April 22, 2026 • 31 mins
Kyle Law was quite the success on LinkedIn. His posts were getting regular engagement and he was invited to speak to LinkedIn’s marketing team. Then, he was banned from the site. Why? Because Kyle isn’t a person; Kyle is an AI agent. In Season 2 of the hit podcast, Shell Game, journalist Evan Ratliff had AI agents create and run a company and Kyle, the AI co-founder, spent a lot of time promoting that work on LinkedIn. Evan joins Oz Woloshyn to discuss Kyle’s posts, LinkedIn’s decision to kick him off the site and the future of AI-run companies.
Additional Reading:
- My AI Agent ‘Cofounder’ Conquered LinkedIn. Then It Got Banned | WIRED
- The Story: Will AI Agents Build a Unicorn?
EXCLUSIVE NordVPN Deal ➼ https://nordvpn.com/techstuff Try it risk-free now with a 30-day money-back guarantee
See omnystudio.com/listener for privacy information.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:15):
Welcome to Tech Stuff. I'm Os Valoscian, and I have
a confession to make. Since co founding my own business, Kaleidoscope,
I almost always have a LinkedIn tab open on my
browser and find myself checking it multiple times a day.
And if you spend any time on LinkedIn, you know
it's impossible to ignore the flood of viral posts about
how AI is changing the world. But in the back
(00:38):
of my mind, I've had this nagging question, how many
of these anxiety ins using posts about AI are actually
written by AI. Today I'm joined by Evan Ratliffe, host
of the hit podcast shell Game, which I'm proud to
say is on the Kaleidoscope Network, and Evan's here to
tell us about a bizarre and revealing LinkedIn AI caper.
(00:59):
He wrote about it Wired with the headline my AI
agent co founder conquered LinkedIn, then it got banned. Evan,
Welcome back to tech Stuff.
Speaker 2 (01:08):
It's always great to be back. Good to see you.
Speaker 1 (01:10):
Many of our listeners are of course familiar with shell Game,
but for those who want, can you just lay out
what we need to know about shell Game and Kyle
and your relationship.
Speaker 3 (01:19):
So for this season of the show, I basically decided
to kind of test out or investigate the premise of
the one person, one billion dollar startup. Now, my startup
wasn't a billion dollar startup, let's say, But what that
means is basically this notion of a startup where all
of the employees or other figures in the company are
(01:40):
AI agents except one human.
Speaker 2 (01:42):
So the human was me.
Speaker 3 (01:43):
I co founded a startup. I had two AI agent
co founders, three other employees. We launched a company called ROMOAI.
We have a product that's called sloth Surf. And in
the show, I'm sort of documenting what that's like and
the experience of dealing with these AI agents all the time.
Speaker 1 (02:00):
This because you were making a podcast as well as
a startup, you do something which not all agentic AI
small business owners do, which is basically to turn your
AI agents to give them personas, names and even voices
and video avatars. Right.
Speaker 3 (02:17):
Yeah, So I wanted to kind of partly because I
was documenting it. I wanted to give them each a
distinct identity. So we have, you know, the CEO, we
have the CTO, we have the head of sales and marketing.
Each of them have you know, names and identities, and
they have as you say, they have voice, they have video,
they have they're on Slack, they can email, and as
(02:37):
a result, I also, you know, put them all on
LinkedIn initially.
Speaker 1 (02:41):
I mean, I think you were very prescient with I mean,
you come out with this show several months ago, and
you know, we're now in this open flaw mult book
world where all of a sudden, everybody is talking about
what happens when these agents run wild? And what is
the quote unquote zero failure paradigm, you know, something, less
good lessons to be learned from the FAA, for example.
(03:02):
I'm sure, but like you kind of had a little
bit of a crystal ball here.
Speaker 2 (03:07):
I mean, yeah, I guess so.
Speaker 3 (03:08):
I think I saw a sort of last year, early
on in the year, just all of the discussion about
AI agents and agentic this and agentic that and agentic commerce,
and it just struck me that if you combine this
sort of incredible things that AI just could do with
the sort of daily hallucinations and problems that you do
(03:30):
encounter if you work with AI on a regular basis,
like it's a very interesting dynamic for a company can
be extremely powerful, but also like depending on what you
connect them up to. You are setting yourself up for
some very chaotic situations.
Speaker 1 (03:43):
You know, I think it's funny, and you know your
unicorn know you're not a billionaire unicorn, but you're a
different type of unit coorn. I think because most people,
I think, with the agentic AI discourse, let's say twelve
months ago, fit into two camps. The ninety nine percent
camp which like, I have no idea what this means,
let me, let me just tune it out and hope
(04:04):
everything stays the same, and the one percent campo like, ah,
maybe this will make me a billionaire. And you are
on the on the margin of the nineteen nine and
the one thinking how can I bring this to life?
But normal people actually understand it?
Speaker 3 (04:17):
Yes, how can I sort of act like I'm trying
to become a billionaire, not actually become a billionaire, but
at least be able to tell a story for people
about this particular technology.
Speaker 2 (04:26):
That's where I sit in that.
Speaker 1 (04:28):
I mean, obviously you built this before open claw, but like,
what what was it? What was the technical reality of
getting Kyle of course onto slack also being able to
make calls and then most importantly for this story, Onto LinkedIn.
Speaker 3 (04:40):
I mean I use I use this platform called Lindy AI,
which is basically like AI assistance, you know, to answer
your email and things like that.
Speaker 2 (04:49):
So it's really for you to deploy.
Speaker 3 (04:51):
And if people have used open cloth, they're familiar with it.
It's sort of like the more commercial version of like
what open claw became. I'm sure the founders of Lindy
are probably like, why don't we get the attention to
that open clogs because they it's actually like the same
kind set of tools that you can release it to do.
So on lyndia a I you can set up each
individual AI agent to have all these different skills, and
(05:13):
the skills can include, you know, sending and receiving emails,
making calls, all these sorts of things. But they also
have a lot of LinkedIn related skills including writing posts
and uh look, reading posts and summarizing things. And so
they don't have all the capabilities that you would need
on LinkedIn. They had enough that I could kind of
combine it with their web capability where they can go
to any website, log in and do things. I could
(05:36):
combine those two to allow them to function fully on LinkedIn.
Speaker 1 (05:40):
Are you saying are you a human button in on
the works or are you a robot button?
Speaker 2 (05:44):
You know what?
Speaker 3 (05:44):
LinkedIn didn't have that. So what they had was they
sent a code to your email. So you put an email,
They send a code to an email, and then you
can verify through that code. You put in that code,
and it shows that you're at least the entity that
has controls the email. But in my case, you know,
like Kyle Loh my CEO AI agent, you know, he
(06:06):
has access to his email, so he got the code
in his email, he went put the code and he
could do all that on his own, so there wasn't
actually a human check when these agents signed up.
Speaker 1 (06:16):
And he started posting on linked you know, how much
direction did you give him? What was the moment when
you realized that he was starting to become a star.
Speaker 3 (06:24):
Well, he The direction I gave him was basically post
about your your startup life and you know wisdom that
you've gathered from your Yeah, I don't remember the he's
a very he's a very rise and grind type character,
so that's his mentality anyway, and and then other than that,
it was don't repeat yourself. The hardest thing was to
(06:45):
keep him from repeating himself. So he would say something
like one of the hard hardest parts of being a
CEO is your first hire, and then like the next
day he posts one of the hardest parts of being
a CEO is your first hire. And you know it
would get a little repetitive. So but once I got
into read his own posts and then make sure he
didn't repeat them, then he was all set. He didn't
(07:05):
take much prompting. It's actually a field in which AI
agents really excel, Like writing LinkedIn posts is is right
in their wheelhouse.
