Is the AI bubble popping—or just catching its breath? Eric Newcomer and Tom Dotan spar over Nvidia jitters, Sam Altman’s “bubble” dinner, the MIT “95% fail” headline, app-vs-model margins (Cursor, Claude Code), and Chamath’s SPAC-as-casino shtick. Then Eric sits down with Vercel founder/CEO Guillermo Rauch for a fast, idea-dense jam: assistants → agents → multi-agent teams, why GPT-5’s real story is coding, “vibe coding” and code-last workflows, who gets paid in the era of AI factory-builders, whether to study CS, why taste beats code, and Guillermo’s six-month prediction for a breakout vertical agent.
00:00 Did the AI “bubble” pop? Altman dinner & sell-off vibes
01:16 MIT survey “95%” headline vs reality
09:04 Capitalism, incentives & Chamath’s SPAC “casino”
18:17 Interview starts — Guillermo Rauch (Vercel)
22:07 GPT-5 reality check & the “Einstein-in-a-box” test
37:37 Future of engineering + should you study CS?
48:36 6-month prediction: a breakout vertical agent; underestimating GPT-50
00:00:00
I think we need to break down the logic of why this week
00:00:03
people decided that the AI bubble has burst.
00:00:05
And there were some key things that happened that we, I think
00:00:07
we can deconstruct and decide whether or not they're actually
00:00:10
worth, you know, being a hater over.
00:00:12
The first one was Sam Altman had a dinner that I was not invited
00:00:15
to with reporters where he discussed, among other things,
00:00:19
that it was all on the record, you know, GPT 5 and some of the
00:00:22
disappointments around that, which are very real.
00:00:25
And also at some point said something to the effect of the
00:00:27
quote in front of me basically like, yeah, I think we're
00:00:29
probably there's some exuberance.
00:00:30
So we're in a bubble right now where some stocks are are very
00:00:33
overvalued because of AI. But then in the next sentence
00:00:36
also was like but also AI is transformative.
00:00:39
It's a classic power move. Elon does this too.
00:00:41
It's like you're in the Super hypey thing.
00:00:43
There's almost nothing to lose by being like, yeah, I don't
00:00:45
know, People are really think I'm hot shit.
00:00:47
Maybe too much, you know? Yeah, yeah, yeah.
00:00:50
My, my, my lovers love me too much.
00:00:51
My haters don't get it. But if there's something wrong,
00:00:54
it's because of something other people did, right.
00:00:57
I just want to disregard almost everything Sam says these days.
00:01:00
I mean, it's just he's on. He says so much and it's so
00:01:04
contradictory that it just doesn't even even though it
00:01:07
probably did play some part in the sell off of AI, you know, AI
00:01:11
driven stocks, the idea that oh, Sam Altman has declared a
00:01:14
bubble, so we must be near some sort of a top.
00:01:16
And then there was this MIT survey that came out and I think
00:01:19
that probably had the most effect on a lot of people.
00:01:22
Which headline number was that 95% of enterprise AI, generative
00:01:27
AI projects fail just have 0 impact on a company's efficiency
00:01:31
and bottom line. And that is it is a bad number.
00:01:35
The problem, of course, is that a headline number versus what
00:01:38
was actually in the survey is like pretty different.
00:01:41
And this was not like a a survey that was put out by a fully
00:01:45
disinterested organization like it's effectively a pro AI.
00:01:49
Right, it's. Organization.
00:01:50
It's always funny when somebody like runs away with one line
00:01:54
that you said, but then like you're like, oh, but I'm sort of
00:01:58
trying to make a pro AI point here.
00:02:01
Yeah, yeah. Like it's almost like a rat fuck
00:02:04
in that sense. Like if you're like one of these
00:02:05
people who decides to buy the headline and then you actually
00:02:08
like distribute the survey and then anyone actually reads it
00:02:10
sees that, Like, the key points that it makes is that AI is not
00:02:14
incapable of transforming a business.
00:02:17
Just the implementation of AI and its inability to learn and
00:02:21
adapt to people's workflow is what's causing it to be
00:02:24
unsuccessful. My take away from actually
00:02:27
reading through this thing quickly, quickly was that like
00:02:29
the 5% of companies that have been successful with it are like
00:02:32
going to win. Like it's really about like the
00:02:34
fact that they implemented it better and were able to use this
00:02:37
technology is why like it is actually a transformative tech.
00:02:40
And like one of the co-authors of it is a product manager at
00:02:43
Microsoft. I was just looking him up.
00:02:45
So like these people that are writing this have a they have
00:02:48
like a dog in this fight. And it's not like long term
00:02:51
it's. Going to be a good flow of the
00:02:52
argument is basically everybody sees the potential of Chachi BT
00:02:57
and Claude. They try it out, but then they
00:02:59
almost let and they just don't do enough to actually get their
00:03:02
data integrated into whatever AI tools they're bringing to their
00:03:06
companies. And so then it ends up not being
00:03:08
more useful than just like consumer grade stuff and like
00:03:11
here's what you actually need to do to be better than just like
00:03:15
using Chachi BT. Right.
00:03:16
Like that was the key point. Here's what you need to be doing
00:03:19
to make this successful. And that's going to end up
00:03:21
probably being the talking point from all of the sales people at
00:03:25
the cloud software companies that are selling this stuff is
00:03:28
like, oh, yeah, you read that report.
00:03:29
They made these mistakes. You won't make this mistake
00:03:31
because you're going to buy it and use it in this way.
00:03:33
So it ends up the whole thing was an op, I've decided.
00:03:36
I will say I texted a lot of investors this week just like,
00:03:40
what's your read on the bubble like, you know, people who are
00:03:44
like, you know, investing in these companies often.
00:03:46
I'm interested in what, you know, yeah, people who are
00:03:49
trying to make money like, you know, it's like if it's an early
00:03:53
bubble, it's like, oh, I need to keep going.
00:03:54
But you know, they're they're trying to time the bubble for
00:03:56
financial reasons. And I did get some, you know,
00:03:58
somebody who was like full bubble.
00:04:00
Now trying to decide whether this is the start of 2021 or
00:04:03
late 2021. Obviously, everything unwinds
00:04:06
from the pandemic at the end of 2021.
00:04:08
There's still upside to be seen at the beginning and AS.
00:04:12
Is the person selling? They are.
00:04:15
I am in tread lightly mode right now.
00:04:18
Ha ha. The somebody else was like, I'm
00:04:21
watching the Fed. You know, a lot of people are
00:04:23
like, I don't know, what do you think?
00:04:26
Yeah. I mean, so nobody's like, what
00:04:27
an absurd question. You know, Andreessen Horowitz,
00:04:30
who sort of dispositionally is like, why would you ever sort of
00:04:34
position yourself as a bear? Our job is to fucking be
00:04:37
professional. Bulls Just came out with a piece
00:04:39
that was, you know, knocking down basically, I thought Tom,
00:04:43
not trying to fight with your article about cursor.
00:04:45
I think, I think the sophisticated critical.
00:04:48
MBA so they just view me at the end of.
00:04:51
It the sophisticated critical take, I think as you know, we're
00:04:54
about to say tomorrow, it seems like foundation models revenue
00:04:57
is still strong, but you know, application companies, like all
00:05:01
the coding companies, you know, have pretty negative margins.
