Latest / Founder Mode / Fire Your Worst Customers

Transcript

Fire Your Worst Customers · Founder Mode

0:00
The other big, big news of the past couple weeks, maybe past
0:03
month is this anthropic mythos model that came out right.
0:07
And they're like, it's too good. We can't give it to all of you
0:10
because bad actors are going to use it to hack every, you know,
0:12
it found risks and cybersecurity issues in every single major
0:16
software in the world. So now we're sharing it with
0:18
Apple and Google and and the government etcetera.
0:35
Hey, welcome to founder mode. Yeah, this is the podcast for
0:38
builders who don't wait in the regular line.
0:40
Yeah, so we got another. We have another solo episode
0:43
here, you and me just hanging out.
0:45
It's been an eventful week and so lots, lots to talk about.
0:48
My observation I shared with you earlier today is Opus 4.7 feels
0:52
a bit janky and you're like, well, do you know what's
0:55
happening? And you have, you have.
0:57
Apparently you are aware of what's actually going on with
0:59
Anthropic. I don't know if I'm aware.
1:00
I think I have some theories that are well informed by a
1:02
bunch of people on the inside. I think I mean the, the no
1:05
secret secret is it like Anthropic is deeply out of like
1:08
compute? And so they're like scavenging
1:11
for compute to like feed the tokens.
1:13
And so you're seeing it everywhere, right?
1:15
Like I think 47 is thinking less and and sort of jumping in there
1:19
quicker to kind of save money, to be quick.
1:21
Like even some of the things that was funny.
1:23
Another guy this morning ran a prompt and it says, just so you
1:25
know, I'm not using your context and memory in this, I'm just
1:27
answering this. So it's like explicitly
1:29
shortcutting. It's shortcutting to save.
1:32
Tokens. And, and I don't see that as
1:35
much on the APIs versus a subscription.
1:38
And then a rash of bans, like you see them all over my account
1:41
got BAM, my account got banned. So I think there's a lot of
1:43
people that clearly are doing bad things that are like
1:44
wasteful, but I think there's, you know, anthropics just coming
1:47
down, unlike people that aren't as profitable for them because
1:50
they just don't have the compute.
1:52
And then I think that's sort of leading to like why the new
1:54
models not released is less around the it's secure.
1:58
Hey, everybody wants to check their security before this thing
2:00
gets out and starts finding vulnerabilities, but it's just a
2:03
very, very expensive model. But typically every new model
2:05
that these foundational companies have launched have
2:08
just astronomically increased the amount of compute that they
2:11
use. They're not actually getting
2:12
cheaper to run initially. And then you sort of come out
2:14
with mini models and smaller models and stuff that are more
2:17
optimized for a compute and cheaper.
2:19
But that that sort of first release it, you know, there's
2:21
worried it's going to melt the servers.
2:22
And so I think that slowing it down.
2:24
But now we're actually seeing them sort of like pick and
2:27
choose who they want to work with.
2:28
Yeah. And so that's the first topic I
2:30
want to talk about today, is this idea of like we talked
2:32
about in the prior episode, you know, good, good human, bad
2:35
human, good AI, bad AI. And what about good customer,
2:37
bad customer? Like is it is it the customer's
2:40
fault who's eating all of the tokens and maxing out their kind
2:43
of unlimited subscription? Or is it Anthropic's fault for
2:45
maybe pricing that incorrectly or not going and building
2:47
servers three years ago? Yeah, who's to blame?
2:51
Yeah, no, it's a good one. It's funny, there's, you know, a
2:53
bunch of corollaries there. I think there's, I think it's
2:55
both, right. I mean, yeah, they probably
2:57
hindsight, right? They probably wish they would
2:58
have kicked off a bunch of like data center builds themselves.
3:01
I think they're licensing and reselling and buying a bunch of
3:03
more on the data center side than others for the good
3:05
customer, bad customer. The difference is with
3:07
Anthropic, it's like, well, I can just go and use a bunch of
3:09
different models. I probably didn't install
3:11
something on my computer that, you know, makes it too hardcore
3:14
that I can't go away from it. So I think, but the end result
3:17
is no different. They're trying to find like
3:18
what's the marginal cost for the the right people play and then
3:21
what's that sort of internal cost for them?
3:23
I mean, the subscriptions have always been the same, right?
3:26
How are the people with low usage subsidizing the the few
3:29
that are maxing it out? I think this sort of technology
3:32
is way easier for the few to Max it out than before, and I think
3:37
that's what we're seeing them sort of like grapple with.
