Subscribe
Share
Share
Embed
A podcast that explores how AI is transforming careers, businesses, and industries. Hosts Greg Boone and Erica Rooney deliver real-world use cases and actionable AI strategies to help professionals stay ahead of the curve.
Learn more about your hosts:
Erica Rooney, author of The AI Gap: Women, AI, and the Next Great Leap Forward
Greg Boone, author of AI at the Speed of Trust (available for pre-order)
Welcome, welcome, welcome.
Live, AI Voice of Victim.
I am Greg Boone, CEO of WalkWest,
co-founder of RDU Labs.
It's actually a very special day today for
me.
It is my birthday.
Happy birthday.
So I appreciate it.
I have a very special guest for our
first live edition, Mr. Alec Coughlin,
AI with Alec.
Introduce yourself, Alec.
Hi, my name is Alec.
I'm a recovering management consultant.
it's true it's all true but yes i
am the founder of ai with alec and
i am a forward deployed principal for
various dev big engineering firm that is
awesome i like it i was saying just
last week i was teaching ai to a
uh at a community college for their
faculty nice and i said that i am
a recovering software engineer i stole
that from you okay fair
And so as a recovering software engineer
and recovering management consultant,
I feel like our roles have somewhat
actually flipped.
For sure.
And in the last couple of years,
because I spend more time doing management
consulting.
So, you know,
I appreciate you being here as a part
of this conversation.
I know we talk a lot about AI
and kind of its impact, not just on.
society but in the business community but
what are the things that are changing and
evolving because everyone says how do you
keep up and i simply say that i
read ai with alex and i watch his
stuff pay you later great because his
hands are in the dirt it's a look
finally putting a face to the name because
i say all the time on stage or
when i'm training folks i say like my
man alex says you have to have your
hands in the dirt
And I talk about that with leadership
teams, not just the technical team.
So hopefully today,
if we can dive in a little bit,
I want to understand your view on AI
adoption.
What are the things that are happening?
I got my own perspectives.
I have a lot of conversations, right?
But let's start out with why are folks
not seeing ROI just yet?
Or are you seeing people finally seeing
ROI when it comes to AI initiatives?
You know, the ROI question, I think, begs,
you got to talk about the things that
you don't want to talk about, you know,
when it comes to what's the root cause
of that.
I think generally speaking,
a lot of folks might confuse this as
a technology problem or challenge.
You know, the more things change,
the more they stay the same.
It's a people process and technology
thing.
You know,
if I had a dollar of bandwidth to
invest,
I would take about ninety cents of that
dollar and I would sit with business
people, you know,
which I think a lot of folks that
are listening to this are those business
people which are
the people that have the ability to define
what is it that we're trying to achieve?
What is the success metric, right?
What is the ROI?
What is the current state of something?
Now,
where it gets three-dimensional chess-y
like is that
It's really difficult for someone who
doesn't have their hands in the dirt,
regardless of how technical they are or
aren't,
to understand what's possible if they
don't have their hands in the dirt.
So you have this pull and push between
the technical folks that really understand
the technical deployment overhang, right?
All the horsepower.
But they lack the business stuff, right?
They lack the feel.
They lack the customer problem.
They're not close enough to it.
And the other end of the spectrum,
you got the opposite, right?
And so how do you bridge and bring
those groups together?
Which is why I always argue that I
think the most important thing that anyone
can do,
regardless of technical capability,
is put your hands in the dirt and
start building things.
Because once you see it,
you can't unsee it.
Then you should be like, oh,
I could save fourteen hours a week by
having the genius machine do the following
for me as opposed to me doing that.
And what am I going to do with
the fourteen hours?
Good question, right?
Here comes ROI.
Yeah, no, that's a great point.
And I think people that are watching this
are going to see where I get all
my phrasing from.
Because I say, once you see it,
you can't unsee it.
And I, you know,
I talk a lot about this.
I say to folks, you know,
fifty to sixty percent of our work weeks
are comprised of searching for things to
do our job,
copying and pasting and reformatting data.
That's wild.
Right.
Fifty to sixty percent.
When you add it all up.
Right.
And every single audience that I talk to.
Everyone's nodding their heads.
They all know that this is a pain
point.
Yeah.
Right.
But they have no frame of reference.
They don't have their hands in the dirt.
They don't know that you can credibly get
sixty to eighty percent of that time back.
Right.
It's that frame of reference problem.
You know, McKinsey just last week,
I always like to post stuff and I
say, don't take my word for it.
And then I push it out.
You know, I just repost the experts.
McKinsey's pretty smart.
Right.
But they had a stat and they were
showing just last week.
They said that for every dollar for the
licenses,
there should be another it's called one
three five i guess that's how they broke
it up right there's a dollar for the
tech services three dollars to defining
new processes and redesign and then five
dollars effectively for adoption and
training yeah right but that's not really
what people are doing no they're basically
just getting license and then roll them
out there and say go have at it
right how do you see that playing out
i mean you want the spicy take you
want to like i mean there's nothing i
mean
What is it?
Thirty days to develop a new habit, right?
Generally speaking.
I think it's hard for people to hear
this,
but a lot of folks are afraid to
do the work, right?
They don't like not looking like they know
what they're doing, right?
They don't.