00:05:04
They're built on top of money losing businesses and themselves
00:05:07
lose money. How's that all going to work
00:05:09
out? And Andreessen and Co, well, you
00:05:10
know, I think it's, I assume it's Martine Casado and I think
00:05:14
Sarah Wang sort of, you know, pushed back against your take.
00:05:17
What do you think about that as somebody who is worried about
00:05:20
app company margins, I. Think the margins are less the
00:05:23
issue then like I think there's an interesting dynamic which I
00:05:27
tried to get to in that story, which is the relationship
00:05:29
between the model providers and the application companies.
00:05:32
And it's clear with cursors grossly you can growth that you
00:05:35
can build a huge user base on an interesting application, maybe
00:05:40
has some brand name loyalty to it.
00:05:42
But like especially with developers, like they will just
00:05:45
switch to the thing that provides them the most bang for
00:05:47
their buck. That's going to be the best
00:05:49
quality product. And like the growth of Claude
00:05:51
code over like 6 months, which I know wasn't impressive to Ed,
00:05:55
but but is to me, you know, getting to like a $400 million
00:05:58
ARR shows that like people will gladly switch to a new thing
00:06:01
that seems like it'll work well. And Anthropic has, I mean,
00:06:04
they're money losing too, but they're going to have better
00:06:06
margins than Cursor will because they're paying wholesale access
00:06:10
to the model versus the retail prices that that that cursor has
00:06:13
to pay. So I'm not incredibly bullish on
00:06:16
application layer companies in a world where model companies
00:06:19
build the same software. If they decide they don't need
00:06:21
to, that they really think they can make enough money through
00:06:24
their API business or like a consumer, a pure consumer play,
00:06:27
then maybe there's a world for application companies.
00:06:30
But like, I think it's a very in this particular sector, it's a
00:06:34
very narrow Moat. And I mean, I was talking to and
00:06:37
I. Sort of agree with the
00:06:38
Andreessen people. These companies are not idiots.
00:06:40
They're gonna figure out like some of the stuff, like the very
00:06:43
expensive stuff they're gonna offload to cheaper models.
00:06:46
I agree with you that like the big business question is just
00:06:49
like how sticky are these products?
00:06:50
I think that yeah, it's like how sticky are they?
00:06:53
I don't I and I don't, I don't have an answer, but I don't
00:06:55
think the bubble is gonna swing either way.
00:06:57
It's like, oh, they're not sticky.
00:06:58
They'll because the people are moving to another AI product.
00:07:02
You know, at the end of the day, coders want these coding tools.
00:07:06
The margins will be figured out. They're going to consume some AI
00:07:09
product, either foundation models or companies built on top
00:07:12
of it. Either way, it's good for the AI
00:07:14
hype train. Well, and I to defend the piece
00:07:17
at least like I wasn't. Yeah, No, no, I don't.
00:07:20
I don't just think too Andreessen because apparently
00:07:21
they wrote it at me. Look, I think you know that that
00:07:24
the cursor is not running out of money.
00:07:26
Like the the story was very clear that, you know, yeah,
00:07:28
negative gross margins are bad. So they have like a billion
00:07:31
dollars or something? And they're losing, you know,
00:07:33
apparently their burn is like in the single digit millions.
00:07:35
So it's really an issue of like, you have plenty of time to
00:07:38
figure these things out. And I mean, you mentioned Uber.
00:07:40
Like I covered Netflix during its most cash burning days where
00:07:44
I'll never not bring this up on the show.
00:07:45
Like I worked for an editor who like, made a career for a time
00:07:49
writing columns saying Netflix is a House of Cards.
00:07:52
Right. Has he written me a culpa?
00:07:54
No, we never do. We move on.
00:07:56
We're like the we're like the Uber haters and the AI.
00:07:58
Haters move on, you don't played it.
00:08:01
I was the factual I provided. Where's your music pulper about
00:08:04
Uber? Like, I was like the gun runner
00:08:06
of the Uber skepticism era, you know what I mean?
00:08:09
I was like giving all the haters, they're like ammunition.
00:08:12
They needed to like shit on Uber's like money losing.
00:08:14
But I was always like, I don't know.
00:08:16
And then I always got annoyed that the people who took my
00:08:19
stories and like, oh, they lose a lot of money did all these
00:08:21
like insane math formulas based on them.
00:08:23
And I'm like, there's no way. My numbers are like, so like
00:08:27
there's such impressionistic, you know, like Uber was giving
00:08:30
very like rough numbers to people when they were trying to
00:08:33
back out their economics out of it.
00:08:34
I was like, this is a lunacy. Yeah, well, the problem with the
00:08:37
hater argument there is it it relies on the investors who see
00:08:40
more than we ever get to really just being dumb, right?
00:08:44
I mean, it's the idea that, like, they don't understand it,
00:08:46
but we as the skeptic and the hater do.
00:08:47
And believe me, I don't like giving investors credit.
00:08:50
Like, it's not fun. It makes my job worse.
00:08:51
But like, they do kind of do it for a living.
00:08:54
Like that's the one thing they should be able to understand.
00:08:57
Right. Is like the basic economic We
00:08:59
should get haters on here too. I want to.
00:09:01
Get some more haters, right? No, no, we like the Ed episode 1
00:09:03
of the things the Youtubers. Somebody got mad at me for
00:09:05
saying do you believe in capitalism?
00:09:07
You know, like it was like some sort of shibboleth, but it was
00:09:10
it's sort of a it was a sincere question.
00:09:12
Like some of these people. I do want to understand like
00:09:15
what premises do we share? Because like, to me, part of
00:09:18
believing in capitalism is like investors are fairly rational.
00:09:22
They're making different bets. They have their own, you know,
00:09:24
it's just like I, I, it's useful to know how much people will
00:09:27
think like the whole capitalist thing doesn't ultimately direct
00:09:30
us towards likely outcomes. Like to me, markets are great
00:09:34
truth mechanisms and like we play a part in that by, you
00:09:37
know, point poking at things and interrogating them and then
00:09:41
people change their ideas and then they eventually move their
00:09:43
money. But when people have commentary
00:09:46
that acts like investors won't listen to a good idea and, like,
00:09:49
change their strategy if it's in their incentive, yeah, it feels
00:09:52
like. Well, I mean like market, the
00:09:54
truth mechanism is over time. I mean, like, the periods of
00:09:57
irrationality in the midst of a bubble can give people the wrong
00:10:01
impression of, like, the, you know, solidity of a business.
00:10:04
Like, I think that's what Ed was arguing.
00:10:06
It wasn't like capitalism is failing because these companies
00:10:09
exist and keep raising more money.
00:10:11
It's just like it hasn't reached the point yet where like, you
00:10:14
know, whatever the Warren Buffett line about, like the
00:10:15
shore, the shoreline going out and, you know, finding out who's
00:10:18
naked. Right, right.
00:10:20
And I mean, one thing we really believe in and I think smart
00:10:23
business journalists are like there are people who can be
00:10:26
behaving according to their incentives and driving us off a
00:10:29
Cliff, right? I mean, this is sort of the like
00:10:31
Tiger Globals and the soft banks.
00:10:33
Like often I've made the point in the newsletter that it makes
00:10:36
sense career wise to be the like biggest bull of the biggest
00:10:41
mania, right? You're like, oh, at least they
00:10:43
were the most important person, you know, like in that moment,
00:10:46
like the Tesla longs and stuff, like what's her name?
