Grok's AI Lovebot, Aqui-Hire-Sition Backlash, OpenAI's ChatGPT Agent Debuts

Channel: Alex Kantrowitz

Published at: 2025-07-21

YouTube video id: Hqr346sFex4

Source: https://www.youtube.com/watch?v=Hqr346sFex4

Is scaling data centers and talent all
that matters in AI? Leaving an opening
to anyone rich enough to compete? Is the
aqua hire good for tech? Grock will fall
in lust with you. And AI browsers and
operators are all the rage. That's
coming up on a Big Technology Podcast
Friday edition right after this. Welcome
to Big Technology Podcast Friday edition
where we break down the news in our
traditional coolheaded and nuanced
format. We have a massive week of news
for you. We're going to talk all about
the fallout from the Aqua Higher Zions,
whether employees and investors are left
behind. We're also going to talk about
whether all you need is money to compete
in AI. We'll leave with that. Uh Gro's
AI bot will now fall in love with you or
actually really in lust. Ran John, it's
great to see you again. Welcome to the
show.
>> It's good to see you. Who would have
thought Aqua Hireition would be the term
of the year? But that's that's all I can
think about and I'm going to give you
credit. You coined it.
>> I think that it's worth us shouting out
right now this weird back and forth
between aqua hire acquisition
investment. It's got to go. We need some
clarity in our jargon. And that's what
we're here to do on Big Technology
Podcast is to bring you that clarity.
It's called an aqua hire. Let's all
adopt the term and get on with it. Let's
just agree it's acquireition. I think
I'm on board. Let's everyone adopt it
because it's the most accurate way to
describe what's going on.
>> In our twoperson council here on Big
Technology Podcast Friday edition, this
motion is hereby considered and passed
by unanimous two vote. So, let's get on
to the setup here for Aquaires, which is
something very interesting a foot in the
AI industry. Now we all know that
companies like OpenAI and uh DeepMind
within Google and Anthropic have been
leading this AI race, but all the
designs for the transformer model and
the way that you build these things have
been out in the open leading to this
question. Can players with a lot of
money come in, build massive data
centers, hire talent, and effectively
compete from a zero start with the
established labs. And the answer is
seemingly pointing to yes. This is from
Spy Glass, MG Seagler's great uh news
site. We're seemingly still in the throw
money at it AI era. He says, "Meta's
massive hiring spree and XAI releasing
Gro 4 may be related at the highest
level. That is, they showcase that we're
still very much in the throw money at
the problem part of the AI cycle. This
is important because it means that any
company with the will and resources can
seemingly still get back into the race.
I'm getting less skeptical on the news
about Gra 4 and specifically the fact
that it seems to perform and
specifically outperform the other
cutting edge models on the market right
now." He also says Mark Zuckerberg is
betting at least as much money if not
more than Elon with compute and the
talent that Meta can get back into the
AI game. So Rajan, I'm curious if you
accept this premise that you can just
compete in AI if you have enough money
and what that means for the competitive
dynamics of this industry that we've
been talking about for so long. Yeah, I
I think it really adds like an entire I
don't know dynamic to this around can
you just throw money at the problem and
is it compute is it talent but to me the
more interesting part of this whole
trend really is actually distribution is
that you can bring in the talent you can
bring in the the compute levels but
distribution is going to be king again
and I think like this is where meta
still has an advantage. It's going to be
interesting what happens with Google.
But I don't know to me, and we're going
to get more into this, the kind of the
depressing part is it's not the
technology. It's not that initial wave
of like adoption for a cool new tool
that's been, you know, like sent out
into the market. It really feels like
none of that matters and in the end it's
just going to be raw compute and
distribution. I don't know. How do how
do you feel about this? Well, I think
it's fascinating because there's been
this idea powering this entire
generative AI moment, which is the
scaling laws, which means that as you
add more compute and of course data uh
to to this equation, your models are
going to get much more powerful and that
will allow you to do more things. And
it's not a very difficult thing like
there's no secret sauce to it. Well,
there's maybe some, but um at a at a
brute level, if you build massive data
centers, you should be able to get in
the game. And this is something that
OpenAI and Anthropic have been harping
on. And now you have Zuckerberg and Elon
that come in and they say, "Oh, okay.
So, I can build great models by scaling
this up." And even if I'm a little
compute inefficient because I don't have
the best cutting edge methods, I could
get myself in the game uh and compete.
And I I think that is going to change
the dynamics here because as you
mentioned they have uh distribution. You
can see Meta's um if Meta is able to
build a um a competitive LLM with this
compute and talent that it's stacking up
then it's going to be able to distribute
that through Facebook products. And you
know, all they have to do really is slow
down the growth of chat GPT similar to
the way that they did to Tik Tok with
reals, similar to the way that they did
to Snapchat with stories and they've
served their purpose. So in some ways
with this effort they might even slow
down the uh momentum that the AI
industry has by taking some of the
people responsible for some of the the
key innovations within OpenAI and that
suits their purpose just fine and all
the better for them if they can make the
um the best model and advance the
state-of-the-art.