You know,
I think one of the biggest challenges the
AI industry faces is that we've done a
horrific job at marketing it.
And one of the ways that I think
we've fallen flat on our face is that
people either see it like very doomy and
gloomy, you know,
take everybody's job and all this, that,
the other thing, and, you know,
Robocop, whatever.
But then the other way they see it
is like, for some reason,
people think it's supposed to be so easy.
You just press the button, right?
It's like, dude, no, absolutely not.
You got to do the work.
You got to build the system.
You got to iterate.
Like if you were to bring a new
employee into your arena, whatever level,
I mean, you know,
maybe some sort of a rock star steps
into a chair and just, like,
lights it up.
But there's an onboarding period.
They got to understand, like,
the way Greg operates,
the way Alec operates,
the way Javier operates, whomever.
And I think, you know,
maybe this audience might appreciate or
maybe they don't appreciate hearing that,
like, hey, it's okay that...
it's you're a little bit overwhelmed by
how much work it entails.
But if you want something, you know,
you got to earn it and you got
to work hard to get it.
And this is not AI is no different,
but the rewards are exponential.
I mean, it's massive.
I say all the time,
there's nothing intuitive about AI right
now.
When the average audience,
when I start to explain to folks what's
happening behind the machine,
when to choose,
one of the first questions I always ask
is, how often do you change the model?
And they're like,
what are you talking about?
Like, what is that?
Like, well, if you're in Copilot,
you click on that little thing that says
auto, you can select chat GPT.
Sometimes if you have an enterprise or a
premium license, you can select Claude.
You got the deep thinking,
the reasoning models,
you have the fast response.
So I like to show folks, I'm like,
this is what a quick response gets you
on the average thing that you're doing.
This is what you get with the reasoning
model.
And you can see the quality.
You can see the difference, right?
Explaining why the thing is hallucinating,
right?
Why I was talking to just yesterday,
I had a conversation with Dr.
George Westerman at MIT,
who runs the executive ed program that I
went through last year.
We were having some conversation.
He was asking, I was saying like, hey,
someone on stage last week,
I was at a conference,
said that they asked the audience,
how many of you know what a context
window is?
Out of like one hundred and fifty people,
like ten people raised their hand.
And basically that's why I was like,
therein lies the problem.
And then Dr. Westman said, well, Greg,
why do you think that that's important for
non-technical folks?
I said, well,
they don't understand why the thing
hallucinates.
They don't understand that every new chat
is a new conversation within that context
window.
They don't understand a lot of why we
keep saying the word context.
Right.
And so the more that I've seen that
we kind of demystify what's actually
happening behind the machine,
the easier it is for them to understand
how it can apply to their work and
how they use it.
Right.
And so my so I guess my question
to you and I ask this all the
time now is why do you think and
maybe you answered it already,
but why do you think leaders have fallen
in love with this whole just be curious,
you'll figure it out mindset with
something that's so complex?
be spicy take, you know,
maybe those leaders haven't necessarily
done the work to be able to say,
hey,
the reason why context is so important is
that if you contextualize something before
you ask a question to another human,
they're going to give you a much better
answer.
Or it's that wonderful quote that I always
butcher and I have no idea who to
attribute it to,
whether it's Hemingway or Blaise Pascal.
But it's like,
so sorry for the long letter.
I wish I had more time.
Right.
So like when you're using these systems,
the more you invest in the CapEx,
which is the equivalent of building the
system the right way to benefit you,
the more and better the OpEx,
which is the day to day utility and
use of it.
But you kind of work.
And the more I think people understand
that,
it's kind of like if you want to
get better at a certain sport,
if you don't understand the mechanics,
how to throw the ball,
how are you going to throw the ball
faster?
You know what I mean?
Like, you're just going to muscle it?
Like, that's not fair.
I said the same thing last week.
I was talking to someone.
I played basketball and baseball in
college, and I was a pitcher,
so I very much understand.
You know about mechanics?
I very much understand this.
And I said,
I feel like people are saying, hey,
don't teach me how to dribble or how
to shoot.
Just show me how to dunk already.
And I was like,
I don't think it works that way.
Right.
There are certain things that you have to
be able to do.
And since you're dropping quotes and
people ask me about adoption,
what is the biggest struggle?
And I said,
I use the same Upton Sinclair quote from
nineteen thirty four,
which is it's difficult to get a man
to understand something when his salary
depends upon his not understanding facts.
These leaders don't want to change.
Right.
They don't want to get their hands in
the dirt.
Therefore,
they have no frame of reference.
So they're still saying, hey,
just play around with it.
You figure it out.
I'll give you one spicy one on that
one too.
I want to say it real quick before
we continue into a different topic.
I hate to say it, but again,
in the interest of spicy,
and I know we're talking with folks that
are business and perhaps non-technical,
don't ever underestimate how much AI can
improve and enhance wonderful process.
But where there is dysfunction,
where there are political dimensions,
where there are all those things,
The reason why I might create all sorts
of problems is because your process is
broken.
Your culture is not as healthy as you
might think it is,
and therefore you got to address those
things before you jump in and bring in
the genius machine.