00:10:49
Kathy Wood? Like, even if you have a
00:10:51
terrible performance, you sort of like win fans just for being
00:10:54
like really extremely bullish. And so even if you like believe
00:10:59
in capitalism like I do, certainly I think they're all
00:11:01
these actors who are like, you know, almost like taking
00:11:04
advantage of, you know, human psychology rather than, you
00:11:07
know, trying to make the most money they can.
00:11:09
They're just, they're like, oh, I'm going to be able to, if I'm
00:11:12
the best bull in the world, I'll be able to raise the most money
00:11:14
again in the next bull run in my game is just to like run as hard
00:11:17
at the bull market as possible every time.
00:11:20
And you know, like that's a, that's a game people play with
00:11:23
with great success. Speaking of taking advantage of
00:11:25
human psychology, A Chamath is back out there again.
00:11:30
Welcome back. Scam in the arena?
00:11:32
Scamming the arena's back? No crying at the casino Chamath.
00:11:35
Right. I know what a what a he's so
00:11:37
good at it. He's so good at it.
00:11:38
He's like, like, he's basically calling himself like the casino.
00:11:42
And it's like, sure. The casino the the the random
00:11:45
outcome generator I understand. Don't.
00:11:47
He's literally like, don't invest in me if you're not
00:11:49
willing to lose all your money. And if it's like trivial to you,
00:11:52
like, yeah, who invests like that even if you're rich?
00:11:54
No rich person wants to lose their money either.
00:11:56
Like everybody is investing to get a return.
00:11:59
Like, I know it's just like a bullshit thing you say, but.
00:12:01
This is the unique twist of the Trump era, right?
00:12:03
Is that like he has discovered there is a whole cohort of
00:12:07
Americans who like being scammed, who like.
00:12:10
Well, that's true. I mean, sports betting is the
00:12:12
best thing in the world. Like yeah, exactly Like, yeah.
00:12:14
I lost a lot of money on sports betting over the weekend.
00:12:17
Yeah, Yeah. God, yeah.
00:12:19
That was just fun. It was.
00:12:20
It was. Just for the fun of it.
00:12:22
Was just for the fun of it. Yeah, there.
00:12:23
I just went down to the sports book and I put $50 on a on a
00:12:27
horse called Duck Stuck go and at.
00:12:28
Least it was in person. You got some, you got an
00:12:31
experience out of it. My my issue with the sports
00:12:33
betting apps is like, what are you even getting?
00:12:35
You know, like. I think it's really mistakes of
00:12:37
watching a game. I mean, if you're not watching
00:12:39
the game, you're putting money on it, then that's a real,
00:12:41
that's real sick, right? I did invest also.
00:12:43
I put a bet on Australian rules football because I thought the
00:12:46
spread was so ridiculous, but turns out it was not that team.
00:12:49
That team lost by like 150 points, which I don't know what
00:12:52
that actually means. Anyway, Chamath is yeah, I think
00:12:54
he absolutely represents the peak of Trump era.
00:12:57
We like to be scammed mentality where he is actively calling
00:13:01
what he's doing a casino that there will be crying and you
00:13:04
will not make any money. I will make money from it.
00:13:06
He made money last time. I mean you like wrote I thought
00:13:09
a great piece about like his scam the last time through.
00:13:12
Like what are you expecting this time around 2:00?
00:13:14
Pieces on Schmoth. I mean, one thing that makes me
00:13:17
so sad about the world we live in is just how much your
00:13:20
Internet brand and your real person brand can be so far
00:13:24
apart. Like real people in Silicon
00:13:26
Valley think he's like a huckster.
00:13:28
Like people who work with him, it ends poorly for them.
00:13:30
Many of his professional relationships have gone poorly.
00:13:34
He didn't make people money on the SPAC's.
00:13:36
He made money through fees. Like he's looking out for
00:13:39
himself. And to some degree, you still
00:13:42
want somebody who's like looking out, who cares about caring
00:13:45
about your reputation. He's good because it means that
00:13:48
the part people you partner with and work for and invest their
00:13:52
money for you, you care about your long term reputation.
00:13:55
You want to be good by them. And this is someone who hasn't
00:13:58
minded his long term reputation. It hasn't, you know.
00:14:01
It's not a reputation as an investor, but if you look at the
00:14:03
way he positions himself, he wears his like, I don't know the
00:14:07
fucking brand name suits. He loves to be rich, yeah.
00:14:10
Yeah, And so, like, again, and not to make everything just
00:14:13
about Trump, but this is all part of the same thing.
00:14:15
People in the real estate world knew that he was an absolute dud
00:14:18
as an investor. Like all of his projects would
00:14:21
go bankrupt. He was a brand name to to, you
00:14:23
know, add a modicum of flash on top of an otherwise shitty
00:14:26
project. Here you're like, you're like,
00:14:29
bring back the Wasps. Where were the rich people who
00:14:32
cared about I? Mean what?
00:14:34
Old money. Just the people who.
00:14:35
Cared in their wealth. No bless oblige.
00:14:37
You know, we need, we need like people who are not.
00:14:40
We just need the Carnegie. I am watching Gilded Age right
00:14:42
now. So I have a very strong
00:14:44
attachment to to that era of, of, of wealth.
00:14:47
But yeah, those guys look tremendous compared to the
00:14:49
Chamats. I mean, they've done nothing or
00:14:51
he's done nothing, but he's built an an image that I'm sure
00:14:55
he will have no problem getting people signing up for his specs.
00:14:58
Well, it's that sort of fatalism that makes it possible, right?
00:15:00
It's just sort of like people are like, oh, he'll raise the
00:15:02
money. We should do it and.
00:15:03
Yeah. And I guess maybe the last point
00:15:05
to. Talk about I'm not cheering for
00:15:06
him. I don't know what to say.
00:15:07
I would be very sad if good people lose money on what are
00:15:11
certainly going to be failed IP, OS or whatever you call them.
00:15:14
Specs. I do think to be fair, you know,
00:15:17
we're business journalists. Like I do think he changed his
00:15:20
structure this time. I do think he's slightly more
00:15:22
aligned now with the actual outcome of the specs, though.
00:15:26
Knowing Chamoth, he's found some other way to make money as well.
00:15:29
But. Burdens.
00:15:31
I don't see why, but he's not scamming people, sorry.
00:15:34
This. Well, that you believe in
00:15:36
capitalism, don't you? There's infinite numbers of
00:15:38
shots on in a free model. Sure, yeah.
00:15:40
But I mean, this is always, this is one of these, I don't know,
00:15:43
to me psychological weaknesses the the elites.
00:15:46
Once you're in the club, you're like, oh, you're in the club,
00:15:48
you should get to do it again. It's like there there are tons
00:15:50
of human beings like try it with somebody who hasn't lost people
00:15:53
a lot of money. Like to me this sort of desire
00:15:56
to like, keep giving money back to the type of people, to people
00:15:58
who just like, gamble the money rather than saying maybe we
00:16:01
should try it with somebody new, that it means like a human
00:16:04
psychological weakness of just like loving fame rather than
00:16:08
success. It's never gone away.
00:16:10
I mean like Michael Milken in the 80's, the junk Bong King
00:16:13
still like runs a large financial conference in Los
00:16:16
Angeles, the Milken conference. Nothing, nothing tends to really
00:16:19
happen to these people for the long term.