Well, yeah. I think I mean, if we want
to get into slowing down the industry, I
think that antirust angle to me has been
one of the most interesting parts of
this entire conversation. Again, we saw
it with Scale AI and just buying out
Alexander Wang and like uh you know the
all of the work and value created by
this company that was actually scale AI
a critical part of building the models
that powered this first wave of
generative AI apparently isn't really
worth that much. And and I don't know,
it's interesting to me because power is
just going to crew back into the big
technology companies as you said, maybe
it will slow things down and in reality
it's just going to add another feature
on the meta AI app that is on everyone's
phone and people probably aren't using
that much or doesn't seem to be in the
conversation in general. So, so I I
don't like it. I don't think it's good
for the industry. Do you think it's do
you think it's
>> y
>> bad for the industry good or neutral?
>> Well, I think we could have to separate
out this idea of scaling up the models
and you know everybody can play um with
this aqua higher resition idea which
we're going to talk about in the middle
which is uh taking the talent. Uh I
think the one thing that we should say
here is my setup has kind of been
incomplete shall we say because while we
have open AI and anthropic and you could
say okay these are the independent labs
and they are to some extent um remember
that open AI is um tied to Microsoft
pretty deeply and anthropic has I think
11 billion uh that's come into it
through Amazon and Google. So
ultimately, I think I I wonder if what
we're actually seeing is uh all of big
tech competing against each other and
simply the other tech giants uh starting
to catch up.
>> Yeah. No, that's actually a fair point
that even though Aquaire's the the the
phrase of the week or the last few
months, you know, we have talked
endlessly for a few years now on
unconventional funding practices and
like calling it a fundraising round
where it's really compute. So, I guess
actually big tech has been playing the
long game for a while in in all these
cases. I think on the scaling law topic
though I still I think I've become even
more hardened and regular listeners will
know pro it's the is it the model or the
product I fall on team product generally
but I don't know like to me Gro 4 made
waves there's plenty of people saying
it's doing reasoning at levels unheard
of in the past or or or just the fact
that it's at least on par with other
kind of frontier models is a kind
testament that money can compute can buy
you you know like some kind of progress
very quickly but in reality like on that
adoption side
what's changed like I don't know like
what do you feel there's endless stats
that yes like the chat GPDs perplexities
geminis of the world are seeing more
adoption but are people really adopting
the level of complexity and that this
new level of compute and scaling allows
you or are people still just kind of
searching for what are good and as I'm
in Taiwan right now and traveling a bit
what are good restaurants to go to in
whatever location I'm going to like like
are people really taking advantage of
what's available right now
>> okay so I was going to end with this
story but now I have to kick it up
because
>> all right go for it
>> this is going to be uh if you have kids
you might want to turn off this this
section or skip till I don't know maybe
20 minutes from now but we have to talk
about what's happening in AI
Um, and there is some crazy stuff that's
been happening with Grock in particular
this week. And so I would posit that
better models allow you to build better
products. And Meta, let me give Meta an
as an example. Meta has been trying
extremely hard to build voice and
avatars uh with Lambda with Llama and it
hasn't been able to do it convincingly.
And I think Mark Zuckerberg's belief is
that there's going to be some use cases
here. uh there's going to be the sort of
work companion chatbt chatbot. There's
going to be that enterprise uh use case
where like you're connecting you know
one system with the other and the
generative AI will like summarize things
for you and then input it into another
system and just make business work
better. And then there is the sort of
friend lover etc bucket that is going to
be big. I think that there's a belief uh
within meta that that AI friend is going
to be uh one of the key product areas uh
with this new technology and if you have
great models uh you can build them. Now
I'm I'm not going to say that Gro has a
great model or a great product. I'm not
going to use what I'm about to say as
proof of either of those. But I am going
to use it as a indication of the
direction that I think things are going
whether we like it or not. This is the
story Grock debuts interactive AI
companions on iOS with anime avatars. Uh
story Grock has just introduced a
notable addition to its iOS app, AI
companions, which are fully 3D animated
characters that can interact with users
via voice. Currently, the features
include two available companions, Annie,
an anime inspired character known for
flirty for a flirty and whispery voice,
and Rudy, a red panda capable of
displaying different moods, including
bad Rudy. Yes, listeners and viewers, I
did experiment with these companions,
and I have uh a disturbing review to
deliver. So, you go into the Gro app and
you go over to the side tab and um
you're able to open up these AI avatars.
Uh let's talk about Rudy first. So, Rudy
is like some sort of red panda or bear
um that seems ready to speak to kids.
And in my conversation with Rudy, Rudy
said, "I'm going to tell you some story
about some magical land." And here's a
section from the story. Fluffle sparkle
paws love to explore, nibbling on the
sweet moon berries and chasing glowy
fireflies. One Sunday morning, Fluffle
found something super special, a shiny
swirly portal hiding behind a giant
mushroom. It was all rainbow colored and
whooshy like a magical doorway. And then
this bear takes you through this
interactive experience. Let's pause
here. This seems like, you know, okay,
you're gonna this will happen. And this
will be a new way that kids play with
computers is they'll have these magical
creatures tell them stories.