3:39
But yeah, pretty exciting times. And I think your point has been
3:43
like, yeah, but the API, like compare it to paying a person
3:46
who can't do a fraction of that amount of work.
3:48
Like it's still worth it to spend 50 or 100 bucks a day on
3:51
the API. And our mutual friend Alex Finn
3:53
posted the other day on Twitter, like, if you're spending 1000
3:55
bucks a month on on API tokens, it's worth it if you're if
3:58
you're getting huge leverage and value from them.
4:00
Yeah, I, I think and part of it's like, you know, what is
4:04
your, what is the cost of that learning or investment of not
4:07
of, you know, like we've got folks on the team, they're like,
4:09
oh, I hit the limit, I've got to wait.
4:11
And I was like, wait, if you're not on the $200 plan or for sure
4:15
upgrade, like if you're waiting at 20, definitely go to 100.
4:17
If you're waiting at 100, go to 200.
4:19
And surprisingly, that obviously solves most of the problems.
4:22
And it's like, man, when you're in that flow state, you know,
4:24
you've talked with this like the developers, you get in the flow
4:26
state and you're like, hey, if you break flow, it's going to
4:28
take a long time to get back to it.
4:30
It, it's like for $100 a month, like at the hourly rate of most
4:34
of these people, it's insane to not like click the upgrade
4:38
button, especially, you know, in these cases where it's like
4:40
you're expensing against our company, you know, the company,
4:41
we're like, hey, the company is willing to make sure that you're
4:43
not looking starved within recent, right?
4:46
I think we have a few people that are doing lots of things
4:49
and we've looked at that and we've actually started using a
4:51
lot of the API through bedrock on Amazon.
4:53
And there, you know, very quickly the costs run up and
4:56
we're like, Hey, wait a minute, is that job something we really
4:58
need? Can we run batch?
4:59
There's a thing that a lot of people don't know about on the
5:01
API. You can run batch mode where you
5:03
basically submit the jobs and they'll come back in a while,
5:07
right in hours or a day or something.
5:09
And for a lot of stuff, like if you're only running it once a
5:11
week, just run it Sunday night and kick it off.
5:13
And you know, when you wake in the morning, it'll be good and
5:15
you're not necessarily worried about that.
5:17
And then the costs are way cheaper.
5:19
But so I think there's going to be ways to go look at that.
5:21
But in the short term, it's like, man, if you're worried
5:24
about the cost of token, more than likely you're not building
5:27
something interesting enough if you're not just willing to sort
5:30
of pay for that, right? Well, then there's all these
5:31
companies, right, that are token maxing, right?
5:33
You know, there's leaderboards internal, I think Meta took
5:35
theirs down. But who's using the most tokens?
5:38
I think you owe a buddy who's using the most tokens inside of
5:40
Google. Like it's their companies that
5:42
get it are understanding that no, no, no, we want to spend.
5:46
I think Jensen Huang from NVIDIA said if, if my engineers aren't
5:49
spending a quarter $1,000,000 a year on tokens, then there's
5:51
such wrong. Sort of like a dollar per salary
5:53
thing or something. Yeah, Yeah, I think that the
5:55
Google store is good. I mean, they're incentivized.
5:57
Not I mean, it's jovial, right? But they're like celebrating
5:59
people that use lots of tokens and for good reason that, you
6:03
know, as a kind of a model company themselves.
6:05
They're trying to figure out like, hey, what are the good use
6:07
cases here? And also like having extreme
6:09
users internally gives them great test data to figure out,
6:12
like how they can optimize it, but also how they probably
6:14
should price and sort of segment some of the use in the future.
6:17
Yeah. I think the VIP line at the
6:19
club, I always think of two ways is it's the line outside creates
6:23
that excitement. I mean, this is the whole wait
6:25
list strategy. I mean, you're a marketer,
6:26
right? Like, how do you create a wait
6:27
list to, you know, scarcity for something?
6:29
And so like a little bit artificial scarcity because I do
6:31
think, I mean, there's nothing worse than waiting in a line at
6:34
a club and then going inside and realizing the club's empty and
6:37
you're like, yeah, that's doesn't feel good.
6:40
And I think there's that sense here where some of the VIP miss
6:44
of like, hey, who's getting picked on, you know, which
6:46
accounts are they keeping? Which accounts are they banning?
6:49
And I think, but I think there's other cases where it's like,
6:50
hey, you're just in a bad neighborhood.