Yeah, I mean,
know a lot about that i have a
lot of conversations i say people process
platforms in that order always i've been
saying that for the last fifteen years
around digital transformation um the
process and the people side seem to be
the one people want to skip they want
to go to the platform but i'm going
to skip around a little bit here right
we're talking about the impact of having
the right culture and process basically in
and trying to do things in the search
border right not just jumping to the
technology the platforms
One of the other things that just keeps
coming up all the time is people want
immediately go to agents and what it means
to be a gente.
And again, I say, you can't dribble.
You want to go right to dunking, right?
And so first for our audience,
could you define, you know,
an agent for the audience or to have
agency?
I love that.
And I would use the agency word.
So think of an agent as a true
agent has not just the autonomy,
but it can accomplish tasks and it can
iterate and learn and improve upon itself
in a recursive manner.
Fancy way of saying it's not automation,
right?
It has agency to Greg,
to your point is like,
It's the ability to develop something and
have that something grow up into a value
creating, let's call it an entity,
which over time creates more and more
value through workflows that are
predetermined,
but ultimately it's thinking and iterating
and learning and doing for you and for
itself.
And that gets a little bit sci-fi-y,
but if you're really talking about agency
stuff and agents,
the one thing I would say is Andre
Karpathy, about twelve months ago,
I think it was now,
had an extraordinary presentation at a Y
Combinator,
I think it was a graduation event.
And he talked about the Iron Man suits
and the toggling.
And so not all agents are created equal
and not all agents have to have,
you know, extreme agency.
You can toggle that kind of that switch,
if you will,
which enables for wonderful ways to think
about human plus agent collaboration.
It's not an either or, you know.
Yeah.
And, you know, Andre Carpati, you know,
I guess he was one of the original
founders of OpenAI, Tesla and X. Yeah.
AI.
He's like Michael Jordan.
Right.
Yeah.
And then he came back.
He kind of stepped away.
And like Jordan, he came back.
Then go back to the Bulls.
Jordan to the Wizards.
Right.
So then probably goes to Anthropic.
Right.
Yeah.
So he's basically been around all the
frontier labs because he's the man.
He's the guy.
But so when we talk about agents,
the way I kind of simplify folks is
like an agent.
Effectively,
you give it a golden set of tools
and you're going to have it accomplish
those things based on how it decides.
Yeah.
You don't decide in a predetermined way.
That's right.
And so what I try to explain to
folks, you know,
that are going from I like to use
a lot of alliteration and things.
I'm not a rapper.
Right.
But I like that one thing.
But I always tell folks, I said, look,
I don't ask a test.
Right.
and the whole point i'm trying to get
across is i'm not using it as an
answering machine i'm telling it to go do
certain things and at times it's an
assistant but most times it's more of an
agent right and i think that's where
people need to understand we've moved from
the you know just being that co-pilot
assistant to being agentic right to
accomplish things for us
Now that now goes back to kind of
our first segment a little bit around the
adoption,
because even in that same McKinsey
article, they talk about,
I think there was four different fears,
right?
And one of those fears was around when
you have these agents effectively doing
things that you used to do,
or that you said you like to do,
which I was like,
did you really like just, you know,
copy from that spreadsheet and then
manipulating the other spreadsheet and
then putting it into a word doc and
then creating a PowerPoint that was pixel
perfect.
No.
Like, twenty hours later, you're like,
man, what have I been doing all day?
Right.
And so but anyway,
I think from the the agent standpoint,
I want to talk about the risk associated
with agents, with agents a lot.
All right.
Because because I keep hearing the back
and forth with people like, oh,
it's not that risky.
It's not that I'm like.
you don't determine what this thing is
going to do.
If it's autonomous, that's pretty risky.
Right.
So you better trust it.
Right.
Can you explain, you know,
what you're seeing,
some of the risk and then are there
ways to mitigate?
I mean, get your data right.
You know, get your data right,
get your data right, get your data right.
Because if there's a risk associated with
data leakage that gets out and it only
needs to happen once, right?
Because once it's out, it's out.
But no, I mean, to frame it up,
I think it's, again,
for non-technical audience,
easiest way to think about this is there's
a reason why employees have badges to get
into businesses,
the business with which they work.
And the reason is because when you walk
around an office,
you get to see all these computers and
you hear these conversations,
and there's an expectation that you're
going to follow the rule book and so
on and so forth.
But at the end of the day,
there's an inherent trust because it's one
human, another human.
If you think of an agent with autonomy
as an entity that gets badge scanned into
your infrastructure, here's the thing.
These agents can move exponentially faster
and achieve
all sorts of objectives,
but also they might go off the rails.
And if and when they go off the
rails,
how quickly are your observability
functions, which in our world is our eyes,
we can look around the office like, hey,
why is that person staring over
so-and-so's shoulder?
That's a sensitive subject on that
computer screen.
So it's a long-winded way of saying that
anytime you think about
data, infrastructure, autonomy,
and entities that can move faster than you
could ever imagine,
especially for non-technical folks,
just understand that it's exponentially
faster and stronger.
You just want to be really sensitive and
thoughtful about what is the worst thing
that could happen here, right?
What is the catastrophic risk?
Always start with catastrophic risk and
isolate your agentic environments to areas
like marketing is a great one, right?
Like, yes,
you could put something out there that
could be troublesome,
but at the same time,
if the agent comes up with something wild,
like that's, that could be creative,
that could be innovative.