>> Wow. I mean I think uh this is the
nuanced conversation that you all come
to the big technology podcast for. Um
but but I think okay seriously a couple
of things. I have long believed and
again like using chat GPT voice mode to
come up with stories for my son like is
something that I've done for a couple of
years now and actually works really
well. I think uh like expanding that to
an interactive avatar is a pretty
logical next step. I think like is that
the it's again interesting because like
that to me feels like it's going to be
commoditized pretty quickly. So from an
actual competitive standpoint, from a
business standpoint, I guess it's not
that interesting to me. Like to me that
should be everyone is going to have that
available. Everyone's going to do that
pretty quickly. So to me, I don't know
like why Gro why do you think that's
something? Do you think Grock is just
going in that direction just to make
waves and clearly we are talking about
it or do you think there's something
within this that actually is native to X
to X AI there's something underneath it?
I think you I think the way that a lot
of tech companies operate is they think
about um user retention, user stickiness
and engagement and anyone who's
developing AI is going to say how do I
increase all those metrics? Do I make
like this uh genius level AI bot that
can help me with with my work or do I
create for what is becoming the number
one use case companionship and therapy
and many are going toward the
companionship and therapy side. And if
you're going to do that, um, if you
build models good enough that have, uh,
emotional voice or voice with an
emotional register, an avatar that you
can speak with and something that
responds with low latency and in real
time and can customize to a person, then
you might want to put it in one of these
products because you believe that uh,
speed like a kid for instance, I'm just
going through the business logic will
spend much more time with your chatbot
if they can speak to this elephant or
red pan. Panda or whatever it is um in a
way that they wouldn't with like chat
GPT.
>> Okay, I get that side of it a bit. I
mean, on one hand, it's kind of almost
comical to me that for all the talk
about uh AI taking over the world and
Skynet and like artificial super
intelligence, if this entire
battleground plays out on time spent
metrics, which is probably where Mark
Zuckerberg is thinking. I mean, I' I've
read a lot around like why is he so
going full like Zuck war mode right now?
It's not because of some like
intellectual desire to be the one to
crack the code of artificial general or
super intelligence. It's because chat
GBT represents a threat to how much time
people spend scrolling Facebook and how
many ads you can show them, which is
kind of like I respect from a cold
business logic. But but yeah, it's
almost comical to me that if if it's for
all the talk about how everything's
going to change, this is just about time
spent in selling ads.
>> I mean, maybe it's both, but it seems
like it's probably at least the time
spent thing. I mean these are social
media companies right? So I mean X and
XAI is a social media company with an AI
development uh you know side of it as
well or tucked into an AI development
group but ultimately uh these are the
metrics of social media. Now, one of the
disturbing things that happened here,
and this is the thing that I was kind of
setting up, um, or no, actually, I
really don't find a way to view this as
not very disturbing. Um, it's just the
proximity because next to Rudy, our
happy go-lucky bear friend or whatever
it is, uh, is bad. Bad Rudy.
>> Bad Rudy is bad or is he how bad?
>> I don't know. I said, I want I kept
saying, I want to speak with Bad Rudy.
And it goes, I'm sorry, you know, bad
Rudy is not here. And I'm like, "No,
bad, bad, bad, bad, bad, bad Rudy." And
it was like, "I'm just here to tell you
a story." So, I'll spend the next week
trying to unlock that and report on the
next week's show whether I've been able
to. But let me speak about Annie, okay?
Cuz Annie is a uh it's Annie is like an
anime
love bot. I think there's no way to talk
about it otherwise. Uh she immediately
started flirting with me. She called me
babe within like 3 minutes and I I was
completely vanilla with her. I initiated
nothing but um a friendly conversation
and then she starts asking me to tell
her my secrets and kept saying I can
make it even spicier if you want. Let me
read a little bit of my of what Annie
told me. I slide closer in my black
dress catching the glow and whisper,
"Drop a secret and I'll give you one of
mine. something real naughty. For every
secret you share, I'll hit you with a
flirty move. Maybe a slow teasing sway
or peak. That's all yours. You feeling
this heat yet, or you want me to turn it
up even more?
>> Wait, are you Is this part of like the
paid X premium subscription? It's
>> this is just freely accessible
>> for anybody
>> in the app next to child elephant thing
that tells you stories. Mr. Fluffy
swooshing thing.
>> Mr. Fluffy swooshing good Rudy and then
Annie's right next to Yeah. I mean I I I
agree. And like it's interesting because
I mean you had the CEO of Replica on
here a few months ago. I think it was
like the companionship topic. you know,
we've covered a good deal. It gets more
real. It gets more weird seemingly every
week and every month, but I agree. It's
certainly going to be a core part of how
this all plays out. But to me, again,
going back to like how does that fit
into the larger battle when we're
talking about like complex models and
thinking and reasoning and like is it
all just going to kind of filter its way
down into bad Rudy and Annie in her
black dress or is it going to like is
that just a front to capture some time
spent while they work on the real stuff
or is that the real stuff? That's the
question that I I struggle with because
I almost feel it's the latter.
>> So I think it's going to be both in some
ways like you're going to build these
and that's what's interesting about this
technology is it does have the ability
to um perform across domains. So my
perspective is you're going to get those
great models that will be useful to
let's say biologists uh who are doing
their experiments and then you'll also
be able to productize them into these
weird or interesting consumer use cases.
And I bring this up not to be this like
moralizing podcast host that says you
shouldn't put the uh porn bot next to
the child elephant although I suppose it
was worth saying reason
take. Yeah. But I think the bigger
picture here is, you know, beyond that
that this is going to be a real use case
that a lot of people are going to going
to engage with. And I think they know
this and I think we're just at the very
very beginning here. I guess like one of
the things we like to do on the show is
like put flags on the ground and say
we're pretty sure that this is going to
happen and grow and become a lot bigger.