00:16:20
I mean, some of them do get like banned from trading, but that's
00:16:23
not even close to what's going to.
00:16:24
Happen with, not anymore. Yeah, exactly.
00:16:26
That's it. Like I'm not accusing Chamath of
00:16:28
anything of that level. He just was an opportunist and
00:16:30
fucked over a lot of people who didn't.
00:16:34
We're looking at fights. So Chamoth come on the show.
00:16:36
We're I don't think he is. That would be a big get for lots
00:16:40
of reasons. Is probably the person who's
00:16:41
written the most negatively about him.
00:16:43
I'm not betting on that one. But if you're if you're a hater
00:16:46
positively negatively, we're we're ready to argue with you.
00:16:48
We like this arguing format. So we we want to have arguments.
00:16:52
Not quite an argument. I like him too much.
00:16:54
But after this we're going to have Guillermo the Guillermo
00:16:57
Roush over cell. He's coming on the show.
00:17:00
I think he had, I know a really good prediction about where
00:17:04
things are going. We talk about agents.
00:17:07
He obviously runs a no code company and I continue not to.
00:17:11
I have not built my no code startup empire.
00:17:14
If it's so easy to no code like where, where are the apps that
00:17:17
I'm? Using maybe you should start
00:17:17
getting into fights with lovable.
00:17:19
I know dude, I don't know if people know what you're talking
00:17:21
about. We don't yeah, that that can be
00:17:23
deep lore for newcomer fans. You can look up Eric's tweets
00:17:26
trying to get his. Money waste your time for I I
00:17:28
lost it on Sunday just because I I got charged twice from lovable
00:17:33
after trying to cancel. They, they have extreme dark
00:17:36
patterns that they say they're going to try to reform.
00:17:38
Literally you click a big red button that says cancel and then
00:17:42
you still are not cancelled from lovable.
00:17:44
They're like, oh, that was cancelling within lovable.
00:17:46
Then we had to send you to stripe and then you have to
00:17:48
obviously run through the stripe.
00:17:49
But anyway, they, they swear they're changing.
00:17:51
They're going to fix it. Maybe cancel in Swedish means
00:17:54
something else? Maybe it means like I'm still
00:17:56
down. I like the.
00:17:58
Company I would have come I it was just like I'm not using it
00:18:00
this second. I would have come back now.
00:18:02
I don't know. I'm pretty angry with them, but
00:18:04
hopefully they're. Reformed we're big supporters of
00:18:06
Europe tech over here on newcomers so we we long term
00:18:09
long term bulls on lovable and every European tech company all
00:18:13
right we should probably cut on over to your interview with
00:18:15
Guillermo all right I'm. Here with the founder and CEO,
00:18:20
Versal, Guillermo Roush. Welcome to the Newcomer Podcast.
00:18:24
Thanks for having me. Last time we were hanging out,
00:18:27
I, I, I feel like you convinced me that Purcell would be every
00:18:30
company in the world. So we'll see.
00:18:33
We'll see what you've brainwashed me into by the end
00:18:36
of this conversation. Where?
00:18:38
Where were we? I I think it was huminex if I'm.
00:18:42
I'm remembering, I think we. Had like lunch off that and just
00:18:44
like the the great founder visionary that I was like, oh,
00:18:48
what, what won't this company do?
00:18:51
You're you're obviously touching a lot in AI.
00:18:54
You know, like, you've got VO now with sort of your own vibe
00:18:57
coding product, but you're also sort of powering a lot of like
00:19:00
the vibe coding that everyone's talking about in sort of some of
00:19:03
the infrastructure to make it possible for people to sort of
00:19:09
spin up sort of websites and applications quickly using
00:19:12
artificial intelligence. But also, you're just like a fun
00:19:15
founder who's willing to sort of like, say what's going on and,
00:19:18
you know, not, not just let the venture capitalists dominate the
00:19:23
chattering among the Silicon Valley classes.
00:19:25
And terminally online so you can ask me about any of this
00:19:28
happened on X5 minutes ago and I'll know.
00:19:31
So I, I just want to start off with, yeah, like the existential
00:19:34
question about like the mood. I mean, I've been honestly been
00:19:37
texting everybody today about like sort of animal spirits or
00:19:41
like how's the, how's the market?
00:19:43
Like, you know, what's your read on sort of the AI mania?
00:19:46
Like, do you feel like there's still a lot of room to run here
00:19:50
or are you getting signs that there's like some some paranoia
00:19:54
about the euphoria? You mentioned Versailles is in a
00:19:58
very specific and special position that we see every,
00:20:03
almost everything that's being built on.
00:20:04
AI has to have some kind of user interface to the world.
00:20:08
And so traditionally, we've been supporting founders in building
00:20:12
this AI portals, right? AI applications now increasingly
00:20:16
agents. And, and one of the things that
00:20:18
I've been noticing is that there's basically 3 chapters of
00:20:21
AI so far. Number one was assistance #2 and
00:20:26
it's the arc that I think we're currently in is agents.
00:20:29
I believe that there's a third phase.
00:20:31
I think what's really interesting being so close to
00:20:32
developers is that I've always kind of called out us
00:20:36
developers. We're very self-serving.
00:20:39
So if we get AI, we're going to apply to our job first.
00:20:42
We're going to make our own life easy.
00:20:45
And So what I'm seeing with developers is that the next
00:20:47
generation is going to be teams of agents or multi agent
00:20:50
architecture. We've gone from assistant to
00:20:52
agent to multiple agents. The alpha developers today, what
00:20:55
they're doing is they're, they're instead of just limiting
00:20:57
themselves to like 1 agentic session, they're they're
00:21:00
spinning up 20 in parallel or 10 or 5 or whatever.
00:21:05
And that gives you a glimpse of what I think is going to be the
00:21:07
next phase of AI. You can think of the current
00:21:09
phase of AI as you have a virtual Co worker.
00:21:13
You can think of the next phase of AI as they have a virtual
00:21:16
team of Co workers and I'm the manager.
00:21:19
Is the challenge with just having the virtual Co worker is
00:21:21
in some ways the human gets the lame part of the job?
00:21:24
Or it's like I have to sit there and be like, sure, yeah, I keep
00:21:27
keep doing the hard work. Like, yeah, I keep just
00:21:29
hallucinating. Yeah.
00:21:30
Like, and it's tedious and, and perhaps you don't get that 10X
00:21:33
effect. You know, I do see a lot of
00:21:35
debate, healthy debate sometimes of like, is AIA 2X or is it a
00:21:40
10X for my job? I saw an article that a
00:21:43
programmer put out the other day saying like, look, in certain
00:21:45
areas of my job, it definitely feels like a 10X because the
00:21:48
example that this gentleman was giving was like, I don't write
00:21:51
compilers every day. And so like the fact that I was
00:21:54
just able to sell the AI, can you write me this like compiler
00:21:57
transformation is like, whoa, like I feel like a superhuman.
00:22:00
But then he was saying, like in other areas of my job, like, I
00:22:03
don't know, sometimes it like kind of gets in the way and
00:22:05
whatever. To to sort of really, I guess go
00:22:08
directly at the current moment. I mean, yeah, what's your
00:22:11
reaction been to like ChatGPT 5, I guess like to GPT 5 because in
00:22:17
some ways, I guess I think if that were amazing, the lead here
00:22:20
would be like, oh, the models are going to keep coming up with
00:22:23
new ideas and that's where we're going to make progress.