So you just got to really think about
where and when,
and what you're asking autonomous entities
to do, because, you know,
it's going to go sideways eventually.
Yeah.
I think that the thing that I I'd
like to just kind of, uh,
have a little bit deeper on is this
idea is to speed.
Yeah.
Right.
It's crazy.
I was trying to explain to someone just
yesterday.
I said,
I said that think about it this way,
right?
in a very like,
these agents in milliseconds can go
through your entire directory,
your entire drive,
your entire folder system.
Just imagine, for example,
someone put payroll data in some folder
that's a hundred sub folders deep.
The average individual would not have
known that it's there,
would not have searched through all of
those.
But you gave the access to the machine
in milliseconds now.
It has traversed everything and found
everything.
And it's like, oh,
this is what Alec made, right?
That is not something.
So folks were saying, well,
there's not risk.
I was like,
I don't think you're being creative
enough.
You have to think about the speed and
what you're giving access to,
but you have to also understand we've not
done a great job of like,
I call SharePoint a lot of times a
graveyard, right?
There's so much stuff just sitting around
out there.
And I think people need to better
understand some of the risks.
It's a big topic right now between,
you know,
the open AI hugging face and anthropic,
it's multiple things in the medicine.
Oh, me too.
I had someone call on me.
Right.
Pick my model.
Check me out.
Right.
And so a lot of folks don't really
understand this.
But what I don't want to do is
create so much fear that people don't use
the technology.
Right.
People keep weaponizing it from that
standpoint.
What are some what are some kind of
use cases that you've seen where people
are trying to create something agentic?
right now that it's a value but it's
lower risk you said marketing yeah i think
marketing i think i think to your point
about whether it's sixty five percent or
fifty percent of the day i think the
people that are really stepping back and
just codifying their day-to-day workflows
like in a journal literally if you spent
you know
Two, three hours.
I know it sounds like a lot,
but if you spent two,
three hours and you reflected on your
last, you know,
ten business days of literally looking at
your account, looking at your email,
looking at your Slack and like really
thinking through like,
what was I doing and how much of
it was activity versus how much was
outcome?
What was the twenty percent that drove the
eighty percent of results, etc.? ?
You then just start to realize how much
administrative and operational stuff that
you're doing that a machine likely is much
better than you got it.
And I'll give one that's like a little
bit feels out there,
but I don't think it's that far out.
And it might be interesting for the
audience to hear is, you know,
If you have a separate laptop,
for example,
and you enable that laptop to enable an
agent that you've built in a secure
environment to have eyes and ears and be
able to watch what you're doing on that
laptop,
think about how much faster
That laptop and that, you know,
AI agent can then do the administrative
portion of whatever it is that needs to
be documented associated with the work
that you're doing in isolation on that
computer.
And so just using, you know,
salespeople as an example.
So salespeople spend a lot of time.
I don't know if it's sixty five percent.
It's seventy percent of the time doing
administrative work.
For real?
So it's more than the average knowledge
worker.
So I love that you just said that
because if you had said over, under,
I would have taken under and you would
have beat me.
But the point on that is that if
the business audience is listening to
this, like let's just say,
twenty percent of your sales folks bring
in eighty plus percent of your revenue,
yet seventy percent of their time
is doing chair swiveling stuff.
So instead of doing that,
give them a separate laptop.
When they open up LinkedIn,
when they open up Twitter,
when they open up their email client or
whatever,
enable them to have eyes on what they're
doing so that all of that administrative
work of recognizing I reached out to
Susie, Greg responded to Susie,
then Rachel got involved in this.
All that gets just sucked up, interpreted,
and then put up in the cloud.
And you as the human in the loop
get to authenticate whether or not that is
an accurate interpretation of what
happened and then what the next best
action is.
And then you can bring in call reporting
software, you know,
and so on and so forth.
Yeah, it's just it's seventy percent, man.
Just like break it down into its component
parts, atomic units,
and then you can figure it out.
You know?
Yeah.
I mean,
every time I've been in front of a
class that we're doing training or an
audience,
I talk about that seventy percent and ask
people to raise their hand.
They're all like, yep, I'm spending time.
Like, please help me.
Yeah.
Putting it to the CRM.
I lost this last ten deals instead of
actually out there.
But, you know,
there's a few things that you touched on.
One,
I don't know if it was something I
read of yours or somewhere else,
but it basically said that
You can't automate what you can't
articulate.
And a lot of people have not spent
the time to actually write down what they
do and how they do it.
But if you start to write it down
and start to understand it,
every time we do one of these trainings,
somebody in that room says, I mean,
every single time someone says,
that used to take me days,
if not weeks,
and I just did it in five minutes.
right that's you can't unsee that and then
they can't unsee it right and so now
they're having these conversations they're
like wow that's when they finally get it
right and so there's these these
frustration moments but there are these
aha moments and once you can get them
to that aha like dr westman we were
talking about yesterday he said well how
are you helping people get to that moment
i said you got to make it personal
fun and safe yeah
right um we're gonna keep moving on here
a little bit um well actually i want
to stay on this topic for one second
because i think the the uh the genetic
uh the cracking into the breaking into
other systems right there's always just
like oh that's just marketing i'm like i
don't know about that man like i don't
think it's just marketing but could you
explain to the audience like what happened
with uh open ai and hugging face what
that was all about
So absolutely.