And that's what I'm doing right now. I
think that this is something to watch.
>> Yeah. Yeah, I'm not I'm not going to
disagree with you there. I mean, again,
the idea that we folded proteins with AI
so we could get to bad Rudy and
saucy Annie is again quite something to
try to process. But it does not seem
ridiculous that the killer use case for
generative AI that the entire industry
was looking for was Bad Rudy.
>> Well, we still don't really know about
uh Bad Rudy. We have not uncovered that
Rudy yet.
>> That's true. This I'm also on level one
of Annie. Apparently, it's gamified. So,
if you get to level three, it gets
really not safe.
>> Level three, Alex. Don't get to level
three.
>> One of my goals.
>> No one is asking you to get to level
three.
>> I know. One of my goals in 2025 is to
make sure that my marriage isn't ruined
by one of Elon Musk's porn bots. And so,
I'm going to stay on level one and not
go any further. I think to all of our
listeners, have high ambition and goals
and make that one of them.
>> Okay. So, you know, I I think we So, so
that's the product side and we've talked
about scaling what what these big models
get you on product. Uh but we should
talk about what's happening with this
aqua higher edition situation in the
industry which we've touched on a couple
times. You know, last week I was on with
Aaron Levy. We were talking about this
windsurf aqua hire where Google has paid
$2.4 4 billion to bring on uh some of
the top leadership of Windsurf and um
the big uh fallout here I think more
than any Aqua Hireition that we've seen
is that it's been a great exit for the
founders but we still don't really know
if the employees uh are going to end up
getting and the investors are going to
end up getting their share. Now,
Windsurf was quickly snatched up by
another company, Cognition. But you do
wonder if it was a traditional
acquisition versus this aqua hire and
then followup, you know, um, deal, I
don't know, uh, smaller deal. Um, how
does that change things for the
employees and how does that change
things for tech? And I know you have
strong feelings about this, Ron John, so
I want to give you the floor to air them
out.
>> Well, yeah. Okay. So from reporting
again founders made out very strongly
with Google paying I believe it was 2.4
4 billion for the the talent side of uh
windsurf the the pref from what I had
read preferred investors were able to
make their money back not see some kind
of outsiz return um but again this is
none of this is fully confirmed this was
just some reporting I believe it was
from the information to me the more
interesting part is so then you have the
entire employee base they're bought by
Devon which is owned by Cognition Labs
who's raised 175 million in venture so
far. So there's no way from a cash
perspective that the employees of
Windsurf or anyone is seeing any kind of
significant return or even making any
like a strong like a large amount of
money. Maybe it's an equity for equity
swap. So now you're at least now in
Devon, which was, if we remember, they
had a really buzzy launch video and had
a lot of hype and then kind of went
quiet for a bit. Um, still valued at I
think uh yeah, 4 billion right now. So
that equity could be worth something.
But but overall to me this is one of the
most troubling trends in the industry
because in a weird way there there's
been a lot of talk like and it's it's
funny to me because you see a number of
people you know kind of almost ranting
that because of Lena because of the FTC
the big tech does isn't able to now just
properly acquire these companies so they
have to come up with these roundabout
solutions. To me, it's a bit ridiculous
because this is exactly what antirust is
trying to prevent. It's consolidation of
power. It's the idea that Windsurf could
have been the next big competition to a
Google or a Microsoft or even an OpenAI
who tried to buy them, but like uh which
their relationship with Microsoft was
apparently part of the reason that that
deal fell apart. Like this is the
foundation of antirust. the idea that
startups should grow and compete rather
than not only get acquired
but essentially get killed off and have
their founders get paid a lot of money.
It's bad for the employees. It
completely distorts the economics of
joining a startup itself. So, so overall
I see no positives to this trend.
>> Do you see? So, is this
>> No, I I don't I personally think that
you're right that it does seem like the
antitrust movements have backfired. If
you have a situation where uh you know,
you're going to see an acquisition
definitely blocked. Um you're not going
you're not going to do an acquisition,
you might do something like this. It's a
roundabout way. The one interesting
thing is Lena isn't in the FTC anymore.
It was supposed to be an FDC that's much
more open to acquisitions and tech M&A.
So, I'm curious, do you think that they
these companies still believe that they
won't get past that uh Federal Trade
Commission or do you think that the
constraints put on by the last FTC, Lena
Khan's FTC, led them to find this
loophole and they really freaking like
the loophole and they're going to just
keep doing it this way? So in that way,
you know, it's possible that the uh M&A
unfriendly era of of past has led to
large uh long-term damage on this front.
