and all three of these investments
fail because React Native is not good
enough for vibe coding a mobile app, now
I have a hedge bet with Bit Rigg that
will cover the difference. If I am
lucky, one to four of these four
companies will end up giving me a 10 to
100x return in approximately 8 to 10
years. I have nothing to gain at this
point in time beyond thinking these
things are cool. I'm not promoting these
companies because I invested in them. I
am talking about them because I want you
to understand what this vibe shift was
and why I'm interested in them. And
really fun question from chat. What if
flutter wins? Really good question. I
actually also had invested in two
separate Flutter companies. Of the many
companies I've invested in, only three
have gone to zero. And of the three that
have actually went bankrupt and I was
able to claim the losses on my taxes,
two of the three were the two Flutter
companies.
So yeah, Flutter will not be the winner
because Flutter can't even get out of
early stage. I do invest in things I
don't agree with because I want to hedge
my bets. If there is a company that is
doing a thing that I think is obviously
wrong and most of my portfolio is built
against, I will invest on the off chance
they are right and I am wrong. Actually,
we have to touch on one of my favorite
things about this batch, too. As I've
been saying, the AI for thing companies
know a lot about thing and don't know a
lot about AI. You would imagine that
these companies probably aren't the most
experienced developers if they're
building a car wash company. The really
interesting trend I saw was for the
first time in any batch, all of these AI
for thing companies were really hyped to
have me there. Nobody outside of SF
knows who I am. But when I go to demo
day, I get a decent number of photo
requests. People who come up asking
like, "Oh my god, you're Theo. It's so
cool to see you here. You don't take a
photo?" That's awesome. I love that. If
you see me in public and you want to do
something like that, I'm cool. It's cool
to know people know who I am and they
resonate with my content. Usually the
people who ask for these things would be
the founders of the dev tool companies.
Like the people who built a zero
obviously knew who I was and were hyped
to see an email from me. We chatted.
They're now good friends. The developer
tool companies in the developer space
should know who I am because like let's
be real, I have the numbers here.
According to last state of AI developer
survey, 54% of respondents watch my
videos. Do you want to be in the 46%
that don't keep up? If your job is to
sell to those people, and also let's be
real, of the developers that use AI that
make it into Y Combinator, are they more
likely to be on the 54% or the 46%.
Realistically speaking, they're more
likely to be locked in. This is how it
is. A lot of the people in the recent
batches have said how useful my startup
and startup adjacent videos have been,
but also how useful my general dev stuff
and AI coverage videos have been for
them. So, a lot of these AI for thing
companies wouldn't be my expected
demographic. I'm expecting the dev tools
to be hyped on me and for these guys to
maybe sometimes know who I am. Almost
every single one of these companies was
really hyped about me. I was shocked.
Like, of the people who came up to me,
the vast majority were working on [ __ ] I
didn't know anything about. Like
healthcare, like mapping software, like
missiles, like car washes. I don't know
[ __ ] about any of that. But they were
coming up to me because they were hyped
that I was there. And that is so cool
and heartwarming and awesome. But it was
counteracted by a very weird change.
Most of the dev tool companies had no
[ __ ] idea who I was. I do not know
what happened here. I am talking about
businesses that are literally building
tools for cursor where none of the
founders have ever heard my name. That's
not even the funniest one. The funniest
one by far was a kind of vibe coding
enterprisey one that's building
developer tools to help businesses
integrate custom software faster. And I
was skeptical. I didn't hear many of the
things I would expect to hear. It was a
recent pivot. The revenue wasn't great.
And they were specifically targeting
things like Superbase and other like
data platforms that were harder to vibe
code against. And I was like, okay, this
is a bit red flaggy for me. You don't
know who I am. you're talking about
Superbase a lot. I don't see the space
here. Like how are you managing data if
you're doing this? I had already like
written this one off like I'm not
investing in this one. I'm just curious
how this founder is thinking about this.
I [ __ ] you not. They reply, "Oh, we
built it all on convex. They're the only
backend platform we have found that does
really well with vibe coding. Have you
seen their thing, Chef? You know, Chef,
the vibe coder on the Codex website.
That was my idea." They didn't know who
I was and they literally built their
startup around a thing that only exists
because of me. I have no idea what the
[ __ ] happened here. I still invested
because I thought that was pretty cool.
That all said, the majority of the dev
tool companies I talked to had literally
no idea whatsoever who I was. Maybe
three of them did and at least five of
them didn't. That was very interesting
to me. What this suggests is that the
dev tool companies are not as locked in
as they were before. that they are
trying to find edges in the industry and
they had a good enough pitch to sneak
through that. And my hotter take here,
and this one might get me in a little
bit of trouble, but uh I can't just
glaze the whole time, right? Why
Combinator puts out a request for
startups right before every batch? The
partners who are all founders, by the
way, YC doesn't have investors that work
there. Everybody there has built a
company. So these founders now partners
will describe things that they want to
see more of in the batch in hopes that
people will apply with ideas that are
similar to this. Most of these people
are what I would call exdevelopers.