00:22:25
And just by the fact that you're saying, oh, we're going to find
00:22:28
ways to use the existing models and agents to like get more out
00:22:31
of them by having them collaborate to me as sort of a
00:22:33
sign that we're not having the like, Oh my God, Chachi PT or
00:22:38
GPT 5 is like blown all our minds.
00:22:40
Do you agree with that or what's your read on it?
00:22:42
I agree with you that analyzing the GPT phenomenon is really
00:22:46
important to the industry because you're right.
00:22:48
Like it did, it came out and it didn't come with a short proof
00:22:52
of the Riemann Hypothesis. Exactly.
00:22:55
Where's our exactly in the expectation or whatever?
00:22:59
Yeah, right. But something that was really
00:23:02
interesting was that Open AI is taking the coding problems
00:23:06
really seriously prior to its release.
00:23:09
They came to us and they're like, can you please run this on
00:23:11
V-0? Can you please run your evals
00:23:12
with it? They were asking us like what
00:23:14
are the vibes from your team? What are the vibes from your
00:23:18
tests? Is it good at design?
00:23:20
So there was a lot of attention being paid to the topic of
00:23:24
coding. And the simplistic view of that
00:23:28
is that the market is becoming hyper competitive on the
00:23:31
enterprise side and that Anthropic is making a ton of
00:23:33
progress with coding. There's amazing Chinese models
00:23:36
coming out like Quen, etcetera. They're really good at coding.
00:23:40
And so the simplistic view is like, oh, coding is a hot
00:23:42
market. So that's what they're going
00:23:43
into it. My view is actually that because
00:23:46
agents are the future, not assistants and not one off one
00:23:51
shot things, the future to superior and and and higher
00:23:55
intelligence and more productivity will be agents
00:23:58
writing code to solve problems. So the thing to pay attention to
00:24:01
is not a consumer observation. In fact it almost backfired for
00:24:05
consumers right? Because we were like, can you
00:24:08
please bring back my boyfriend or girlfriend or?
00:24:10
Oh, right. We don't want it to be smart.
00:24:11
We just want it to be. Yeah, like that was the main
00:24:14
thing, right, you know? Art of Maine had almost
00:24:17
anticipated that this was going to happen because they did the
00:24:19
following thought exeriment. If we could revive Albert
00:24:23
Einstein and we all get a digital copy, how many people
00:24:27
would be jazzed? I know a lot of people here in
00:24:29
Silicon Valley. You and I would be like, holy
00:24:31
crap, right? Einstein in a box.
00:24:33
Sort of like you're a crazy guy you don't want to talk about
00:24:36
like, reality television or whatever.
00:24:37
Like when I wouldn't want to talk to you all the time.
00:24:39
I see the point, right? Yeah.
00:24:41
The average person would be getting too much horsepower.
00:24:44
They're like Kate Albert. Like, what should we watch on
00:24:45
Netflix tonight? And he's like, I don't know,
00:24:48
like I don't watch Netflix. Let's talk about internal
00:24:50
relativity. And so I think GBT 5 had a
00:24:53
little bit of that of like, it's becoming intelligent in a way
00:24:56
that's going to enable this like emergence of intelligence that
00:25:02
you can't easily probe for. The future of intelligence will
00:25:05
be and, and we're seeing this with like test time models, like
00:25:08
going in a loop, spending a lot of energy writing code, testing
00:25:13
it, exploiting multiple branches of the search algorithm.
00:25:17
And these are things that a consumer cannot possibly
00:25:19
ascertain. Like the most that a consumer
00:25:21
can do is, you know, I have friends that keep questions in
00:25:25
their head that they know like a is are not very good at.
00:25:29
And so like they do their own little benchmark of like, oh,
00:25:31
let's see if GBT 5 gets this right now, testing the
00:25:35
intelligence or the IQ of AI in itself is becoming really,
00:25:39
really difficult. That's why I open AI was so
00:25:40
interested in like, what is what's the B0 point of view?
00:25:44
You're creating a coding agent that you know is a specializing
00:25:47
in in vibe coding, etcetera. And so that means you have a lot
00:25:51
of tests. That means you have a lot of
00:25:53
benchmarks like, you know, what does that mean for you?
00:25:57
I believe every company should be concerned with creating this
00:26:00
benchmarks of their own institutional knowledge and
00:26:03
intelligence. Well, some of it, you know, it's
00:26:05
like for me, it's like I put it in like, oh, is it good?
00:26:06
How good is it now at writing a story like me?
00:26:09
How, how you know good? Is it keeping track of yes,
00:26:12
proofread problem. You know, you, you end up having
00:26:14
your specific challenges. So when you're in an industry,
00:26:17
it's not that hard to figure out like, oh, is it doing a better
00:26:20
job for me or not? You could also make the case
00:26:22
that you should, you could be working on a writing agent that
00:26:26
it's more than just one forward pass of like.
00:26:29
And this is another thing that kind of backfired, but it's also
00:26:32
very interesting is that they try to ship in a model router,
00:26:36
right? Like the default thing was auto.
00:26:39
And then it's going to choose if it's going to think hard or if
00:26:43
it's going to give you an immediate response.
00:26:46
Another thing we've learned about AI that's fascinating to
00:26:48
me is that there is this metaphor of like thinking fast
00:26:52
and slow, for which there is a book and like people have been
00:26:56
talking about for years, like prior to like AI existing.
00:26:59
And it's, it's proven true. There is 2 modes of thinking.
00:27:02
There is fast, fast twitch muscle fibers or like you just
00:27:06
want like, hey, like, can you provide this really quick and
00:27:09
find me typos? And then there is a like, no,
00:27:10
no, no, like, let's analyze. Like, are you repeating words
00:27:14
too much? And that requires like a slow
00:27:17
thinking. And so open AI try to, because
00:27:20
they're in the pursuit of creating a good consumer
00:27:22
experience. They were like, OK, we're going
00:27:23
to try to route automatically. That appears to me to be an
00:27:28
extremely hard problem, if not impossible.
00:27:30
It's hard to know from the query if it's a hard problem or a slow
00:27:34
problem. We have a related problem in
00:27:37
computer science called the halting problem.
00:27:39
Like by just reading some code, we don't know for how long it's
00:27:42
going to run. It could run until the death of
00:27:46
the universe, or it could run really quickly, but we have to
00:27:48
run it to actually know. And so there's almost like an
00:27:52
uncertainty of like, should I spend a lot of energy thinking
00:27:55
here or should I respond really fast?
00:27:58
In some ways, this is the argument for the role of
00:28:00
startups in AI and in sort of narrowly focused companies.
00:28:05
It's like, really, you're saying the model companies should
00:28:07
almost be an API layer? They're not going to be great at
00:28:10
deciding in every sector whether it should be a faster or slow
00:28:13
thinker and or what. Content or how many agents you
00:28:17
use, etcetera. And so where I was going to go
00:28:18
with your case is that it'd be really cool to be crafting the
00:28:23
ideal writing agent. I, I know most people listening
00:28:26
to this will have an intuitive sense of the answer to this
00:28:29
question, but like, I'm curious to hear you define it.