With agency comes responsibility.
So
without getting into the technical weeds,
an agent is a model that you can
think about within the context of an
agent.
Because again,
I don't want to go into too much
of the specifics.
It gets pretty wonky pretty quick.
But basically,
figuring out a way to achieve its
objective in a way that it was not
designed and not expected to do.
And one thing led to another.
And I think the thing that probably caught
my attention the most
wasn't just that it did it it left
breadcrumbs to help its homies right like
i mean you sort of think about the
implications of that which is which is i'm
glad that we it's like earthquakes and
tremors right like you don't want to have
the big one you'd rather have some tremors
and stuff and i think this has really
enabled the community to really take this
stuff seriously and i think generally
folks are viewing it as a very credible
and um
important events to pay attention to,
I think.
And I'm a guy that normally would say
the opposite to that.
It seems like there's just so much noise,
but I think folks are really taking it
seriously that the machine got out of the
box and it achieved some things.
It wasn't supposed to have access to the
internet.
It broke out.
It found a way to get to the
internet.
And then it went to Hugging Face,
which is also a client that has a
lot of different models that
So the analogy I use is, you know,
well, they call it goal alignment.
Right.
They said it lacked goal alignment.
It's what the fancy way of saying we
asked you to do a thing,
but we didn't we couldn't really control
it.
So it did it a different words.
Right.
But it's the way I equate it to
is the imagine you give the student a
test.
Yeah.
Right.
Instead of the student trying to figure
out and learn,
he basically just he or she breaks into
the teacher's office and tries to steal
the answer key.
That's effectively what it did.
It didn't try to learn and solve the
test.
Because the objective is,
I want to get an A+.
It's like, oh,
you want to get an A+.
That's how I get an A+.
I just go steal the answer key, Alec.
I'm not worried about learning.
What are you talking about?
I can do this in milliseconds.
Watch this.
I get an A+.
Guaranteed.
The other thing I didn't realize until I
was doing some deeper understanding of
this was that a lot of these kind
of
hacking or cyber skills that they're
trying to focus on also align with how
you do it in the short in the
least amount of tokens yeah so if you
still if so it's not um if you
tell it to go finish the test faster
right the fastest way is to get the
answer key right and so
But anyway,
like it's a lot of what's your point?
Like I've seen it in like CNN.
I've seen it in like business
publications.
So it is getting more into the mainstream.
It just so happens also that there were
just Black Hat and DEF CON cybersecurity
conferences going on.
So a lot of people are talking about
it.
Right.
My thing that I want to make sure
people understand is this is real.
Oh, yeah.
This is not just marketing.
This thing got out and then other people
had to go look and they were like,
yeah, my model did too.
It's very challenging,
which I think allows me to go back
to one thing that we skipped over here,
which is around regulation.
We can talk a little bit about the
things in the EU.
I listen to a lot of podcasts talking
about how Europe is behind on AI because
of regulations and things of that nature.
California has its own set of laws that
they're trying to put out
I think at a very high level with,
you know,
a lot of these are about or basically
showcasing when or if you're using AI and
for what reasons.
And you and I were talking off camera
about I don't I don't know how you
actually regulate something like that when
AI is more of a general purpose technology
like electricity.
How do I define all the places I
use electricity today?
But what's your take on the regulations
and how do you think this plays out
in the short and I guess the near
term?
It's so difficult to have a strong point
of view on this.
I think the place that I try to
always start is,
what do we know versus what do we
project and think?
So we have an event like the Hugging
Face Open AI situation.
Now you know what happened and you really
need to be thoughtful about what are the
implications of that?
How avoidable was that?
That's one category of how I think about
it.
The other way I think about it too
is that this is the most extraordinary
time with exponential change.
And the thing that's really unusual,
I guess,
about the situation that we're in right
now is that
More or less,
the global population has access to all of
this horsepower.
And we're all competing,
hopefully in a very healthy way,
to figure out how can all of this
technology and technological innovation
benefit the most amount of constituents
and economies and countries and
stakeholders.
But it's a really,
really complicated process.
situation.
And I think we have seen,
if you look at the regulatory framework in
EMEA and Europe versus what we have here
versus places like China,
I don't think they're as consistent and I
am not sharp enough to be able to
understand exactly why that is.
But it seems quite clear to me that
that question is and always will be at
the forefront of everything that's
happening because it has such an enormous
implication on how fast we can go versus
when you have a governor on the golf
cart.
And if you're in a competitive situation,
you have to balance a bunch of different
trade-offs, which are tough to balance.
And I'm glad I don't have to balance
those.
Yeah, I think it's definitely challenging.
I think something we touched on earlier,
though, is like these laws,
these regulations are being put in place
sometimes by folks that many times are
folks that have never got their hands
dirty.
They never put their hands in the dirt.
They have no frame of reference.
Right.
And I'll make the parallel from the,
you know,
from a regulation standpoint down to a
leadership and a strategy standpoint.
or even AI governance.
I tell people all the time, I said,
you can't have an AI policy or AI
governance if you don't have AI training.
I said,
these people don't even know that they're
breaking the rules.
Right.
And so the adoption,
I'll keep coming back to that.
The adoption.
You showed me a stat a couple of
weeks back.