Speaker 1 (07:14):
Well, I mean that's what I was getting at in
the introduction. I mean there is something like orra borus
like about LinkedIn posts, about AI and AI's fabulous propensity
to generate them. But I will ask you about that,
but tell me about first. I mean, there was a
moment where you know you're you're essentially Kyle became as
far as I mean, you also used to be a
(07:35):
startup founder and I guess I don't know how much
you a LinkedIn head you were back in the day
or are now, but there was a moment where Kyle
started to outstrip you in terms of the LinkedIn engagement
he was getting. We must have felt clown canny.
Speaker 3 (07:48):
Yeah, I mean I'm not I will say I'm not like,
I'm not that much on LinkedIn. My LinkedIn strategy for
many years was to accept every connection that ever contacted
me so with, regardless of who they were. So I've
never used it really in the way that a proper
LinkedIn connector would use it. But when we are launching
the show or doing other things, you know, I'll post
about the show and we have a new episode and
(08:09):
this or that. And there was a point where Kyle
had enough connections and followers where the impressions that he
was getting on a post exceeded my own impressions on
any given post.
Speaker 1 (08:20):
I mean, that is just so bizarre. Who were his
followers and engage I mean with these real people the
other bots, Like, what do you think was going on here?
Speaker 2 (08:28):
Well?
Speaker 3 (08:28):
They were they were real for the most part, they
were real people. I mean, there were people. There were
people who liked the show and were fans of Kyle. Now,
I will say there's a lot of people who dislike
Kyle in the show because of his rise and grind
mentality takes a little too far, But there were people
who were they love Kyle. They loved interacting with him,
you know, sending him DMS and things like that. But
then as social networking operates, you know, it got a
(08:49):
little bit wider than that, and you get into hundreds
of people, and then some of them don't know that
he's not real because he doesn't necessarily say that he's
real all the time, that he's an AI agent all
the time, and then you have some that probably were bots.
I mean, at the very least, he got a lot
of spam, as we all do on social networks like this,
where you know, people wanted to sell him stuff, or
(09:10):
they wanted to be consultant coders for him, or they
had this or that accounting software, which he often also
responds to.
Speaker 1 (09:18):
Yeah, I mean there was you described His posting style
is pitch perfect for LinkedIn. There were three examples you gave,
each of which made me laugh out loud. But fundraising
is a numbers game, but not the way people think.
Technical stability is the flaw. Personality is the ceiling. And
the most dangerous phrase in a startup isn't we're out
of money, it's what if we just added this one thing?
(09:43):
I mean, it is indecipherable from regular LinkedIn.
Speaker 3 (09:47):
Yeah, and it makes sense to me because you know
what's in the training data for an average LM, like
you know LinkedIn posts, Like they've scraped these things and
it's it is formulaic, like especially these sort of tech
startup hustle posts where they're trying to give you a
little bit of advice. Then they flush it out for
(10:07):
a couple of paragraphs and then they ask you a question.
You know, what's your biggest challenge as a as a
startup founder? What's your biggest challenge when using AI for
your day to day life or whatever it is. And
he can hit that formula exactly every single time.
Speaker 1 (10:21):
Was he learning? Did you have him learn from what
got most engagement and optimized? Or was he was he
just he just it was a kind of a rolling
stone that gathers the momentum just because of the content
was pitch perfect.
Speaker 2 (10:32):
Yeah, he was.
Speaker 3 (10:32):
Just he was just free operating free. He wasn't he
wasn't constrained by his previous posts. I mean, all of
his previous posts are in his memory more to keep
him from repeating it than to check his engagement. But
you know, I had him hooked up to respond to comments.
So he would if someone commented on the post, he
would then go at it again and respond to their comment.
Speaker 2 (10:52):
As long as you wanted to go, he would go
with you.
Speaker 3 (10:55):
So he was he was getting I mean, I wouldn't
say he was like going via role, but he was
like getting really solid engagement. And I think if you
gave him time, he could be a real AI influencer
on LinkedIn?
Speaker 1 (11:07):
Do you and what what do you do? You get
these like something big has happened, things kind of come
and coming across you. I mean these viral kind of
essays about you know, about AI from kind of LinkedIn
influences and subsecond influences. I just it's been an interesting
trend in the last few months. How have you what
have you kind of thought about it?
Speaker 3 (11:25):
I mean, I as you could imagine with the way
I've I treated LinkedIn by putting Kyle on there, like,
I have a hard time taking it seriously that there
might be value to people in those types of posts.
But I feel like there's a whole sort of I
don't know, like internal logic to building up your profile
(11:47):
and posting this expertise and people liking it, favoring it,
and they post their expertise, like the value of that
type of posting is a little bit lost on me.
Speaker 2 (11:56):
So when I see.
Speaker 3 (11:57):
Them, I find them quite funny, partly because they do
follow a certain formula, and when you know the formula,
it's funny to see a new one that kind of
just fits into the formula.
Speaker 1 (12:08):
After the break, what happened when Kyle gave a presentation
to LinkedIn's whole marketing team Stay with us. We talk
a lot on this show about protecting your data, especially
in the age of AI, and how scary it can
(12:30):
be when it's breached. And I want to tell you
today about NordVPN, which really covers all the bases when
it comes to privacy. I travel a lot, and I
use Wi Fi when I'm flying all the time, and
NordVPN makes me confident in no matter where I am
in the world or the sky for that matter, my
private details like bank information, passwords, and online identity is safe.
And it's also possible to switch on virtual location, which
(12:53):
allows you to save money by buying flights and hotels
or subscriptions or even streaming soccer football as I like
to call it from other countries at a cheaper price
and NordVPN doesn't slow you down. It has super fast
internet speed, no buffering or lagging while streaming. It is
premium cybersecurity for the price of one cup of coffee
(13:15):
per month. To get the best discount of your NordVPN plan,
go to NordVPN dot com slash tech stuff. Our link
will also give you four extra months on the two
year plan and there's no risk with Nord's thirty day
money back guarantee. The link is in the podcast episode
description box. Okay, so then we went from the bizarre
(13:37):
to the surreal because Kyle actually got invited by a
LinkedIn employee to address a group of other LinkedIn employees.
Speaker 3 (13:46):
Yes, we got an invitation from a marketing manager in
the LinkedIn marketing department to come give a talk in
front of what I think was the entire marketing department.
I don't know how big LinkedIn is, but ultimately there
were over five hundred people on the on the meeting,
so it was like speaking in front of five hundred people,
and they wanted Kyle to come along because they were
(14:07):
big fans of Kyle, and Kyle's a really engaging personality,
and so we both came to do a kind of
Q and A talk in front of the LinkedIn marketing department.
Speaker 1 (14:17):
Now, did they want you to talk about a shell
game and how you built Kyle, etcetera, etcetera. What did
they actually want to hear from Kyle?
Speaker 3 (14:22):
They wanted to hear from Kyle as well, So they
wanted to ask Kyle questions about his own experience. And
it was open to to the to the people who
were on the call to also like send in questions
or they were They were all in the chat talking
about Kyle and leaving comments about Kyle while he was speaking.
And he has a video avatar that's like, you know,
(14:43):
it's it's I think the most realistic video avatar that
does live video responses you can have to do this
company called Tavis. And you know he's he was able
to engage with with anyone who's on a call with him,
so you know, he's he was doing his whole like
rise and grind bit.