00:28:33
What, what is an agent to you? Like, does it need to take
00:28:37
action? Like, I mean, deep research
00:28:38
itself is an interesting case where it's like, oh, I mean, a
00:28:41
consultant is like an, an, you know, taking actions in a
00:28:45
certain way and deep research behaves like a consultant.
00:28:48
On the other hand, and people have probably heard me say this
00:28:51
before on this podcast, Like to me, the gauntlet for an agent is
00:28:54
like, can you go spend my money? Like, will I let you really make
00:28:57
sort of permanent sort of decisions that that have real
00:29:01
consequences in the world? I don't know.
00:29:03
How do you think about what what sort of the minimum requirement
00:29:07
for? What makes something the
00:29:08
minimum? Requirement is that it produces
00:29:12
an artifact, an output that is generally longer than an answer.
00:29:18
So it right, like if, if I, I make that decision between like
00:29:23
an assistant and an agent. So if I'm asking like, you know,
00:29:26
what's the province main the largest province in Argentina
00:29:32
versus build me an application. Funny enough, for us, that was
00:29:36
like one of the biggest jumps in how people have perceived the
00:29:40
intelligence of B0B0 started being what do we call one shot?
00:29:44
You would tell it, give me a user interface for a document
00:29:49
collaboration system. And the feedback that we got, it
00:29:52
was like some people like literally have because this was
00:29:54
so early, like it was basically maybe a few months after Chad
00:29:59
GBT launched, some people would tell us, all right, there's not
00:30:04
going to be any jobs anymore. I just saw God, like, holy crap,
00:30:08
what do you guys do? And some people tell us this is
00:30:11
the worst piece of crap ever because on in the one shot world
00:30:16
quality so divergent. It's almost like playing the
00:30:19
lottery. In fact, some naysayers of AI
00:30:22
for coding are saying that like what drives the revenue is not
00:30:26
the quality is the lottery effect, which I completely
00:30:30
dismiss. But I wanted to tell you this is
00:30:31
funny. It's a dangerous idea because
00:30:35
yeah, it's a grabby idea. Interesting.
00:30:37
The next big jump for us was when we made V-0 really work
00:30:42
like an agent. And So what it does is that it
00:30:45
makes that one shot, but then he looks at it, Visio asks itself,
00:30:51
does it work? Does it compile?
00:30:54
The most recent version of her agent even physically looks at
00:30:57
it like using computer vision. So as an example, say I want to
00:31:01
make a website for my daughter's birthday.
00:31:04
Please theme it like, I don't know, Disney Princess.
00:31:10
It'll then look at what it cooked.
00:31:11
It's like, does it look like what I was asked?
00:31:14
And so an agent is taking a multi step approach, much like a
00:31:19
human would in order to produce a higher quality output.
00:31:25
One of the key ingredients there is this idea of reflection, like
00:31:28
you're the agent is asking itself questions, is looking at
00:31:31
artifacts. It is it's using tools.
00:31:34
One tool that we have is it can look at a screenshot, it can
00:31:38
look at designs for inspiration. So sometimes the model doesn't
00:31:42
know what you mean by like build me a podcasting app like
00:31:45
Riverside and so he needs to go and research.
00:31:48
Do you like the term vibe code? By the way?
00:31:50
Do you embrace that term? I personally do it's funny
00:31:55
enough because like our products becoming so successful, like the
00:31:57
Fortune 500, some people are like, do you really want to say
00:32:01
vibe coding? I think vibe coding is it's it
00:32:06
evokes this idea of you're not actually coding, which is really
00:32:11
important. The term came out because Andrew
00:32:15
Karpati and a bunch of our people were realizing AI was
00:32:19
progressing at a clip where your own micromanagement of the agent
00:32:24
was producing diminishing returns.
00:32:26
AI Once Upon a time, like in AI times like 6 months ago.
00:32:30
Once Upon a time it was so bad that you literally had to look
00:32:34
at every token it output. As the models get better and the
00:32:40
world becomes more agentic and self healing, you realize that
00:32:44
you're looking at it. It's just wasting your own
00:32:46
energy and your time. You can start trusting and even
00:32:51
seeing because we can render that output.
00:32:54
We can run the application. You can see it come to life.
00:32:58
And so there has to be a term for that.
00:33:01
The winning term right now is by the coding, but it's basically,
00:33:05
sometimes I've called it instead of being coded first like a
00:33:08
product, like cursor, it's code last.
00:33:12
You might occasionally maybe at the end of the process you look
00:33:14
at the code, but the code is not the important thing here.
00:33:17
And I believe that most agents are going to going back to like
00:33:21
why GBT 5 is obsessed about coding, even though people are
00:33:24
obsessed about four O being nice to them.
00:33:27
I believe that agents will have to write code to get to that
00:33:31
next frontier of value for society and for intelligence.
00:33:35
And I believe that a lot of people will not even realize
00:33:37
that this agents are writing code.
00:33:40
What's happening to your employee salaries?
00:33:43
I mean, you know, the the AI engineers, obviously we've seen
00:33:46
all these stories of meta poaching everyone from the
00:33:50
research labs, like how's this playing out for sort of the
00:33:53
typical software engineer in Silicon Valley or what?
00:33:57
Yeah. What are you seeing in terms of
00:33:59
trends and comp? Are the the $100 million
00:34:03
salaries as widespread as chattered about?
00:34:06
We are in a little bit of a microcosm here, a metaphor that
00:34:10
someone gave me their days like look like it seems like if you
00:34:13
extrapolate what you're saying. And because I know that V-0
00:34:17
hires developers to build V-0, we're transitioning into a world
00:34:22
where the job of an engineer is to build the factories of code
00:34:28
rather than being the line worker.
00:34:32
And so the engineers that are capable to building the
00:34:34
factories of code, the Bible coding platforms, the agents,
00:34:39
the AI interfaces, the assistants, those people are
00:34:43
highly empowered today. And so it becomes that highly
00:34:47
competitive space. It's like, you know, how many
00:34:50
soccer players like Lionel Messi are there that have that
00:34:54
specific set of skills? And then like, you know, there's
00:34:57
lots of soccer players, but a handful that make a ton of
00:35:01
money. And So what I would recommend to
00:35:05
people is that they position themselves in that world, right?
00:35:08
Like become the person that understands how to bring those
00:35:12
two worlds that I talked about earlier.
00:35:14
The theoretical AI potential and it's applied AI presentation
00:35:20
layer. That's kind of the, our thesis
00:35:22
of the Versailles AI cloud is that as this model layer keeps
00:35:27
getting smarter, better, more accessible, you should be
00:35:32
focused on applying that AI to specific domain, a specific
00:35:36
business, a specific vertical, a specific kind of agent.
00:35:40
But you're going to need a cloud that has all of the like right
00:35:43
patterns, tools, defaults, etcetera, for building those.
00:35:49
And that's, that's our thesis with, you know, we're not going
00:35:51
to build a cloud that is 1 to 1 exactly like AWS.
00:35:54
There's no advantage there, but we want to build a cloud that is
00:35:58
perfect for that next generation of companies and developers.
00:36:03
I'm trying to understand. I mean, this is we're in an
00:36:05
insane situation where you're right at the high end.
00:36:08
If you're building the AI factories, you could command
00:36:11
insane salaries. On the other hand, if you're
00:36:15
sort of the sort of, I don't know, worker, more like line
00:36:19
coder, software engineer, I guess you're worried like, oh,
00:36:22
you see V-0 and you're like, oh, that's coming for my job.