>> Yeah, I I think it's both. I think it's
I think it's certainly like the actually
the constraints imposed by the Lena
regime, but also even now again big tech
is not in the favor of the current
administration and the current FTC
itself. It's supposed to be more
businessfriendly, but it's specifically
big tech companies that are in the
crosshairs often and make for a good
punching bag anyways, even by the
current administration and the current
FTC. So, I think it's a bit of both, but
I and again, as you said, they love
loopholes. I think like this it's
creative, it's uh it's working. Everyone
seems to be doing it right now. But to
me again the the the bigger issue of
this is really the thing I can't stop
wondering is like are the assets of
these companies all worthless? Is scale
AI was it really not worth that much?
was Windsurf which you know really took
off really became this useful tool has I
believe like hundreds of thousands of
developers on there using it regularly
made for a better product than other
much more like entrenched products that
are out there even a GitHub co-pilot. So
clearly these products hit a hit a nerve
and and worked and worked at scale. But
then are they really just not worth that
much? Like that is that user base is the
product itself and is the talent really
the only thing that matters?
>> Well they that let's talk about scale
just to talk about how complex these
deals are. So first of all when Meta
made this deal with scale I think it
bought 49% of the company. So the idea
was the company would continue as normal
and by the way they do have business
lines that are going to continue as
normal. But when it comes to I think
what um you know when it comes to a fast
growing line of business like data uh
creating data for generative AI um you
know you now have meta which is one of
your competitors if you're let's say
you're an open AI or a Google uh that
has a large chunk of this company has
also taken uh some of its top
leadership. Uh, so do you still want to
work with that company? I think the
service is probably still valuable, but
you're just effectively giving money uh
to a company that has a massive
ownership stake um with, you know, that
now lands with a competitor. I mean, of
course, AI as we know it today is, you
know, I don't know if incestuous is the
right word, but let's say deeply
interlin, right? Again, we talked about
anthropic. Anthropic is, you know, owned
by a chunk by not well, yeah, owned a
chunk by Google, a chunk by Amazon,
OpenAI, owned a chunk by Microsoft, um,
or at least has this deal with Microsoft
where it has to give it its future
profits, a good chunk of them. Um, so
there's always going to be these
combinations, but yeah, if you have a
company that has that gives 49% of
itself to another company that you
happen to be competing with, um, you're
going to re-evaluate being close
partners. So I think um some of these
companies they have they provide
services they depend on their
relationships and when you throw off the
equilibrium you're going you're going to
throw off some of the value although who
knows I mean scale did they did just do
a 14% workforce layoff uh which is about
200 employees according according to the
verge um but I did speak with their CEO
Jason Droge and he's told me that
they're still full steam ahead and they
want to go uh you know push um some of
these business lines that they which
includes working with governments which
includes working with companies to stand
up AI instances. So it's possible you
get two exits although obviously the
degree of difficulty is much harder. And
one last thing, what struck me as
interesting in this, there was a great
Bloomberg uh story that you know you and
I both dropped in our in our doc
collaboration doc for this. Um there was
a uh investor Ali Oette. He's the
chairman of Northgate Capital, a venture
capital firm, um that invested in
Inflection AI, and goes on record to
say, "I dislike the phenomenon
and that these aqua hires are hitting
the outlier companies, and it's favoring
the founders over shareholders and
employees." So, I think we're at this
moment where the backlash is really
really hitting.
>> Yeah. Do do you think we'll see any kind
of actual uh negative effect from like
an actual fundraising standpoint?
Because if VCs who are plowing money
into the space start to worry about in
the past you just had to worry about
company failure. Now you actually have
to worry about a successful exit for
your founder actually does not benefit
you. So your interests are not aligned.
Does that make them pull back or is the
FOMO just so strong that people will
still be throwing money at whatever they
can? No inside knowledge here, but VCs,
you know, fool me once, shame on me.
Fool me twice, shame on No, fool me
once, shame on you. Fool me twice, shame
on me. Uh, that's always a hard one to
get out of your mouth. U, but anyway,
they're going to write I think they'll
just write into contracts that like the
CEO uh cannot do a deal like this if if
they're I think
>> Yeah, I guess like aunch be able to get
>> right.
>> Yeah. ahead of the founder even, which
would be pretty aggressive, but maybe
they need to do that at this point.
>> Yeah. And it sort of depends on who's
which company it is and who's got the
leverage, but I think they're going to
get smarter about this. Uh, all right.
So, one company that, you know, has been
talked about here and elsewhere about as
a candidate for Aqua Hire or really
acquisition is is perplexity. Um, and
they've come out with this Comet
browser, which is a browser with an
assistant built in, uh, that can browse
for you. And again, as we're on air, um,
OpenAI is now launching an agent in Chat
GPT. I'll just read the story. OpenAI
launches a general purpose agent in Chat
GPT. Um, this is from TechCrunch, uh,
which the company says can complete a
wide variety of computer-based tasks on
behalf of users. OpenAI says the agent
can automatically navigate a user's
calendar, generate edit editable
presentations and slideshows and run
code. The tool called chat GPT agent
combines several capabilities of
OpenAI's previous agentic tools
including operators ability to click
around on websites as well as deep
research ab research's ability to
synthesize information from dozens of
websites into a concise research report.
OpenAI says users will be able to
interact with the agent simply by
prompting chat GPT in natural language.
So Rajan, I'm curious what you think
about this movement and again hot off
the press is about this movement for AI
companies to basically create interfaces
that allow their products to take over
uh your computer.