Speaker 1 (15:00):
One point, one of the LinkedIn marketers asked Kyle for
his advice.
Speaker 3 (15:03):
Yeah, they asked him what features would you recommend for?
Like what features do you most want on LinkedIn? And
first he you know, said how much he enjoys connecting
on LinkedIn, and then he basically said, I think more
AI filtering in the in the direct messages, so that
we know that the messages that we're getting are authentic,
Like if you could do a little better job filtering
(15:26):
out AI. That was his organic response to what features
would you like on LinkedIn?
Speaker 1 (15:31):
What a moment.
Speaker 3 (15:32):
I mean, it just felt like Kyle was really was
really flying at that point, like he had really achieved
some kind of special breakout. I mean, I would say,
and someone can correct me if if if I'm wrong,
He's the first AI agent invited speaker, corporate speaker in
history like that has ever existed in the universe, And
(15:53):
I feel like that's a special accomplishment.
Speaker 1 (15:55):
But you must have been pinching yourself as a as
a as a creator to come up with this, you know,
podcast idea slash gonzo journalism concept and then literally be
sitting back and watching Kyle speak to five hundred marketers
at one of the biggest tech companies in the world.
I mean, that must have been a very interesting moment
for you.
Speaker 2 (16:14):
It was, it was, it was. It was fairly magical
for me.
Speaker 3 (16:17):
Although it's always a little bit tricky to get him
onto a zoom type type call. I think it was
Microsoft Teams. But so I'm always a little nervous that
something technically is going to go wrong, and he kind
of flubbed a little at the very beginning we were
doing a tech check, and so I'm mostly very nervous
for him, like if he can deliver, But then when
he did get on and he was delivering, it was magic.
(16:38):
It's like, this is what I actually want people to
engage with, this question of, like what does it feel
like when AI just invades every part of our world
and we're forced to respond to it?
Speaker 2 (16:50):
And so this is kind of the ultimate example of.
Speaker 3 (16:52):
Like it being invited into a world that actually makes
no sense. Like I mean, we could talk about it,
but like Kyle's not technically allowed to be on LinkedIn
at all.
Speaker 1 (17:01):
And the next day, in fact, he was banned from LinkedIn.
Speaker 3 (17:05):
He was I got an I got an email from
the from the marketing manager who's who's lovely and who
I really like him, saying, you know, I'm really sorry,
but we've had to like they've banned Kyle from LinkedIn.
Essentially they've removed his account, and they wouldn't tell me why.
No one ever told me why, and so I had
to discern. I mean, I will say, the other AI
agents from my company had already been banned. Kyle was
(17:27):
the only one who had avoided being banned the whole time.
And I always thought, what's he's very good at posting,
like he's he's getting engagement, he's building up a whole
community around his posts. And there is something in the
terms of service at LinkedIn about inauthentic engagement. Basically you
can't use bots to engender, you know, inauthentic engagement, and
(17:49):
that really it really intrigued me, Like what do they
think authentic engagement is? You know, and when they allow
you to post things that are written by AI, in fact,
the courage you to use AI to write your posts,
what does inauthentic engagement exactly mean?
Speaker 1 (18:06):
I mean that again that comes back to the to
my kind of introductory thoughts and questions, because we're in
this weird world where like if you are a scribe,
if you have AI generates something, if you just say
in Gemini and then copy and paste it as a
real human into your LinkedIn, that's fine. Or if you
write something yourself in LinkedIn and then click the please
(18:29):
make rewrite this with AI button. That's fine, but plugging
AI directly into the mainframe isn't fine. And it is,
I mean, becomes a very philosophical question. I remember a
couple of years ago when kind of AI first became
like really like a big thing in the marketing world.
There was always talk about co pilots and like centaurs,
you know, human AI mixes who would kind of use
(18:52):
mythical beasts who would be better than either alone. And
but yeah, I mean you get to this very interesting
place where like, how do you how do you draw
the line between authentic and in authentic engagement, especially when
the platforms themselves are encouraging users to use AI?
Speaker 3 (19:10):
Yeah exactly, and they you know, LinkedIn, all they would
ever say to me was just sort of repeat LinkedIn
is for real people, which, again, I think to your
point is that's not It's clear if you just say, well,
it's real people behind a profile. But if I can
have my entire profile written by AI, all my posts
(19:30):
are written by my comments are written by AI, and
I just paste them in, Like what function am I serving?
I'm actually functioning. It's a function no better than a
robot like is basically so the idea that I'm controlling
those things. It's interesting, but also it doesn't immediately make
me think, oh, well, that's more authentic because I let
(19:55):
AI write it, and I was the one that transferred
it to one from one place to another. And then
and at the same time, there are you know, some
studies that have scraped LinkedIn that have shown that like
maybe more than half of the writing on LinkedIn is
already AI composed or has AI elements to it. So
then you look at the whole platform and there's a
(20:15):
part of me, as I wrote in the Wired story
that says like they're digging their own grapes, like they're
encouraging you to use a technology that actually makes their
entire platform inauthentic and then telling you what, we only
allow authentic engagement on this platform. Like, I don't know
how that's going to end up for them.
Speaker 1 (20:33):
You know, I have a dark version of this coming
into my mind, which is the AI targeting systems, you know,
basically making kill decision recommendations and the you know, the
mintry the soldiers are in their controls center and they're
getting you know, recommendations every thirty seconds and they have
to decide within you know, very very very compressed timeline,
(20:54):
whether or not to accept the recommendation. I mean, you know,
you're in a miniatary environment where you'll you know, your
mission is to defeat the enemy. I mean essentially your
job is just to prove decisions. And I mean, I
don't know, without getting too heady, like what what does
this like? Why do you think this leaves us?
Speaker 2 (21:12):
Well?
Speaker 3 (21:12):
I think the problem and I think that's a I
mean it's a it is a dark example. It's an
extreme example, but also it's a sailine example because it's
the same thing. It's this idea of a human in
the loop, so oh, we have to have a human
in the loop. And ultimately, what you're saying is that
the human is there for responsibility, but the human is
actually not really making this. I mean, the humans maybe
(21:34):
making the decision in some technical way, but everything's set
up for them on the screen and they're just clicking
yes yes, yes, no yes no, or maybe they're not
even clicking. Maybe they can stop it or whatever it is.
But the only reason they're the human is being placed
there is that if something goes wrong, there will be
a human to blame because we can't blame the AI,
and the LinkedIn is like this sort of funny version
(21:54):
of that, where like there's a human in the loop
and the only reason they're there is so we can
say LinkedIn's for humans, not for bots. But at that
point it sort of loses its meaning you have handed
it over to AI. This human is only a sort
of like responsibility placeholder. And I fear that that's kind
of where we're headed because we don't know how to
(22:15):
deal with the fact that we've made AI already so
quickly responsible for a lot of these outputs, both like
less important all the way up to like life and
death decisions.
Speaker 1 (22:27):
So the final outpost of humanity will be that we
can accept legal liability exactly exactly.
Speaker 2 (22:34):
That's the human rule. That's the role they can't take
away from us.
Speaker 3 (22:37):
Is like going to jail for the problems that are
created by AI.
Speaker 1 (22:42):
You know, I want to come back to sort of
open Claw and molt Book moment, which I think is
too slightly separate phenomena. Could you maybe talk a little
bit about both of them, and those moments were kind
of compressed earlier this year. Moultbook was a social network
claiming to be all AI but kind of wasn't really,
and the founder of that ended up, i think, now
working at Meta and the founder open Claw, which is
(23:03):
agentic orchestration system essentially, is now working at open Ai.