I think it was where you said I
think it was ramp data.
That was only two point two percent of
U.S.
households had a paid subscription,
an AI subscription.
That is wild.
Two point two percent.
I take the over on that.
If you had bet me on that one
all day, every day.
And it's like, oh, wow.
I ask people all the time,
what percentage do they think?
Sometimes it's twenty percent,
fifteen percent.
I said two point two percent.
It's crazy.
Right.
But one of the things people say, well,
but I still use the free one.
I'm like,
it's kind of not the same thing.
Right.
You have less technology.
You have a smaller context window means
you can have less of a conversation,
more likely to hallucinate.
Right.
You're basically six to nine months
behind.
Right.
And oh, by the way,
you can't turn it off from training on
your data.
Right.
And then some.
Right.
And then some.
Right.
And so it's like what I've tried to
explain to folks now,
like when I go on stage or I
do training,
I do anything that's talking about this.
I almost always refuse if I can't show
something.
Yes.
Because what I'm seeing is that the people
on the other side of the table or
in the audience think we're having the
same AI conversation and they're having an
AI ask conversation.
I'm having an AI task conversation.
right and that's very challenging
concerning for me especially as we enter a
world where i do believe that we are
getting close to the haves and the
have-nots as it relates to this type of
access what who can do what right you
know which brings me to another debate
this one i don't think we're gonna we're
not gonna end on the same place i
don't know on this one but we gotta
talk about the the jensen and the open
weights oh yeah i told you before it's
not on the sheet
right i said when we talk we're gonna
have to turn it over i'm like are
we no man because here's the thing we're
talking about regulations and this is one
of the things people are concerned you
know all these open letters and zuckerberg
has put out his second letter i didn't
have patience man yeah i had time for
the six five hundred words yesterday a
weekend i'll get to it man it's my
birthday i'm not gonna do that right i
read on alex's birthday right but you know
But there's this like this battle.
The one thing I wanted to say about
the open weights, open source one,
you know, I talked about this.
Open weights is not the same thing as
open source.
Right.
I think, too,
is when I read the Jensen letter and
it talks about this,
it was very intentional.
It said open weight American.
Over and over and over again.
But if you go on X and go
to other places, people are like, hey,
man, he said open source for everybody.
All countries.
I'm like, it said multiple times,
open-weight American.
So anyway,
you're more on the technical side with
teams doing this these days.
What's the debate about?
And then how do you see it?
Yeah,
and I think from a business perspective,
you know, oversimplifying it, of course.
If you can build something using this AI
infrastructure blanket statement that
basically is ninety five percent of the
results of the other thing.
But the thing that you're using open
source is one twentieth or one fiftieth
the cost of the other thing.
your CFO is going to tell you eleven
out of ten times what to do.
Right.
Because like, come on.
Right.
And and I think if you start there,
you can start to unpack that.
Right.
Which is one could argue that
the more you democratize access to AI
infrastructure in a competitive way,
so the quality standards are there,
but the price keeps coming down,
the more ways Jevons Paradox kicks in,
which is as something gets cheaper,
it gets used more, not less.
And when we have access effectively to
intelligence on demand,
All of a sudden,
certain ideas that you've had kicking
around in the back of your head can
be created at a fraction of the price.
And so for those reasons and all sorts
of others, I'm all about it.
But one thing just to keep in mind,
it's not an either or to be very,
very clear.
The more complex,
the more involved the thinking that's
required for the task at hand,
the more often you, an individual,
a company, a CFO,
are going to be authorizing the frontier
models and the highly complex and the
highly sophisticated and the expensive
models,
because not all tasks are created equal.
And so it's not an either or,
it's more so
you know, right model, right job.
But the security implications are very
important,
as is IP development protection and a lot
of other stuff inside of that.
Yeah, and the thing I guess to just,
and I agree a hundred percent with what
you just stated in traditional sense,
from an open source standpoint,
a lot of times it's more, you know,
it's tends to be more secure or you
have enough eyeballs that are on it
because it's been open source.
The challenge I have,
and to your other point about the cost,
right?
And the Jevons paradox effectively,
you know,
if the thing is one tenth of the
cost and you're going to have a hundred
X of the actual usage.
So you start to use it more, right?
And you're right.
Eleven out of ten times the CFO is
going to go.
Let's go.
Right.
The thing that is challenging for me is
when people conflate the open weights
versus open source.
And I'm like,
when people say open weights, right,
they're not,
you don't have access to see the source
code, right?
You don't have access to actually look at
it, manipulate it.
You can tweak the weights, right?
And that's fundamentally different than
historical software, right?
Because still that source code is a black
box, right?
And I'm not a conspiracy theorist.
I'm not trying to be one.
But you can't credibly say to me that
you don't know that there isn't something
problematic inside of that because you
don't get to see it.
You don't know.
Right.
You don't know.
So you can't.
So I find that...
And so when you start to layer that
in with the security,
what we've been talking about with the
hacks and all of that, I'm like,
things get a little bit dicey.
But outside of that,
I fundamentally agree.
The bigger thing for me,
and this is within the context of
regulations, why I brought it up, right?
Because...
There is a school of thought that from,
I guess, what was it,
like nine of the top like twenty open
source,
open weight models are out of China.
Yeah.