I am a part-time developer who will
eventually be an ex-developer. I just
cannot justify writing code for the
majority of my time nowadays, especially
with one hand. I miss it dearly. I'll
get back to it soon. But most of the Y
cominator partners have been writing
code for some amount of their career and
they all notice this trend where we need
the companies that really understand a
field well. So let's go to Nautilus
again for a sec. If you don't know jack
[ __ ] about car washes and this you have
two potential investments. One is a
company that knows AI really well that
doesn't know car washes that well and
they say a bunch of things you don't
really know if they're true or not. And
the other option, we're on the
carwash.com podcast. We help run
multiple webinars about car washes. I've
been working at them since I was 16 and
here's all the things I've done. One of
those is obviously the good investment.
And it's very easy to know that without
having any experience in the space. My
spicy take, and this hurts, and I'm
sorry to the partners that feel like
this is a call out. It is not meant to
be. It is more an observation than a
call out. I am sorry. There's a Dunning
Krueger thing here. If you know nothing
about car washes, you pick the person
who knows about car washes. If you know
a little about car washes, you pick the
person who you think based on their
knowledge of whatever you're interested
in, you think has the most capability of
success. Once you know enough more at
the end of the Dunning Kruger, you
realize the person who knows more is
much more likely to succeed than me. You
pick them again. So if we draw this out
as a Dunning Krueger curve where as you
start learning the thing, you get really
confident. As you get further, you
realize you're clueless. And then you
slowly build up real knowledge. If your
current understanding of car washes is
here, you pick the person who knows what
they're doing. If your current
understanding of car washes is here, you
pick the person who knows what they're
doing. Your understanding is here.
Doesn't matter. You're going to pick the
right person anyways. But what if your
understanding is here? If you know a
decent bit about car washes, but you
think you know everything about them,
this is when you start making edgy bets.
If the person that the car wash company
talked to knew a bit about car washes,
they might have made dumb decisions.
Replace the word car wash with software.
Now imagine you don't know jack [ __ ]
about coding. And you talk to somebody
who's really deep in the software
development space. They have years of
expertise. They've been working in open
source forever. They are involved in the
open source communities that I and
people like me care about. They watch
all my videos. They're deep in the
space. and now they're talking to a YC
partner who doesn't know jack [ __ ] about
software development. If this person
seems deep and excited about the thing
you don't know jack [ __ ] about, good
chance they know what they're doing. If
you've been writing software for a
while, but haven't in a long time, and
you know what it took to keep up, you
know how important it was to be locked
in, to keep up with what's going on in
the industry, you know how the trends
shift and change and all of that. You
pick the person who deeply knows the
thing. Let's say you recently got into
vibe coding after not coding for 10
years. And you're like, "Oh my god, I
had no idea coding was so easy. Being a
good coder doesn't matter that much
anymore. Anyone can just vibe code their
way through Cursor and Windsurf." By the
way, Windinsurf and Cursor basically the
same thing. Windsurf's a little cheaper.
You should probably use that. Those
types of people will pick the startup
that they think has the most likelihood
of being worth billions, not the one
that understands the space the best. My
honest belief is that the recent surge
of YC developer tools that don't know
jack [ __ ] about the developer space is
because there are partners at YC that
are currently here in the Dunning
Krueger curve of how software
development works. And they are saying
stupid things, attracting stupid
founders that bark back the same stupid
beliefs they have that result in
mediocre at best and destructive at
worst developer tool companies ending up
in the batch. And I am not saying this
is all of them. I would say this is at
worst half of them, but there are more
of these clueless developer tools
companies than there are clueless
non-developer tools companies altogether
combined. So, it's funny that we went
from the AI dev tool companies
understanding dev well cuz you had to
understand dev to do anything to the AI
dev tool companies being the only AI for
X companies in the batch that don't
actually understand the developer space
that well. Of the 18 investments or so I
did in this batch, I'm pretty sure I hit
two developer tool companies total.
Depends on if you count Bitriig or not.
Their audience isn't devs, but you get
the idea. Two to three of my 18
investments were dev tools. In previous
batches, it's over half. It's also a
smaller number. It's usually like 8 to
10. So, this is a really good batch of
people who were really locked in on the
thing that they are doing. That is a
huge industry shift and I genuinely
really hope it maintains because the
summer 25 batch is the best YC batch
I've ever seen as long as you don't look
at the dev tools too hard. Think that's
all I have to say about this one. What a
chaotic experience it has been talking
to dev tools who have no idea what we're
doing in the space and talking to car
wash companies that couldn't be more
hyped to talk with me about the dev
stuff that they're doing to make car
washes better. I did not make a lot of
money from this yet. In fact, it was a
very expensive week for me. Thank you to
my sponsors and thank you to all of you
for watching and making this possible.
But hopefully, if all goes well in 5 to
10 years, the sponsors for this channel
will be the companies I invested in
instead of the companies funding my
investments. Thank you again to everyone
for watching this. I hope it's helpful
for y'all to better understand how these
startups work and how I think about them
and most importantly, how I make my
investment decisions. I am not like many
other investors. So, don't assume that
what I talk about here is going to get
you an investment because the majority
of the investors I talk to don't know
jack [ __ ] about anything. I am trying to
explain this is this cool weird
in-between position that I happen to be
in and I hope it was useful to somebody.
I know this video would have been very
useful to me not very long ago. Let me
know what y'all think. Until next time,
peace nerds.