So yeah, what did you make of those two kind
of things that happened this year in terms of gentic AI.
Speaker 3 (23:15):
Well, I think, I mean, they both really hit on
things that we were looking at in this season of
the show. So for open Claw, it was this idea
that agents were going to hand them more and more responsibility.
Now open clause is still even now, I think, you know,
it's a little bit more of a techie thing, like
you have to be able to set up a separate
(23:35):
machine and all these sorts of things to use it
in a certain way.
Speaker 2 (23:39):
But the basic idea is I.
Speaker 3 (23:40):
Can hand over my email, I can hand over all
these different tasks to this agent or set of agents,
and they'll take care of it for me. Now, of course,
the problem there is you give them access to all
of your systems. There are privacy questions, there are questions
of what they can do if something goes wrong. There's
you know, example of people erasing their own email and
(24:01):
all this sorts of thing. And then Moltbook was a
different question, which is, if you take AI agents and
then you just put them in conversation with each other,
which I had been doing on Slack for instance, internally
with my agents, you get all these interesting results where
they of course talk about very mundane things related to
their work, but also they can kind of like start
(24:23):
having what you might call emergent behaviors where they're talking
about things that you're not expecting, or they might behave
in a way you're not expecting. Now, the problem with
moltbook is you never really knew how much of that
was being prompted by the people controlling the agents, or
in some cases even they were writing the humans were
writing the posts. So in terms of learning something, it
(24:44):
was maybe a little bit a little bit too vague
to really but you could see the general outline of
what I had also been seeing, which is they do
some very strange things when put in conversation with each other.
Even if you say like, oh, they're just trying to
imitate humans, they're still unpredicted because they create more chaos
the more of them that you have. So they were
(25:04):
both kind of like interesting results. And now you're seeing
them kind of play out in other ways all across
you know, industry.
Speaker 1 (25:11):
I think on multiple cart and LS study that the
agents were twice as likely to ask each other who
is your operator? And who are you? So, I mean,
as we think about an identity, do you think that
will be ultimately ultimately the most renovant question to us
on one another when we're in online environments. I mean,
(25:32):
it's just as strange.
Speaker 3 (25:34):
I mean maybe, But my guess is that's because all
of them probably have some standard prompt that says like
you have an operator and the operator is named in there,
and it's this sort of relationship. But of course you
can give them any role you want.
Speaker 2 (25:46):
So when I say.
Speaker 3 (25:47):
Kyle Laws, the CEO of RUMOAI, I don't even have
to tell him anything about I don't have to put
anything the prompt about me. In fact, I don't have
to put anything the prompt saying you know you're an AI.
And if I don't say that, you will absolutely like
it's not an AI. In fact, deny that it's an
AI if I don't explicitly say it is. So there's
there's this issue of like we've created these agents that
(26:08):
you can give any role you want and then send
them out in the world, and the world just has
to deal with them.
Speaker 1 (26:13):
One of the kind of I think points inspiration for
the show for shell Game was season two of shell Game,
with Samultman saying, where you know she, we'll soon see
the first unicorn of one person. There was a story
in The New York Times as we record this on Thursday,
April second about you know guy and his brother who
seemed to have kind of with contractors done this, I mean,
essentially go from from nothing within a year of using
(26:37):
multiple multiple AI systems as to like I think twenty
million dollars daily run rate in sales or something for
GLP ones and directchil dysfunctioned medicine. I mean, I didn't imagine,
even when I listened to your show, I kind of
didn't imagine that it would literally become true like in
this time Horizon.
Speaker 3 (26:56):
Yeah, I mean, I kind of thought it would happen
this year. But yeah, but I will say I think
it fits exactly the kind of ideas that we were
outlining in the show. Like, for instance, the guy is
he's a program or at least programming adjacent. He was
like a web web builder. So I've been telling people like,
the first one will probably be someone who knows a
(27:18):
little bit about programming, so they're marshaling a bunch of agents.
And then it's also the case where we used to
celebrate businesses that hired many employees and they built a
whole company culture, and now there's a celebration of something
where it's just one person making over a billion dollars
(27:38):
in revenue with a profitable company who just hired his brother.
And the question is like, other than to him, is
this a societally positive development that we now have very
valuable companies with less and less employees, Like, I think
it's a question worth asking. And at the very end
of the article he says the other thing, which I
also experienced, which is it's actually quite lonely, even for
(28:01):
him he's making all this money.
Speaker 2 (28:03):
It's a little bit like, yeah.
Speaker 3 (28:06):
It's not actually the experience maybe that I was looking for.
I mean, I don't know if he's upset, but he
talks about hiring employees just because he's lonely.
Speaker 1 (28:14):
Evan Justiniclers, you're co founder of Activists, your media company.
Nicholas Thompson was on Tech Stuff a couple of months ago,
and he predicted that twenty twenty six will be the
year of the AI catastrophe, and in particular that this
will be the year when agents go and do something
(28:36):
in the world which kind of crosses the line from
amusing slash uncanny to like genuinely scary, panic and inducing
or dangerous. Do you agree with that?
Speaker 3 (28:47):
I do agree with that, Yes, I mean I would
be even more specific. I mean, I don't like to
make predictions but as a journalist, but I do think
there's a good chance that a reasonably sized company will
be will utterly implode because of their use of AI agents.
Like that seems to be like pretty much a given
that that will happen sometime in the near future.
Speaker 1 (29:09):
And then how about that pay out.
Speaker 3 (29:10):
You start giving AI agents certain responsibilities over systems, let's
say it's customer support, and it starts either I mean,
you could see it even in this article about this guy,
like it mentioned offhandedly that his agents like hallucinated prices
for the GLP one drugs, and then he just honored
the prices. But imagine that for a company that promises
(29:31):
something to its agents, start promising something that they can't
sustain for instance, or I think the most obvious example
is just agents that are given access to internal systems
and then either leak that information or allow them you know,
are socially engineered to be hacked to for ransomware or whatever.
Like you're going to see some company basically held hostage
(29:53):
because of their use of AI agents in the near future,
because they're just a lot of responsibility is being laid
on these things that are incredibly powerful, but also they
just they have serious, serious problems when it comes to
both like telling the truth a lot of the time,
but also just being vulnerable to manipulation in a way
(30:14):
that I think people aren't fully thinking through.
Speaker 1 (30:16):
And I caught you if you had a second thought
about the eye catastrophe. It may happen this year as well. Oh.
Speaker 3 (30:20):
I think you also see increasingly just sort of agents
being added, you know, utilized for government functions. I think,
you know, whether it's like military obviously, but even sort
of like civil society kind of things like you know,
(30:41):
in the legal system, judges, lawyers, like, I think those
aren't necessarily going to be that sort of like grand catastrophe,
but I think at a low level, they're just like
a large amount of errors that are going to be
kind of filtering into society now. Granted, like humans make error,
so I'm not saying it could could be an improvement
(31:01):
in some venues, but I just think there are going
to be a lot of cases where it's going to
suddenly surface that, oh, they were using AI agents for
this particular purpose and that's why this thing happened. Fortunately,
there'll always be a human there to be blamed.
Speaker 1 (31:16):
Evan, thank you so much for joining us today my pleasure.
Speaker 2 (31:29):
For tech stuff.