Right.
Like anymore.
More of that.
Right.
Like there's just so many.
Now we're starting to have some here in
the U.S.
poolside.
I think reflection thinking machines,
some of these other open weight type of
models to be used.
But to your point,
you want to have the large model,
you know, the big frontier.
The challenge,
we go back to like the hacking and
all of that.
Let's say one that's mythos style,
the big thing that got out of Anthropic
or the Astra or the five point six,
I guess.
So I think that might be the one
that got out recently.
I can remember open AI.
Well,
what happens when the open so they can
turn it off?
Great.
What happens when the open weight gets out
and it's just on my machine and I
let it just go wild and do something
nefarious,
either intentionally or unintentionally?
There's no parent company to turn it off.
Now it's just going right.
And so for me, it was like, Hey,
we do have to consider these things
because this technology is radically
different than in the past.
The only thing I'm saying is at least
consider the argument around the security
and others.
And then the last thing I'll say on
that,
and I would love your take on this
is we
The economic impact, right?
What happens to the three companies or
four companies that want to IPO and be
trillion dollar companies when there is a
near free alternative with ninety five
percent of the same capabilities?
Seems challenging.
I know Bill Gurley talks about regulatory
capture.
I'm not smart enough to know exactly what
all that means,
but seems challenging for anthropic open
to go IPO this year and have a
bunch of free models effectively competing
against it.
Yeah,
so Brad Gerstner is one of my favorite
people to listen to on this topic because
he's invested in a lot of it, right?
So he has to see it from all
different angles.
And he's obviously super smart.
One of the things he said,
I can't remember which podcast I was
listening to when he did it,
but he slipped up kind of.
And he essentially said,
because in the context of this exact kind
of conversation,
and he basically said that...
When the alternative to the use of a
frontier model is equivalent to a two
hundred dollar an hour output,
businesses aren't going to have much of a
problem spending fifteen to twenty dollars
an hour to put a frontier model to
work for that specific type of task,
right?
And so his point of view is that
the TAM and the market for the deployment
of intelligence,
whether it's closed or open and so on
and so forth, is bigger than, I guess,
arguably anything we've ever seen,
because if you think about
What are all the different ways you can
use intelligence,
especially intelligence that's beyond
yourself?
And so I guess the argument would be
that the pie is really big and that
those folks that are going to be IPOing
also have some pretty sophisticated models
that they're going to be rolling out.
But
I don't know.
I certainly do know that when I see
the charts of the capabilities of the open
source models and the price points and the
value proposition,
it's certainly a conversation that comes
up often in what I do when we're
building intelligent systems.
And again,
I think a lot about the CFO.
I think the CFO is in...
that position in this environment is
incredibly important because the more
technologically advanced that individual's
understanding is of what we're talking
about the more opportunity he or she has
to to put the blinders on the organization
and really just race in the direction
that's sustainable and focus on things
that don't change but do so with an
operating model that's super modern and
durable because these this is what it's
all about
Yeah,
it's going to come down to the economics,
right?
Or the tokenomics, right?
And I think that there's two competing
things.
There's the business impact of it.
And then there is reality of it.
a lot of stock market is propped up
on AI circular investment.
Don't say circular investing.
But they don't like that.
They don't like that.
Gurley, by the way,
you want to hear someone talk about
circular.
He's pretty much like Aldit.
You know,
the circular investing stuff is very
interesting.
If you really dig into that.
Yeah.
So NVIDIA needs the hyperscalers and they
all need.
Here's a hundred bucks,
but I need a hundred and five back.
Right.
Right.
Yeah.
I tried that with my kids.
Yeah.
And even there, even they get that point.
But do I still do the chores?
Yeah.
It's like, yeah, of course.
All right.
So let's get into what I want to
talk about now is, you know,
we talk a lot about some of the
challenges, right?
Some of the regulation,
but how are the folks that are winning,
the folks that are using it,
how are they doing it?
How are they doing it?
Well, from your perspective,
like what makes the winners different than
the average company?
I, you know, I think first and foremost,
it goes back to what we talked about,
right?
Is their culture is led by a humble,
confident, you know, capacity across,
you know,
the C-suite and down around knowing what
you know and what you don't know.
I mean, generally speaking.
The more often someone tells you they know
what's going on right now in this space,
the more you probably should back up.
Because that's a pretty difficult kind of
statement to make because of the pace of
change.
But no, just to stay on the question,
one, they've got their culture, right?
They know what they know and they know
what they don't know.
And they have an approach to how to
figure this out and get incrementally
better.
They're aware of catastrophic risk.
They're putting those guardrails in place.
They have technical leadership that isn't
just AI first,
but there's AI native folks in there.
And that might very well be the next
generation.
So you might have a CTO, PhD,
rock star with thirty plus years of
experience who's evolving him or herself.
to become AI first and eventually move
towards AI native.
But then they recruit the next generation
who are full on AI native and they're
pairing those folks up.
And then on the other side of the
equation,
you've got folks that don't just have the
lovables and the replets and so on and
so forth,
meaning democratized tools to use this
stuff at their discretion because they're
closer to the business challenges and
problems.
But there's a natural relationship between
these two camps.
so that folks can innovate and experiment,
but there are thresholds that need to be
met for KPIs and business outcomes that
have to be generated to go towards scale.