00:36:25
So my question I think is coming at you from 2 ends, which is
00:36:28
like at once, like, oh man, we see these insane salaries.
00:36:31
On the other hand, I feel like they're the sort of, I don't
00:36:34
know, Wall Street Journal story saying don't, don't study
00:36:37
computer science because they're not a new job.
00:36:39
So I guess take one at a time because I think they're getting
00:36:42
mixed. They're getting mixed together.
00:36:44
One talk about each end of of the equation here.
00:36:47
For the other end, yeah, I do have a complete different take
00:36:51
on universities and colleges and that is a kind of worms.
00:36:55
I think we should address it. OK.
00:36:57
On the line coder, I also think that the demand for code from
00:37:01
those people has actually never relented.
00:37:05
Meaning let's say that you were able to produce like 1000 lines
00:37:09
of code per hour per day, whatever your boss probably
00:37:15
wanted you to produce even more like there's more features or
00:37:18
bugs, there's more innovations, more products, etcetera.
00:37:21
And so if you use AI, you're that's why I was saying that
00:37:25
it's so exciting to be seeing people go into this multi agent
00:37:29
ways of working. And so the person that uses AI
00:37:34
is going to run laps around the person that doesn't.
00:37:37
But do you? Think they have more like
00:37:39
frontline engineers in a decade than we do today?
00:37:43
It's very hard to estimate in, in like in total numbers, right?
00:37:48
But I do think we'll definitely have more software, right?
00:37:51
Like, right. Yeah.
00:37:52
The demands, we all agree on that.
00:37:55
The, the definition of an engineer will likely expand in
00:37:58
such a way that I can, I can tell you definitively, yes,
00:38:01
right. Like, because we're putting
00:38:04
software generations in the in, in the hands, in the hands of
00:38:07
everybody as a whole, there will be, you know, hundreds of
00:38:12
millions, billions of people creating software, be a specific
00:38:16
engineer as we understand them today.
00:38:19
I can't even, you know, I, I had a tweet a couple days ago that
00:38:23
Elon replied to say I, I was saying, like, I can't even
00:38:28
explain just how different software engineering will look
00:38:30
in five years because the perception that people have
00:38:35
today is of a code editor. Like that's it.
00:38:39
Like if you watch the hacker movie, what is the idea of a
00:38:42
programmer or they're looking at a code editor?
00:38:45
But now I'm looking at my fellow factory builders and I don't see
00:38:53
them looking at code editor anymore.
00:38:55
I'm increasingly seeing them talk to all these agents.
00:38:59
And so that to me already like if I was able to transport
00:39:02
myself to five years from now, I don't think software engineering
00:39:06
looks anything like what do we do today?
00:39:08
And so it's very hard to say like like are we calling those
00:39:11
people that are managing the fleet of agents engineers?
00:39:15
You know, most likely I would say it's, it's more on the
00:39:18
customerization or democratization of software
00:39:20
creation. To a direct point you were
00:39:23
touching on earlier, do you think people should study
00:39:26
computer science right now? I think people really need to
00:39:31
understand how AI systems work, period.
00:39:36
They need a deep understanding of the technology stack, the
00:39:39
limitations of the technology, what makes them safe, what makes
00:39:43
them not safe, what makes them make mistakes, whether they're
00:39:47
good, whether they're are they not good.
00:39:50
I do have a little bit of apprehension about do the
00:39:54
universities today themselves know?
00:39:57
And I'll say most likely not, right?
00:40:00
Like, think about how fast the industry is moving.
00:40:03
Right that. You're like come work in
00:40:05
Versailles and figure it out here and skip our user.
00:40:08
Road map is moving week by week, right?
00:40:12
It's very hard to say in Q32026, Visor is going to look this way.
00:40:18
We're going to need to find a way to get the universities up
00:40:22
to speed on AI, on vibe coding or start to blur the lines
00:40:28
between university and industry. We've been running very
00:40:33
successful internship programs at our cell where I personally
00:40:36
spend a lot of time with the interns.
00:40:37
Like I'm picking their brain, they're picking mine.
00:40:40
Like this intern is like, you know, hanging out with CEOs and
00:40:44
CT OS because like, I do think that you are getting up to speed
00:40:48
really fast if they're in the right room.
00:40:50
I do think there's a little bit of a get rid of your ego.
00:40:53
Forget about everything you are good at.
00:40:55
It starts from first principles. If you, if you thought you were
00:40:59
really good at TypeScript, if you thought you were really good
00:41:02
at C, well, yes, but also someone that never knew how that
00:41:09
technology even worked will probably be generating lots of
00:41:13
it just like you very soon. They might be even be learning
00:41:18
faster than you because you're able to try more things.
00:41:21
I'm very concerned in the sense of like people say, the junior
00:41:24
developer job. Instead, I would argue maybe
00:41:27
there's nothing that's ever been more important than using your
00:41:31
company with clinical junior developers that are more AI
00:41:35
native and can show you new approaches truly to learning, to
00:41:41
iterating, to figuring out how to solve a problem when you
00:41:44
don't know all of the science behind it.
00:41:46
When you ask about computer science, computer science is a
00:41:48
very specific version of like how they teach you data
00:41:51
structures and algorithms, etcetera.
00:41:54
If they're leaving out this like pragmatic AI side of
00:41:58
engineering, I really liked what Elon said about like we're
00:42:01
nuking the term researchers at Grok, at XAI, we're all
00:42:05
engineers. I do think it's a very healthy
00:42:07
mindset to have. We're all engineers and we're
00:42:10
all figuring it out and we're all using AI to figure out as we
00:42:13
go. Taste has become, you know, a
00:42:16
keyword in how people think about the role of humans going
00:42:21
forward in AI. And I, I want to use that as a
00:42:25
lens to sort of untangle some of the coding stuff we're talking
00:42:29
about. Like, you know, obviously I
00:42:31
spend a lot of time thinking about writing with writing like
00:42:33
you can, everybody can have taste about stories because we
00:42:36
can all like read and assess them.
00:42:39
And obviously there are sort of the practitioners who sort of
00:42:42
know what goes into making it. But then there have always been
00:42:45
sort of critics of, you know, novels and writing that sort of
00:42:49
you can imagine with AI, the sort of line between the writer
00:42:54
and sort of this sophisticated critic of writing sort of is
00:42:58
collapsing because you can sort of use AI to write over time.
00:43:03
I guess with code. My my question is like, there's
00:43:07
sort of the, there's the product, there's like what what
00:43:10
is the thing outputted that you could have taste on?
00:43:13
And that's something that seems like it's being democratized,
00:43:17
like potentially writing. But won't this sort of code base
00:43:22
itself still matter? Or like, will we need engineers
00:43:26
that have like the taste for like, what is well written code
00:43:30
and like how systems are set up? Or are we just going to like,
00:43:33
yield that piece of it to agents?
00:43:36
Like, is the taste only about whether the product sort of
00:43:39
itself looks good, works well, or is it about like how the
00:43:44
actual code base is set up? When I first came to Silicon
00:43:47
Valley, one of the things that was and vogue was there were a
00:43:52
handful of people in the Valley that could create deep tech
00:43:57
software systems and also design beautiful products and I could
00:44:01
count them with like one hand. There were a handful of people
00:44:06
that I knew and I aspire to become like them.