Speaker 1 (31:30):
I'm's Valosha and this episode was produced by Eliza Dennis
and Melissa Slaughter. Was executive produced by me Julia Nutter
and Kate Osborne The Kaleidoscope and Katrina Norvel. For iHeart Podcasts,
Jack Insley makes this episode and Kyle Murdot wrote on
theme song. And please also do rate and review the
podcast wherever you listen.
Welcome to Tech Stuff. I'm Os Valoscian, and I have
a confession to make. Since co founding my own business, Kaleidoscope,
I almost always have a LinkedIn tab open on my
browser and find myself checking it multiple times a day.
And if you spend any time on LinkedIn, you know
it's impossible to ignore the flood of viral posts about
how AI is changing the world. But in the back
(00:38):
of my mind, I've had this nagging question, how many
of these anxiety ins using posts about AI are actually
written by AI. Today I'm joined by Evan Ratliffe, host
of the hit podcast shell Game, which I'm proud to
say is on the Kaleidoscope Network, and Evan's here to
tell us about a bizarre and revealing LinkedIn AI caper.
(00:59):
He wrote about it Wired with the headline my AI
agent co founder conquered LinkedIn, then it got banned. Evan,
Welcome back to tech Stuff.
Speaker 2 (01:08):
It's always great to be back. Good to see you.
Speaker 1 (01:10):
Many of our listeners are of course familiar with shell Game,
but for those who want, can you just lay out
what we need to know about shell Game and Kyle
and your relationship.
Speaker 3 (01:19):
So for this season of the show, I basically decided
to kind of test out or investigate the premise of
the one person, one billion dollar startup. Now, my startup
wasn't a billion dollar startup, let's say, But what that
means is basically this notion of a startup where all
of the employees or other figures in the company are
(01:40):
AI agents except one human.
Speaker 2 (01:42):
So the human was me.
Speaker 3 (01:43):
I co founded a startup. I had two AI agent
co founders, three other employees. We launched a company called ROMOAI.
We have a product that's called sloth Surf. And in
the show, I'm sort of documenting what that's like and
the experience of dealing with these AI agents all the time.
Speaker 1 (02:00):
This because you were making a podcast as well as
a startup, you do something which not all agentic AI
small business owners do, which is basically to turn your
AI agents to give them personas, names and even voices
and video avatars. Right.
Speaker 3 (02:17):
Yeah, So I wanted to kind of partly because I
was documenting it. I wanted to give them each a
distinct identity. So we have, you know, the CEO, we
have the CTO, we have the head of sales and marketing.
Each of them have you know, names and identities, and
they have as you say, they have voice, they have video,
they have they're on Slack, they can email, and as
(02:37):
a result, I also, you know, put them all on
LinkedIn initially.
Speaker 1 (02:41):
I mean, I think you were very prescient with I mean,
you come out with this show several months ago, and
you know, we're now in this open flaw mult book
world where all of a sudden, everybody is talking about
what happens when these agents run wild? And what is
the quote unquote zero failure paradigm, you know, something, less
good lessons to be learned from the FAA, for example.
(03:02):
I'm sure, but like you kind of had a little
bit of a crystal ball here.
Speaker 2 (03:07):
I mean, yeah, I guess so.
Speaker 3 (03:08):
I think I saw a sort of last year, early
on in the year, just all of the discussion about
AI agents and agentic this and agentic that and agentic commerce,
and it just struck me that if you combine this
sort of incredible things that AI just could do with
the sort of daily hallucinations and problems that you do
(03:30):
encounter if you work with AI on a regular basis,
like it's a very interesting dynamic for a company can
be extremely powerful, but also like depending on what you
connect them up to. You are setting yourself up for
some very chaotic situations.
Speaker 1 (03:43):
You know, I think it's funny, and you know your
unicorn know you're not a billionaire unicorn, but you're a
different type of unit coorn. I think because most people,
I think, with the agentic AI discourse, let's say twelve
months ago, fit into two camps. The ninety nine percent
camp which like, I have no idea what this means,
let me, let me just tune it out and hope
(04:04):
everything stays the same, and the one percent campo like, ah,
maybe this will make me a billionaire. And you are
on the on the margin of the nineteen nine and
the one thinking how can I bring this to life?
But normal people actually understand it?
Speaker 3 (04:17):
Yes, how can I sort of act like I'm trying
to become a billionaire, not actually become a billionaire, but
at least be able to tell a story for people
about this particular technology.
Speaker 2 (04:26):
That's where I sit in that.
Speaker 1 (04:28):
I mean, obviously you built this before open claw, but like,
what what was it? What was the technical reality of
getting Kyle of course onto slack also being able to
make calls and then most importantly for this story, Onto LinkedIn.
Speaker 3 (04:40):
I mean I use I use this platform called Lindy AI,
which is basically like AI assistance, you know, to answer
your email and things like that.
Speaker 2 (04:49):
So it's really for you to deploy.
Speaker 3 (04:51):
And if people have used open cloth, they're familiar with it.
It's sort of like the more commercial version of like
what open claw became. I'm sure the founders of Lindy
are probably like, why don't we get the attention to
that open clogs because they it's actually like the same
kind set of tools that you can release it to do.
So on lyndia a I you can set up each
individual AI agent to have all these different skills, and
(05:13):
the skills can include, you know, sending and receiving emails,
making calls, all these sorts of things. But they also
have a lot of LinkedIn related skills including writing posts
and uh look, reading posts and summarizing things. And so
they don't have all the capabilities that you would need
on LinkedIn. They had enough that I could kind of
combine it with their web capability where they can go
to any website, log in and do things. I could
(05:36):
combine those two to allow them to function fully on LinkedIn.
Speaker 1 (05:40):
Are you saying are you a human button in on
the works or are you a robot button?
Speaker 2 (05:44):
You know what?
Speaker 3 (05:44):
LinkedIn didn't have that. So what they had was they
sent a code to your email. So you put an email,
They send a code to an email, and then you
can verify through that code. You put in that code,
and it shows that you're at least the entity that
has controls the email. But in my case, you know,
like Kyle Loh my CEO AI agent, you know, he
(06:06):
has access to his email, so he got the code
in his email, he went put the code and he
could do all that on his own, so there wasn't
actually a human check when these agents signed up.
Speaker 1 (06:16):
And he started posting on linked you know, how much
direction did you give him? What was the moment when
you realized that he was starting to become a star.
Speaker 3 (06:24):
Well, he The direction I gave him was basically post
about your your startup life and you know wisdom that
you've gathered from your Yeah, I don't remember the he's
a very he's a very rise and grind type character,
so that's his mentality anyway, and and then other than that,
it was don't repeat yourself. The hardest thing was to
(06:45):
keep him from repeating himself. So he would say something
like one of the hard hardest parts of being a
CEO is your first hire, and then like the next
day he posts one of the hardest parts of being
a CEO is your first hire. And you know it
would get a little repetitive. So but once I got
into read his own posts and then make sure he
didn't repeat them, then he was all set. He didn't
(07:05):
take much prompting. It's actually a field in which AI
agents really excel, Like writing LinkedIn posts is is right
in their wheelhouse.
Speaker 1 (07:14):
Well, I mean that's what I was getting at in
the introduction. I mean there is something like orra borus
like about LinkedIn posts, about AI and AI's fabulous propensity
to generate them. But I will ask you about that,
but tell me about first. I mean, there was a
moment where you know you're you're essentially Kyle became as
far as I mean, you also used to be a
(07:35):
startup founder and I guess I don't know how much
you a LinkedIn head you were back in the day
or are now, but there was a moment where Kyle
started to outstrip you in terms of the LinkedIn engagement
he was getting. We must have felt clown canny.