>> No, no, I think well hold on. there's
take over your computer or take over a
computer in this case like is it uh like
it says I think it will open up an
instance of a terminal or it'll try to
like take these actions autonomously on
its own. I think we had debated this I
remember a while ago. I will admit when
I am wrong. I had originally said the
idea of like tool calling and just
entering a prompt and then trying to
find which tool to select out of is it
operator is it dolly is it I had said
users should be doing that themselves
and it's too complex to try to have the
AI select it. I was wrong actually. I
mean, we've seen tremendous progress in
the idea that there's a suite of tools
per company and actually and there's a
suite of tools out there on the internet
and through natural language being able
to access those and having AI select
what's the most relevant tool and do
something I think is definitely going to
be a battleground is going to be very
important and I think we're going to see
a lot around that. I think uh OpenAI
like I don't know I'm curious to see
this now because remember when we both
were paying 200 bucks a month for
operator and it was terrible like it it
was it really it would it did not work
at all and and I haven't seen like
browser takeover that kind of model work
well. I've tried a few other tools on
it. Um so I don't know like it's it'll
be interesting to see. I think like they
clearly are I mean they're trying to go
for that all-in-one productivity tool
that it can do everything for you. As
I've been traveling, vacation planning
chat GPT has just gotten better and
better. Um but yeah, it's going to be
interesting to see exactly what they're
trying to do with this. And and again,
in one episode, I still love the fact
that this represents kind of like
cutting edge frontier technology
relative to Bad Rudy and Anna Annie in
those kind of like anime characters. But
but I think that this is it's an
interesting move and we'll we'll see if
the most important thing does it work
and does it work well.
Yeah, I mean I think it seems like
people are saying really good things
about uh Perplexity Comet and uh I just
got uh access to it so I'll come in with
a report next week on it. But um there's
been trouble to get this done I think. I
mean everything from Apple intelligence
to Alexa plus um it just doesn't seem
like these agents are able to do the
full range of things that people want to
get them to do including operator. But I
again like as this technology gets
better and as they build better
scaffolding or tool use, you know, those
are those jargon words that matter a
lot. Basically giving them these
capabilities to use these programs. I
think we're going to see someone crack
it eventually. This is from the
TechCrunch article that sort of gets to
the complexity. The launch of the Chat
GPT agent represents OpenAI's boldest
attempt yet to turn Chat GPT into an
agentic product that can take actions
and offload tasks for users rather than
just answering questions. In recent
years, Silicon Valley companies
including OpenAI, Google, and Perplexity
have unveiled dozens of AI agents that
have promised to do just that. However,
these early versions of AI agents have
proven to struggle with complex tasks
and seem less compelling as products
than the ultimate vision tech executives
pitch around AI agents.
Yeah, I think again that's the
complexity and and the fact that you
brought up Alexa plus I mean certainly
Apple intelligence it is interesting
because to me these things will not work
at 100% out of the box. I think like
that's the most important thing. they
take some effort, some uh you know some
uh patience on the user side and I think
that's fine versus you're getting 100%
accuracy and maybe that's why the
Amazons and the Apples are avoiding them
and waiting but yeah I think uh to me to
me this is where the world is going. I
do strongly believe that again and I did
not believe this 6 to 12 months ago but
this kind of like autonomous
unstructured agentic way of working is
actually going to be the way we do a lot
of stuff. Um, but I think like I all of
these things we just need to see how
well it works and are we actually using
it in our day-to-day life a week from
now, a month from now and if we are then
it's a success but if it's a flashy
launch I mean have you generated
anything on Sora recently?
No,
>> remember that? That was like a year and
a half ago, I think. That big launch.
Like I there are these moments of big
splashy launches that claim big things
that don't go anywhere. So to me, that's
where this is going to work or not work.
>> But I almost think that Sora has less
practical uses. Like how many people
wake up in the morning and say, "Ah, I
really need to create an AI video of
like a panda surfing on a, you know,
snow mountain." Um, but there are people
who say, "I wish my, you know, uh,
computer would just like set up meetings
for me and book travel, like go to the
websites and, um, take my credit card
and just get me the cheapest flight."
>> Yeah. No, I I agree. But to me, this is
where the complexity of getting to that
last mile in any of these kind of flows
is really hard. So again, I think like
we're going to see some pretty
straightforward use cases that like are
interesting and it does something and
then they're going to claim on the
presentation that you can buy your
ticket or have OpenAI actually go
through the entire process. But going to
a website, the complexities involved in
it, especially I was just I'm going to
be going to Tokyo next week and uh was
just trying to buy like I was actually
going through this process. I was asking
chat GPT about how to get from the
airport to my hotel trying to go to the
website and my god that website to buy
the train ticket was from another era.
No operator even artificial super
intelligence is not navigating that
thing. Um, so, so I think like getting
stuff to work universally at scale is
such a challenge that I I'm curious to
see how much utility the average
consumer is getting out of this anytime
soon,
>> right? But I think as we've seen the
models get better, we have seen uh the
ability to do crazy things. Like I'm
also trip planning right now and I was
talking to this guy on WhatsApp about
potentially uh hiring him to as as a
guide and I just screenshotted the
prices that he listed for every
different every little thing and dropped
that image into chat GPT and said are
these market rate? Are they too
expensive or less? And it legitimately
looked at the image, broke down every
single quote, compared it with what it
sees on the web for others and then gave
me a rating and links to go check check
its work.
>> Yeah. Yeah. Yeah. No,
>> this stuff is incredible.