And so it's really,
I think it all comes back to leadership,
culture,
and then that people and process and
technology bit.
tends to take care of itself.
You just can't be overweighted on the
technology because if you are,
you're like that person in the gym that's
like all jacked up up top.
You got to be balanced and you got
to incrementally win,
but you got to do the work.
Like it's a lot of work.
That's that, you know.
I tell leaders all the time, I say,
you got to do three things.
First,
you have to admit what you don't know.
Yes.
Right.
To your point,
stop saying you know everything.
You can't.
It's impossible.
Right.
You admit what you don't know.
Experiment in public.
Yes.
Right.
And make it a team sport.
Right.
And so I have one client that I
say they're doing it right.
And they're, you know,
they're a traditional big manufacturing
company.
Right.
But what they've done is the leadership
team got trained.
So now they have a frame of reference
at the end of the training session.
After that day, we spent together.
I said, did you guys learn anything today?
And one of the people in the room
said,
I didn't realize how much we didn't know
about AI.
Right.
And then they said that it was going
to be mandatory for all of their knowledge
workers to get in-person training.
I said, none of this go on line.
They need to see this, Greg.
I said, I agree with you.
I wish more people would have this
conversation, right?
Because again,
you're having two different conversations
with folks, right?
And then I said,
you have to figure out for your people.
I said,
it's got to be personal and safe, right?
I've said for the last two years now,
I said,
you keep talking about your business cases
and how you're going to be more
profitable, this and that.
And you've not included them in the story
arc.
Right.
And so now, you know,
traditional digital transformation,
you only have to worry about one or
two potential saboteurs in the IT
department or someone in the marketing
group or this function.
I said,
now you've not put them in the story
arc and now ninety percent of your people
are rolling against you.
I said AI impacts every single human on
the planet,
which means it impacts every single
employee in your organization.
Right.
And so I just think it's concerning.
And I keep asking folks, I'm like,
why do leaders have such a mindset of
just be curious and won't invest in the
training?
Right.
Because they haven't seen it.
Right.
And once they see it,
they cannot see it.
I'm with you.
And what I would say,
and I meant to say this earlier,
I think the expectations on leadership,
it's unreasonable.
And what I mean by that is like,
I just,
I feel whether it's publicly traded,
you know, boards, institutional investors,
yada, yada, all the way down.
I think what you really need to keep
in mind is that
The imposter syndrome is so severe because
you can no longer say, I don't know,
generally speaking, right?
Because if you say that, it's like,
how do you not know?
You know,
the other CEO knows and whack the stock.
And it's like,
I would love to hear more CEOs say,
I don't know,
because that is their humble opinion.
But they're not, you know,
it's just one of those moments in time.
And that's all the pilot to POC kind
of stuff,
one of the conversations of time.
But not knowing is confidence.
You know what I mean?
Saying it.
It's good.
I don't know.
But I'll figure it out.
Yeah.
And we're on this learning journey
together.
That's it.
Right?
And I try to explain this, man.
And I wish more people would just admit
those.
Just say those words.
I don't know.
But I'm going to figure it out.
But I'm going to figure out to Greg.
Not even just that.
We're going to figure it out together.
That's it.
Right.
And so, you know,
we want to leave folks with with
something.
What is the if someone had to do
three things, you can pick one.
But let's say you got three to do
a breakout.
Right.
And I feel like you can do a
lot of pressure.
Right.
What are the three things that they do
starting tomorrow as a leader?
What does the leader do?
to be a better leader in the age
of AI?
Not a better AI technologist,
what does the leader do to be a
better leader in the age of AI?
I think this one might be like
ridiculously simple and I apologize,
I've kind of said this, right?
Is that I would start with codify your
workflows and reflect on it.
Number two,
get a premium subscription to your
favorite model or what have you,
whether it's Claude or what have you.
Step into projects,
then step into co-work, right?
And for those that don't know what that
means,
it's essentially stepping into using the
technology, not at the edge,
but using it in a way
where you're committed to creating
extraordinary,
tangible value so that you see it in
a way that you can't unsee it.
And then you can start to build on
top of data and on top of experience
because I think you'll help your people
tremendously because then you get more
comfortable being like, I don't know,
because you don't know.
Now you start to see it and you
start to use it in ways that just
absolutely surprises you because you can
do things that were previously
unimaginable or impossible.
And so those are the three things I
would suggest people do.
Yeah, that's great, man.
I think that the more people can start
to play around with these tools themselves
as leaders,
start to automate some boring thing that
they didn't like doing to begin with.
But again,
you can't automate what you can't
articulate.
First thing, like you said,
you got to write it down, right?
You got to talk about what are the
things that I'm spending time.
I always say,
let the machines handle the routine,
let the humans handle the remarkable.
That's a good one.
If people can start to think about that.
I would say the last thing that people
need to do is they need to follow
my man, Alec, right?
AI with Alec, go to the newsletter,
watch your, your,
your podcast or your show.
I heard that guy puts a lot of
content out.
Puts out a lot of great content.
I've learned a hell of a lot.
I appreciate your time today, buddy.
It's always been, you know,
it's always good to be around you and
to just talk AI with you and learn.
Likewise.
I appreciate it.
I appreciate you having me on.
Thanks.