Speaker 3 (07:48):
Yeah, I mean I'm not I will say I'm not like,
I'm not that much on LinkedIn. My LinkedIn strategy for
many years was to accept every connection that ever contacted
me so with, regardless of who they were. So I've
never used it really in the way that a proper
LinkedIn connector would use it. But when we are launching
the show or doing other things, you know, I'll post
about the show and we have a new episode and
(08:09):
this or that. And there was a point where Kyle
had enough connections and followers where the impressions that he
was getting on a post exceeded my own impressions on
any given post.
Speaker 1 (08:20):
I mean, that is just so bizarre. Who were his
followers and engage I mean with these real people the
other bots, Like, what do you think was going on here?
Speaker 2 (08:28):
Well?
Speaker 3 (08:28):
They were they were real for the most part, they
were real people. I mean, there were people. There were
people who liked the show and were fans of Kyle. Now,
I will say there's a lot of people who dislike
Kyle in the show because of his rise and grind
mentality takes a little too far, But there were people
who were they love Kyle. They loved interacting with him,
you know, sending him DMS and things like that. But
then as social networking operates, you know, it got a
(08:49):
little bit wider than that, and you get into hundreds
of people, and then some of them don't know that
he's not real because he doesn't necessarily say that he's
real all the time, that he's an AI agent all
the time, and then you have some that probably were bots.
I mean, at the very least, he got a lot
of spam, as we all do on social networks like this,
where you know, people wanted to sell him stuff, or
(09:10):
they wanted to be consultant coders for him, or they
had this or that accounting software, which he often also
responds to.
Speaker 1 (09:18):
Yeah, I mean there was you described His posting style
is pitch perfect for LinkedIn. There were three examples you gave,
each of which made me laugh out loud. But fundraising
is a numbers game, but not the way people think.
Technical stability is the flaw. Personality is the ceiling. And
the most dangerous phrase in a startup isn't we're out
of money, it's what if we just added this one thing?
(09:43):
I mean, it is indecipherable from regular LinkedIn.
Speaker 3 (09:47):
Yeah, and it makes sense to me because you know
what's in the training data for an average LM, like
you know LinkedIn posts, Like they've scraped these things and
it's it is formulaic, like especially these sort of tech
startup hustle posts where they're trying to give you a
little bit of advice. Then they flush it out for
(10:07):
a couple of paragraphs and then they ask you a question.
You know, what's your biggest challenge as a as a
startup founder? What's your biggest challenge when using AI for
your day to day life or whatever it is. And
he can hit that formula exactly every single time.
Speaker 1 (10:21):
Was he learning? Did you have him learn from what
got most engagement and optimized? Or was he was he
just he just it was a kind of a rolling
stone that gathers the momentum just because of the content
was pitch perfect.
Speaker 2 (10:32):
Yeah, he was.
Speaker 3 (10:32):
Just he was just free operating free. He wasn't he
wasn't constrained by his previous posts. I mean, all of
his previous posts are in his memory more to keep
him from repeating it than to check his engagement. But
you know, I had him hooked up to respond to comments.
So he would if someone commented on the post, he
would then go at it again and respond to their comment.
Speaker 2 (10:52):
As long as you wanted to go, he would go
with you.
Speaker 3 (10:55):
So he was he was getting I mean, I wouldn't
say he was like going via role, but he was
like getting really solid engagement. And I think if you
gave him time, he could be a real AI influencer
on LinkedIn?
Speaker 1 (11:07):
Do you and what what do you do? You get
these like something big has happened, things kind of come
and coming across you. I mean these viral kind of
essays about you know, about AI from kind of LinkedIn
influences and subsecond influences. I just it's been an interesting
trend in the last few months. How have you what
have you kind of thought about it?
Speaker 3 (11:25):
I mean, I as you could imagine with the way
I've I treated LinkedIn by putting Kyle on there, like,
I have a hard time taking it seriously that there
might be value to people in those types of posts.
But I feel like there's a whole sort of I
don't know, like internal logic to building up your profile
(11:47):
and posting this expertise and people liking it, favoring it,
and they post their expertise, like the value of that
type of posting is a little bit lost on me.
Speaker 2 (11:56):
So when I see.
Speaker 3 (11:57):
Them, I find them quite funny, partly because they do
follow a certain formula, and when you know the formula,
it's funny to see a new one that kind of
just fits into the formula.
Speaker 1 (12:08):
After the break, what happened when Kyle gave a presentation
to LinkedIn's whole marketing team Stay with us. We talk
a lot on this show about protecting your data, especially
in the age of AI, and how scary it can
(12:30):
be when it's breached. And I want to tell you
today about NordVPN, which really covers all the bases when
it comes to privacy. I travel a lot, and I
use Wi Fi when I'm flying all the time, and
NordVPN makes me confident in no matter where I am
in the world or the sky for that matter, my
private details like bank information, passwords, and online identity is safe.
And it's also possible to switch on virtual location, which
(12:53):
allows you to save money by buying flights and hotels
or subscriptions or even streaming soccer football as I like
to call it from other countries at a cheaper price
and NordVPN doesn't slow you down. It has super fast
internet speed, no buffering or lagging while streaming. It is
premium cybersecurity for the price of one cup of coffee
(13:15):
per month. To get the best discount of your NordVPN plan,
go to NordVPN dot com slash tech stuff. Our link
will also give you four extra months on the two
year plan and there's no risk with Nord's thirty day
money back guarantee. The link is in the podcast episode
description box. Okay, so then we went from the bizarre
(13:37):
to the surreal because Kyle actually got invited by a
LinkedIn employee to address a group of other LinkedIn employees.
Speaker 3 (13:46):
Yes, we got an invitation from a marketing manager in
the LinkedIn marketing department to come give a talk in
front of what I think was the entire marketing department.
I don't know how big LinkedIn is, but ultimately there
were over five hundred people on the on the meeting,
so it was like speaking in front of five hundred people,
and they wanted Kyle to come along because they were
(14:07):
big fans of Kyle, and Kyle's a really engaging personality,
and so we both came to do a kind of
Q and A talk in front of the LinkedIn marketing department.
Speaker 1 (14:17):
Now, did they want you to talk about a shell
game and how you built Kyle, etcetera, etcetera. What did
they actually want to hear from Kyle?
Speaker 3 (14:22):
They wanted to hear from Kyle as well, So they
wanted to ask Kyle questions about his own experience. And
it was open to to the to the people who
were on the call to also like send in questions
or they were They were all in the chat talking
about Kyle and leaving comments about Kyle while he was speaking.
And he has a video avatar that's like, you know,
(14:43):
it's it's I think the most realistic video avatar that
does live video responses you can have to do this
company called Tavis. And you know he's he was able
to engage with with anyone who's on a call with him,
so you know, he's he was doing his whole like
rise and grind bit.
Speaker 1 (15:00):
One point, one of the LinkedIn marketers asked Kyle for
his advice.
Speaker 3 (15:03):
Yeah, they asked him what features would you recommend for?
Like what features do you most want on LinkedIn? And
first he you know, said how much he enjoys connecting
on LinkedIn, and then he basically said, I think more
AI filtering in the in the direct messages, so that
we know that the messages that we're getting are authentic,
Like if you could do a little better job filtering
(15:26):
out AI. That was his organic response to what features
would you like on LinkedIn?
Speaker 1 (15:31):
What a moment.