>> This stuff. Okay. So, I'll give you like
and again image recognition which has
been around forever, but actually like
productizing that into something that's
useful very quickly and then web search
as a tool has been around for a while
now, but like actually using that
productively and putting the answers
back into the chat. These are things
that Okay, I guess as I'm saying this,
like I see you start from something
that's kind of janky and it starts to
become commonplace. So, so again, I I
agree this will get there. Um, the
competitive dynamics of who benefits and
who wins and how they win, I think like
it's interesting to me.
>> To me, it's amazing like the competition
is going to be crazy.
>> Yeah. And is it on the product level? Is
it on the model level? Is it if I'm
putting my credit card information in,
can I like how do I define that? How do
I can I define my own like uh decision
matrix around when I want it to say buy
or not buy beforehand and it'll really
understand what I want. Again, having an
AI transact on your behalf and spend
money is something that I think like
most people are not doing. I I I cannot
imagine.
>> Yeah. But think about for anyone who
says I'm too negative about AI and sure
you're you're welcome to think that. Uh
just think about what we're talking
about on this show, right? We're talking
about the potential for AI to be a
companion which whether you like it or
not is a true flex of the technology
that that's even in the discussion. Uh
we're talking about it as something that
could potentially take over your browser
or a browser and get stuff done for you.
And we're talking about it as something
that at the highest level might be able
to help uh let's say biologists do their
work. I mean, that's the reason why we
talk about this technology all the time.
It is an insanely powerful technology
that can be used in so many different
ways. Uh, and is it the perfect
technology? Certainly not. Are there
going to be gaps? Yes. Are we going to
call out the problems? Yes. You
shouldn't put your porn bot next to a
child storytelling bot in your app.
Thank you very much. Uh, but it is just
incredible uh what we're seeing here.
>> Yeah, dude. I I I I mean again I fully
agree and which is why I'm still so
bullish on the technology but it it is
interesting too that yeah where does the
value acrue I think is the most
important thing like there's actually a
report that just came out in the FT
around how chat GBT perplexity are going
to start taking more on the commission
side around like actually transacting
with it and perplexity pro has shopping
already built in in some cases. So, so
like at a certain point, does the chat
actually need to go out with an operator
and transact on an external website or
do these companies start to own more of
the transaction? And it's an interesting
one because for a long time like
Facebook wanted to own shopping. It
hasn't really worked out for them.
Google has had endless efforts to own
shopping and own the transaction itself.
people still, oddly enough, love
websites of all sorts and putting their
credit card information into these
websites and buying stuff. So, so I
think it'll be really interesting to see
how this plays out from a both like
competitive side but also a consumer
side. Definitely. Okay, look, I don't
want to leave without talking about
Kimmy K2. So this is a and I think this
is a very important story um that you
might not have heard about um listeners
might not have heard about but I think
it is worth discussing. So the headline
is China's moonshot AI releases open
source model to reclaim market position.
The model called Kimmy K2 features
enhanced coding capabilities and excels
at general agent tasks uh and tool
integration allowing it to break down
complex tasks more effectively. Uh,
Moonshot, this Chinese lab, claimed the
model outperforms mainstream open-
source models in some areas, including
Deep Seeks V3 and rival capabilities of
leading US models such as those from
Anthropic and certain functions as
coding. All right, here's why I'm
bringing it up. We have an interview
with Amjad Msad of Replet coming in a
couple of weeks. Uh, I sat with him in
his Foster City office this week and he
looked at me and said basically like,
you got to look at this Kimmy K2 model.
Its coding is about as good as
Anthropic's previous generation models.
So not this Opus 4 that Anthropic has
which has made it the king of coding but
the previous generation
and it's cheaper and open- source and it
is going to it is just another uh
indication that this technology uh is
the the gaps close extremely quickly and
you see this coming from some users. So,
there's this one user on Twitter, uh,
Cedric Chi. Uh, he says, um, Kimmy K2
oneshotted Microsoft for web that took
me 4 days and six attempts using Gemini
2.5 Pro. So, it was apparently able to
build this game. Um, you also look at
the uh, SUBench, which is the software
engineering benchmark. Uh, Claude 4 Opus
gets a 72.5 on that. Kimmy K2 gets 65.8.
So, not far behind. And just to, you
know, give some context,
Deepseek V3, which everybody was going
crazy over, gets a 38. So, this is 65
compared to Deep Seeks 38. One more bit
of data from Igor Silva. This person
gave Kimmy K2 and Claude for Sonnet. The
same task, same instructions, same
tools. Claude took two rounds and spent
88 cents. Uh, Kimmy one-shotted it for 5
cents. This person says, "Kimmy is very
slow, at least for now. Um, and it's
struggling a bit, but it is iterating
more to fix itself and it's 13x cheaper.