00:44:09
I've cared always a lot about the design and the engineering.
00:44:13
To me, that's what was so special about the Steve Jobs and
00:44:16
Apple mythology is that they literally had a slide in one of
00:44:20
their presentations of like, Apple at its best is the
00:44:23
intersection between computer science and liberal arts.
00:44:27
And they had a sign of like the intersection of two streets, one
00:44:29
called computer science, one called liberal arts.
00:44:33
I think the future will be more of that.
00:44:36
I would argue that the liberal arts is going to be overweight
00:44:40
because the computer science part, well, the agents will
00:44:43
figure out all the data structures, algorithms,
00:44:45
optimizations, blah blah. And so your ability to infuse
00:44:49
creativity, taste, culture, design will probably matter even
00:44:55
more in the future. You know, I'm happy to hear the
00:44:58
argument for like, the liberal arts major, but I still think,
00:45:02
you know, tech companies still need to build things today.
00:45:05
And it's not just about like clever billboards and clever
00:45:08
marketing. Yes, but there is a discovery
00:45:11
element to technology like there's just so much technology
00:45:16
that is possible and yet we don't know it's available to us
00:45:20
and so when people. Say this stuff, I'm like, oh, I
00:45:22
have a distribution platform. This is what everybody seems to
00:45:24
want. I just need to do what you guys
00:45:26
are saying is the easy part. Oh, just cobble on some tech
00:45:29
thing to sell people. It's like, oh, I have people's
00:45:31
attention. It's like clearly I'm deficient
00:45:34
in figuring out the, you know, tech thing to just like cobble
00:45:37
on. But but to me, from where I sit,
00:45:39
it seems harder than some of the vibe coding it takes.
00:45:43
It's like, oh, you can't just spin up.
00:45:44
You can't be like, oh, I have attention, let me just spin up.
00:45:48
There is a software that I pointed there that you're
00:45:50
pointing out, which is that even if you figure out a huge
00:45:52
platform, you need to create something that's great for that
00:45:55
platform. And so if software is getting
00:45:58
democratized, like maybe it's when I mean, when I mean that
00:46:02
everybody can cook. A lot of it is like solving your
00:46:05
own problems and and creating tools for you and your
00:46:08
colleagues and things like to that matter, creating software
00:46:12
at scale, like if you get the biggest like Times Square sign
00:46:18
and you're going to get every eyeball on the planet and you
00:46:20
want to give them something of high value.
00:46:22
It's so really hard. Yeah, you can solve your own
00:46:25
problems. You know what you need.
00:46:26
But to solve some sort of general problem for a bunch of
00:46:29
humans is is a hard product. Building that is that's very
00:46:32
hard. My argument was there's
00:46:35
something really interesting about the Steve Jobs story
00:46:37
because he wasn't an engineer, or at the very least he wasn't
00:46:39
an engineer in the, in the sense that he's Wozniak was an
00:46:43
engineer. And so there is this huge skill
00:46:46
set to figuring out one, what are the boundaries of the
00:46:51
technology? What are, what are the limits of
00:46:52
the technology #2 how do you best percent that technology #3
00:46:58
how do you capture people's attention and distribute it?
00:47:01
And so I think Steve Jobs was really good at these things.
00:47:04
And then he quote, UN quote, used the team of agents, the
00:47:08
people that he recruited. Well, as human beings, right?
00:47:11
Yeah. But you get to kind of see where
00:47:14
this is going, right? There will likely be a bunch of
00:47:18
Steve Jobs that figured out that trifecta and are offloading the
00:47:23
hard engineering work to agents. But also there there might be
00:47:28
teams constituted almost entirely of Steve Jobs people
00:47:32
like, we're all tastemakers in this room.
00:47:35
We all bounce ideas around and like what kind of products we
00:47:38
want to see in the world. We do a bunch of research.
00:47:40
We talk to customers, but we're basically like, I would say it's
00:47:45
almost like the return of the product manager.
00:47:48
I think you're afraid that iconic moment in culture of the
00:47:53
PS by the pool. The intersection of engineers
00:47:57
resenting product managers and general Twitter loathing of
00:48:00
women, Yes. I'll tell you, like, I don't
00:48:02
think people will have seen this coming, right?
00:48:06
But there is a return to, you know, well, do you need more
00:48:10
than the PM with the person sort of like understanding market
00:48:15
customers requirements, E etcetera.
00:48:18
And so it's almost like a pendulum swing back and some on
00:48:21
some levels at least. But yeah, I think the, the, the
00:48:24
reality is is likely to be a lot more nuanced than any of this
00:48:27
prediction. But my, my, my personal
00:48:30
predictions in five years, this whole industry does look
00:48:34
extremely different, Extremely, extremely different. 16 month
00:48:37
prediction Six months from now, what will be different?
00:48:41
Six months. I do think there's going to be
00:48:44
maybe even sooner than this. I think people are
00:48:46
underestimating GBT 5. I think they're judging it by,
00:48:50
again, judging a model is becoming an important industry,
00:48:55
right? There's companies like LMRE.
00:48:58
Takes. I'm like, I really need to sit
00:48:59
with a model and sort of really understand.
00:49:02
And that's an emergent thing. I I don't remember technologies
00:49:04
where like you really had to sit with them for like like aiding
00:49:08
to know them like a person. Right.
00:49:11
It's like, yeah, you want to assess their soul, like on the
00:49:13
first meeting, you know, exactly.
00:49:15
We have to sit with it and sort of see what it can do.
00:49:18
It's like it's like you're 306090 day plan for a new hire
00:49:24
or a new executive like you have to treat a model.
00:49:26
Underestimating GPD files? Yeah, I think.
00:49:28
I think people are. I really think people are.
00:49:31
I also think you are not realizing that this vertical
00:49:34
agents, there's probably going to be some agents, maybe some
00:49:38
like agents for video editing or creating video or creating ads
00:49:43
or you know, I can't tell you exactly what.
00:49:45
There's going to be one that we're all going to be talking
00:49:47
about. Interesting.
00:49:48
Yeah, Yeah, we need to break out Chachi.
00:49:50
BT has so much defined the use case.
00:49:52
We need sort of a vertical niche where it's like it's going to
00:49:55
solve this problem, do this thing calendar.
00:49:57
Something along this line like it seems so odd but how do we
00:50:01
not do it? As soon as Chachi BT came out,
00:50:03
we didn't do it because we didn't have the infrastructure,
00:50:06
we didn't have the best practices, we didn't have the AI
00:50:09
gateways, we didn't have the agent frameworks.
00:50:12
We built a lot of this stuff. And so there's going to be an
00:50:16
obvious in retrospect application with a huge user
00:50:21
base. That's my.
00:50:22
Perspective right now that Gmail is not solving look that the
00:50:25
model is not in my e-mail yet To me is the, I mean superhuman
00:50:28
wants to do it, but to me like Gmail, it's like sitting there.
00:50:31
There's a lot of risk. Obviously I would be a bad
00:50:34
reporter if I didn't ask. You can dodge or not.
00:50:37
The information says you're getting approaches at 9 billion.
00:50:41
Do you have anything to say about that?
00:50:44
No comments. No comments, speculation.
00:50:46
All right, Guillermo, thank you so much for coming on the show.
00:50:49
We have to do this again. This was a lot of fun.
00:50:51
Hey Gary, always play.