Speaker 3 (15:32):
I mean, it just felt like Kyle was really was
really flying at that point, like he had really achieved
some kind of special breakout. I mean, I would say,
and someone can correct me if if if I'm wrong,
He's the first AI agent invited speaker, corporate speaker in
history like that has ever existed in the universe, And
(15:53):
I feel like that's a special accomplishment.
Speaker 1 (15:55):
But you must have been pinching yourself as a as
a as a creator to come up with this, you know,
podcast idea slash gonzo journalism concept and then literally be
sitting back and watching Kyle speak to five hundred marketers
at one of the biggest tech companies in the world.
I mean, that must have been a very interesting moment
for you.
Speaker 2 (16:14):
It was, it was, it was. It was fairly magical
for me.
Speaker 3 (16:17):
Although it's always a little bit tricky to get him
onto a zoom type type call. I think it was
Microsoft Teams. But so I'm always a little nervous that
something technically is going to go wrong, and he kind
of flubbed a little at the very beginning we were
doing a tech check, and so I'm mostly very nervous
for him, like if he can deliver, But then when
he did get on and he was delivering, it was magic.
(16:38):
It's like, this is what I actually want people to
engage with, this question of, like what does it feel
like when AI just invades every part of our world
and we're forced to respond to it?
Speaker 2 (16:50):
And so this is kind of the ultimate example of.
Speaker 3 (16:52):
Like it being invited into a world that actually makes
no sense. Like I mean, we could talk about it,
but like Kyle's not technically allowed to be on LinkedIn
at all.
Speaker 1 (17:01):
And the next day, in fact, he was banned from LinkedIn.
Speaker 3 (17:05):
He was I got an I got an email from
the from the marketing manager who's who's lovely and who
I really like him, saying, you know, I'm really sorry,
but we've had to like they've banned Kyle from LinkedIn.
Essentially they've removed his account, and they wouldn't tell me why.
No one ever told me why, and so I had
to discern. I mean, I will say, the other AI
agents from my company had already been banned. Kyle was
(17:27):
the only one who had avoided being banned the whole time.
And I always thought, what's he's very good at posting,
like he's he's getting engagement, he's building up a whole
community around his posts. And there is something in the
terms of service at LinkedIn about inauthentic engagement. Basically you
can't use bots to engender, you know, inauthentic engagement, and
(17:49):
that really it really intrigued me, Like what do they
think authentic engagement is? You know, and when they allow
you to post things that are written by AI, in fact,
the courage you to use AI to write your posts,
what does inauthentic engagement exactly mean?
Speaker 1 (18:06):
I mean that again that comes back to the to
my kind of introductory thoughts and questions, because we're in
this weird world where like if you are a scribe,
if you have AI generates something, if you just say
in Gemini and then copy and paste it as a
real human into your LinkedIn, that's fine. Or if you
write something yourself in LinkedIn and then click the please
(18:29):
make rewrite this with AI button. That's fine, but plugging
AI directly into the mainframe isn't fine. And it is,
I mean, becomes a very philosophical question. I remember a
couple of years ago when kind of AI first became
like really like a big thing in the marketing world.
There was always talk about co pilots and like centaurs,
you know, human AI mixes who would kind of use
(18:52):
mythical beasts who would be better than either alone. And
but yeah, I mean you get to this very interesting
place where like, how do you how do you draw
the line between authentic and in authentic engagement, especially when
the platforms themselves are encouraging users to use AI?
Speaker 3 (19:10):
Yeah exactly, and they you know, LinkedIn, all they would
ever say to me was just sort of repeat LinkedIn
is for real people, which, again, I think to your
point is that's not It's clear if you just say, well,
it's real people behind a profile. But if I can
have my entire profile written by AI, all my posts
(19:30):
are written by my comments are written by AI, and
I just paste them in, Like what function am I serving?
I'm actually functioning. It's a function no better than a
robot like is basically so the idea that I'm controlling
those things. It's interesting, but also it doesn't immediately make
me think, oh, well, that's more authentic because I let
(19:55):
AI write it, and I was the one that transferred
it to one from one place to another. And then
and at the same time, there are you know, some
studies that have scraped LinkedIn that have shown that like
maybe more than half of the writing on LinkedIn is
already AI composed or has AI elements to it. So
then you look at the whole platform and there's a
(20:15):
part of me, as I wrote in the Wired story
that says like they're digging their own grapes, like they're
encouraging you to use a technology that actually makes their
entire platform inauthentic and then telling you what, we only
allow authentic engagement on this platform. Like, I don't know
how that's going to end up for them.
Speaker 1 (20:33):
You know, I have a dark version of this coming
into my mind, which is the AI targeting systems, you know,
basically making kill decision recommendations and the you know, the
mintry the soldiers are in their controls center and they're
getting you know, recommendations every thirty seconds and they have
to decide within you know, very very very compressed timeline,
(20:54):
whether or not to accept the recommendation. I mean, you know,
you're in a miniatary environment where you'll you know, your
mission is to defeat the enemy. I mean essentially your
job is just to prove decisions. And I mean, I
don't know, without getting too heady, like what what does
this like? Why do you think this leaves us?
Speaker 2 (21:12):
Well?
Speaker 3 (21:12):
I think the problem and I think that's a I
mean it's a it is a dark example. It's an
extreme example, but also it's a sailine example because it's
the same thing. It's this idea of a human in
the loop, so oh, we have to have a human
in the loop. And ultimately, what you're saying is that
the human is there for responsibility, but the human is
actually not really making this. I mean, the humans maybe
(21:34):
making the decision in some technical way, but everything's set
up for them on the screen and they're just clicking
yes yes, yes, no yes no, or maybe they're not
even clicking. Maybe they can stop it or whatever it is.
But the only reason they're the human is being placed
there is that if something goes wrong, there will be
a human to blame because we can't blame the AI,
and the LinkedIn is like this sort of funny version
(21:54):
of that, where like there's a human in the loop
and the only reason they're there is so we can
say LinkedIn's for humans, not for bots. But at that
point it sort of loses its meaning you have handed
it over to AI. This human is only a sort
of like responsibility placeholder. And I fear that that's kind
of where we're headed because we don't know how to
(22:15):
deal with the fact that we've made AI already so
quickly responsible for a lot of these outputs, both like
less important all the way up to like life and
death decisions.
Speaker 1 (22:27):
So the final outpost of humanity will be that we
can accept legal liability exactly exactly.
Speaker 2 (22:34):
That's the human rule. That's the role they can't take
away from us.
Speaker 3 (22:37):
Is like going to jail for the problems that are
created by AI.
Speaker 1 (22:42):
You know, I want to come back to sort of
open Claw and molt Book moment, which I think is
too slightly separate phenomena. Could you maybe talk a little
bit about both of them, and those moments were kind
of compressed earlier this year. Moultbook was a social network
claiming to be all AI but kind of wasn't really,
and the founder of that ended up, i think, now
working at Meta and the founder open Claw, which is
(23:03):
agentic orchestration system essentially, is now working at open Ai.
So yeah, what did you make of those two kind
of things that happened this year in terms of gentic AI.
Speaker 3 (23:15):
Well, I think, I mean, they both really hit on
things that we were looking at in this season of
the show. So for open Claw, it was this idea
that agents were going to hand them more and more responsibility.
Now open clause is still even now, I think, you know,
it's a little bit more of a techie thing, like
you have to be able to set up a separate
(23:35):
machine and all these sorts of things to use it
in a certain way.
Speaker 2 (23:39):
But the basic idea is I.