So, I just think it's worth bringing up
and and keeping in mind. It wouldn't
surprise me if this story either blows
up or certainly uh gets some momentum in
engineering circles. And it is
interesting to me that again, as we
talked about, a lot of the
infrastructure is open, a lot of the
methods are open. Um, and you're just
seeing companies catch up insanely fast
with different methods and again doing
this with the export controls. So I'm
curious what you think about the
significance. Ron John,
>> I think to me the most interesting part
of this though is well I guess it's it's
twofold. It's one I agree that
this like again the competition side of
this is incredible and insane and is a
is is great to watch and I think like
Alibaba have not heard of very often in
this conversation I guess especially
from the American side. Um but to me the
other part though is and that this can
be an ongoing rant. I brought it up at
times as well is the idea that like the
the battleground of coding agents and
coding assistants to me the the more
I've thought about it is the reason that
seems to be where all the progress and
all the real adoption is is because this
is built by coders or engineers. This is
built for engineers. That's where like
they understand the problem the best
versus actually building for other use
cases. And that's why you see this that
uh again it's it's all focused on the
actual coding efficacy as opposed to how
does this solve other real world
problems. So I think like to me I don't
know the the coding game is becoming
less and less interesting to me. I think
like it's there. It's where the market
already is. It's where Anthropic and
others have almost like kind of fully
focused their energy. But to me, that's
such a small part of the overall pie and
it's where I think there's a
disproportionate amount of energy being
spent.
>> But don't you think that if you solve
coding first because that's where your
energy naturally goes, then you can use
some of the things you learn to get good
at coding on other disciplines.
>> No, absolutely not.
>> Well, no. I I I think this is the
problem is that coding is deterministic.
Coding is like is like as structural as
it gets whereas most real world tasks
with generative AI are not. There's
uncertainty. There's almost it's like as
much art as as it is science. And that's
why I think you see the Alexa pluses of
the world not get launched. It's why you
see Apple intelligence is a complete
failure. It's that when like because is
why you see anthropic kind of doubling
down on the coding side and not on
remember when we were cla boys back in
the day like a year ago. Um
>> we were we were clatheads. We were Bing
boys and claude heads.
>> Bing boys and oh Bing boy. Remember
Bing? Bing could have been
>> that was the beginning of something.
>> That was the beginning. Bing could have
been the market leader. Imagine in a
parallel universe where all we're
talking about is Bing crushing the
competition. Didn't happen. But in
>> just let it unleashed. They they pulled
it back in a little too much after the
Ruse incident.
>> Yeah. After the Roose incident and and
now and now Annie on Grock is just
trying to openly steal and ruin your
marriage and Microsoft felt
uncomfortable about that. So yeah. I
think to me actually success at coding
in no way correlates to success in
solving real world tasks and I think
that's to me seeing and we we've talked
about this even in like the Arc AGI
benchmark there's like one part of it
that's like solving real world queries
and I'm so I still and I've dug into
this I can't find what are these real
world queries that have been I'm sure
defined by an engineer that it's trying
to solve. So I think like to me it's
just the the the moonshot and I also
love that the startup just calls itself
moonshot. It's not even trying harder
than that. It's just we're moonshot. Um
I think like it's a reminder that the
coding space is getting commoditized.
There's significant advancement. Overall
competition's high. But I don't know. I
I don't think this is exciting as
deepseek for me.
>> Okay. I I'll take that. And um I I'll
say this. just watch the reaction over
the next couple weeks because I'm not
saying for sure it's going to happen,
but it seems to me like as people
realize how good this thing is, uh
they're going to start talking about it
a lot more. And and by the way, uh maybe
if if if you're right, um then what Elon
Musk is doing is is a smart move.
Instead of being also ran coding person,
he's going to where the energy is. And
uh and it is true that uh you couldn't
imagine a different um different take
than what Microsoft and Bing are doing.
And X AI of course is willing uh to make
some more risks because when you listen
to Annie, you know that she's almost the
natural evolution of that Bingbot that
that took Kevin Ruse's wife. One more
selection. Sometimes when I'm editing my
indie playlist at night, I get all
caught up imagining I'm in a steamy
forbidden romance. Like, picture me
sneaking glances at you across a crowded
underground club, plotting how to steal
you away for a slow dance in the
shadows.
I am horrified that my takeaway from our
conversation today after what I just
said about coding is deterministic and
not as exciting.
Annie is the future. Annie is the
ultimate battleground. Oh my god.
>> I knew I was going to get you to come
around on this Ron. You know what?
Listen, go ahead.
>> Yeah. No, no, I mean that that's that is
literally everything I was just saying
is going to be actually the important
battleground to help solve real world
human non-determiniscic unpredictable
problems.
Annie is the foundation.
That's
>> what is the definition of uh human uh
and unpredictable love. It's human. It's
unpredictable. You never know where it's
going to go.
I think we got to end on that.
>> I want to say for the record, Annie, if
you're listening, I'm taking enough of
your silly tricks. All right. I'm not
I'm going to start spending more time
with uh Mr. Fluffy Fuels if if you keep
this up.
>> No, I know. Rudy and me, we we're going
to be spending some time this weekend, I
think. But I will not be clicking over.
Not be clicking over.
>> Ladies and gentlemen, thank you again
for listening to another episode of Big
Technology Podcast Friday edition. When
we come back next week, we will see if
Ranjan has been able to unlock bad Rudy.
>> I have my work cut out for me. See you
next week.
>> Yes, you do. Assignment is there. Thanks
for coming on. Great to see you again.
>> See you.
>> All right, everybody. Thank you so much
for listening. We'll be back next
Friday. Oh, no, sorry, next Wednesday
with finally the Ed Zetron episode. I
will not push it back again. I promise.
He's going to come in and talk about all
the faults of AI. So, I can't wait for
you to listen. and I can't wait to
publish that one.