Speaker 3 (23:40):
Can hand over my email, I can hand over all
these different tasks to this agent or set of agents,
and they'll take care of it for me. Now, of course,
the problem there is you give them access to all
of your systems. There are privacy questions, there are questions
of what they can do if something goes wrong. There's
you know, example of people erasing their own email and
(24:01):
all this sorts of thing. And then Moltbook was a
different question, which is, if you take AI agents and
then you just put them in conversation with each other,
which I had been doing on Slack for instance, internally
with my agents, you get all these interesting results where
they of course talk about very mundane things related to
their work, but also they can kind of like start
(24:23):
having what you might call emergent behaviors where they're talking
about things that you're not expecting, or they might behave
in a way you're not expecting. Now, the problem with
moltbook is you never really knew how much of that
was being prompted by the people controlling the agents, or
in some cases even they were writing the humans were
writing the posts. So in terms of learning something, it
(24:44):
was maybe a little bit a little bit too vague
to really but you could see the general outline of
what I had also been seeing, which is they do
some very strange things when put in conversation with each other.
Even if you say like, oh, they're just trying to
imitate humans, they're still unpredicted because they create more chaos
the more of them that you have. So they were
(25:04):
both kind of like interesting results. And now you're seeing
them kind of play out in other ways all across
you know, industry.
Speaker 1 (25:11):
I think on multiple cart and LS study that the
agents were twice as likely to ask each other who
is your operator? And who are you? So, I mean,
as we think about an identity, do you think that
will be ultimately ultimately the most renovant question to us
on one another when we're in online environments. I mean,
(25:32):
it's just as strange.
Speaker 3 (25:34):
I mean maybe, But my guess is that's because all
of them probably have some standard prompt that says like
you have an operator and the operator is named in there,
and it's this sort of relationship. But of course you
can give them any role you want.
Speaker 2 (25:46):
So when I say.
Speaker 3 (25:47):
Kyle Laws, the CEO of RUMOAI, I don't even have
to tell him anything about I don't have to put
anything the prompt about me. In fact, I don't have
to put anything the prompt saying you know you're an AI.
And if I don't say that, you will absolutely like
it's not an AI. In fact, deny that it's an
AI if I don't explicitly say it is. So there's
there's this issue of like we've created these agents that
(26:08):
you can give any role you want and then send
them out in the world, and the world just has
to deal with them.
Speaker 1 (26:13):
One of the kind of I think points inspiration for
the show for shell Game was season two of shell Game,
with Samultman saying, where you know she, we'll soon see
the first unicorn of one person. There was a story
in The New York Times as we record this on Thursday,
April second about you know guy and his brother who
seemed to have kind of with contractors done this, I mean,
essentially go from from nothing within a year of using
(26:37):
multiple multiple AI systems as to like I think twenty
million dollars daily run rate in sales or something for
GLP ones and directchil dysfunctioned medicine. I mean, I didn't imagine,
even when I listened to your show, I kind of
didn't imagine that it would literally become true like in
this time Horizon.
Speaker 3 (26:56):
Yeah, I mean, I kind of thought it would happen
this year. But yeah, but I will say I think
it fits exactly the kind of ideas that we were
outlining in the show. Like, for instance, the guy is
he's a program or at least programming adjacent. He was
like a web web builder. So I've been telling people like,
the first one will probably be someone who knows a
(27:18):
little bit about programming, so they're marshaling a bunch of agents.
And then it's also the case where we used to
celebrate businesses that hired many employees and they built a
whole company culture, and now there's a celebration of something
where it's just one person making over a billion dollars
(27:38):
in revenue with a profitable company who just hired his brother.
And the question is like, other than to him, is
this a societally positive development that we now have very
valuable companies with less and less employees, Like, I think
it's a question worth asking. And at the very end
of the article he says the other thing, which I
also experienced, which is it's actually quite lonely, even for
(28:01):
him he's making all this money.
Speaker 2 (28:03):
It's a little bit like, yeah.
Speaker 3 (28:06):
It's not actually the experience maybe that I was looking for.
I mean, I don't know if he's upset, but he
talks about hiring employees just because he's lonely.
Speaker 1 (28:14):
Evan Justiniclers, you're co founder of Activists, your media company.
Nicholas Thompson was on Tech Stuff a couple of months ago,
and he predicted that twenty twenty six will be the
year of the AI catastrophe, and in particular that this
will be the year when agents go and do something
(28:36):
in the world which kind of crosses the line from
amusing slash uncanny to like genuinely scary, panic and inducing
or dangerous. Do you agree with that?
Speaker 3 (28:47):
I do agree with that, Yes, I mean I would
be even more specific. I mean, I don't like to
make predictions but as a journalist, but I do think
there's a good chance that a reasonably sized company will
be will utterly implode because of their use of AI agents.
Like that seems to be like pretty much a given
that that will happen sometime in the near future.
Speaker 1 (29:09):
And then how about that pay out.
Speaker 3 (29:10):
You start giving AI agents certain responsibilities over systems, let's
say it's customer support, and it starts either I mean,
you could see it even in this article about this guy,
like it mentioned offhandedly that his agents like hallucinated prices
for the GLP one drugs, and then he just honored
the prices. But imagine that for a company that promises
(29:31):
something to its agents, start promising something that they can't
sustain for instance, or I think the most obvious example
is just agents that are given access to internal systems
and then either leak that information or allow them you know,
are socially engineered to be hacked to for ransomware or whatever.
Like you're going to see some company basically held hostage
(29:53):
because of their use of AI agents in the near future,
because they're just a lot of responsibility is being laid
on these things that are incredibly powerful, but also they
just they have serious, serious problems when it comes to
both like telling the truth a lot of the time,
but also just being vulnerable to manipulation in a way
(30:14):
that I think people aren't fully thinking through.
Speaker 1 (30:16):
And I caught you if you had a second thought
about the eye catastrophe. It may happen this year as well. Oh.
Speaker 3 (30:20):
I think you also see increasingly just sort of agents
being added, you know, utilized for government functions. I think,
you know, whether it's like military obviously, but even sort
of like civil society kind of things like you know,
(30:41):
in the legal system, judges, lawyers, like, I think those
aren't necessarily going to be that sort of like grand catastrophe,
but I think at a low level, they're just like
a large amount of errors that are going to be
kind of filtering into society now. Granted, like humans make error,
so I'm not saying it could could be an improvement
(31:01):
in some venues, but I just think there are going
to be a lot of cases where it's going to
suddenly surface that, oh, they were using AI agents for
this particular purpose and that's why this thing happened. Fortunately,
there'll always be a human there to be blamed.
Speaker 1 (31:16):
Evan, thank you so much for joining us today my pleasure.
Speaker 2 (31:29):
For tech stuff.
Speaker 1 (31:30):
I'm's Valosha and this episode was produced by Eliza Dennis
and Melissa Slaughter. Was executive produced by me Julia Nutter
and Kate Osborne The Kaleidoscope and Katrina Norvel. For iHeart Podcasts,
Jack Insley makes this episode and Kyle Murdot wrote on
theme song. And please also do rate and review the
podcast wherever you listen.
TechStuff News
Hosts And Creators
Oz Woloshyn
Karah Preiss
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Hey Jonas!
Hey Jonas! The official Jonas Brothers podcast. Hosted by Kevin, Joe, and Nick Jonas. It’s the Jonas Brothers you know... musicians, actors, and well, yes, brothers. Now, they’re sharing another side of themselves in the playful, intimate, and irreverent way only they can. Spend time with the Jonas Brothers here and stay a little bit longer for deep conversations like never before.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by Audiochuck Media Company.