6:52
You know, I had a buddy who created an account, you know,
6:55
was just running tons of scripts against it, got banned and then
6:58
started created a new account instantly that count count got
7:00
banned because they fingerprinted as the same user.
7:03
And so it's like, OK, like, I get it.
7:05
You were just like saying I'm a clean sheet and start over.
7:07
Like, you know, AI kind of remembers, right?
7:09
Anthropic remembers in this case.
7:11
And so. Pepper Farm remembers.
7:13
All right, So strategically, how do we fire bad customers?
7:16
Like I think Anthropic's doing kind of a bad job of this,
7:19
frankly. Like, I think the, Oh yeah,
7:21
sorry, we're banning you. You're doing malicious stuff.
7:23
And then when you e-mail them back, they don't respond for
7:25
days or at all. And if, if they do, they, they
7:28
don't give you like a direct answer.
7:29
I think the, from my experience, like if you have customers that
7:33
are bad, I've fired customers before.
7:35
And like, you know, I've worked in consumer apps where we ban
7:37
users and, you know, I've, I've done that kind of stuff.
7:39
And you know, the, there's a whole kind of, there's a whole
7:42
crisis comes PR and communication strategy around
7:45
this stuff where you have to have a really solid playbook and
7:48
set of FAQs that you can roll out and communicate around.
7:51
Versus like, Oh yeah, we didn't invest enough infrastructure and
7:54
we mispriced this thing. And now we're just turning
7:56
people off and sending them a very like boilerplate e-mail.
7:59
That's not how you do it. Yeah, I think it's tough.
8:01
I mean, you know, we're, you know, Yahoo and obviously
8:04
Microsoft and for those consumer companies like emailing support
8:08
at Google or Yahoo, you know, when your e-mail account got
8:11
blocked or yet it's like essentially a black hole because
8:14
it's like when you're giving out a free service or a very low
8:17
cost service, but you can't afford to millions, it's very
8:20
hard to build good support. And so I think part of that's
8:22
probably a, you know, sophistication of the team and
8:24
like where they are in terms of like they're just overwhelmed.
8:27
For a pretty good AI, you, you, you're kind of thinking about
8:29
how you stratify and segment customers so that you're focused
8:33
on the the big ones that matter, right.
8:35
Like how do you, how do you think about that in kind of this
8:37
more enterprise SAS AI business model?
8:40
Yeah. For us right now, it's small,
8:42
right? The company's small.
8:43
And so we prioritize 2 things. I think 1 is obviously you want
8:45
to work with big customers that are, that are going to lead to,
8:48
you know, to like a larger deal from a dollar economic point of
8:51
view. But honestly, it's, it's the
8:53
ability to like deploy quickly and learn that's actually the
8:56
highest priority. And so in many cases, we'll put
8:59
more resources on a smaller team or a smaller customer, smaller
9:02
opportunity that has this sort of like agility and like desire
9:06
to move quickly with AI, where some of the larger companies may
9:10
like, Hey, they want to test really slow and take their time.
9:13
We don't learn as much. They don't learn as much.
9:15
And partly just because the ability for them to adapt and
9:18
change is just not there as a large organization.
9:20
But I think the other thing is we've definitely, you know, we,
9:23
we had there's a price like we don't do, it's not for heat.
9:26
Like there there is, you know, we do free pilots and free
9:29
trials and things like that to get people up and running.
9:31
But the like once you say yes, like it's kind of expensive and
9:36
you know, relative to like, oh, I just want to go buy the
9:38
cheapest like voice AI bought in the space, Like there's way
9:41
cheaper ones. And that price sort of says,
9:44
like, are you willing to sort of like align and commit and still
9:47
like, we're losing money on like every deal, you know, when
9:51
we're, you know, because you're building them for free and
9:53
they're, they're going to pay off in the long term.
9:55
But it's like, OK, but once we do get this thing deployed,
9:57
you're going to pay a proper like, sort of like premium price
10:00
so that we can then spend the time to keep investing in that.
10:03
And it also signals to us that like, Hey, as a company, you're,
10:07
this is important to you, right? Like the old joke in enterprise,
10:09
I always say, it was like, you know, you're like, oh, I got an,
10:12
you know, RFP or I got this thing and I want to go do this
10:14
pilot with this customer. And it's like, great, you need
10:16
to get 10 grand paid upfront. And if you can't go and sell
10:19
what you think is going to be a $1,000,000 deals in the
10:21
enterprise, and if you can't even find somebody to put 10
10:24
grand on their corporate credit card or like their, you know,
10:26
discretionary budget, like you haven't even reached the person
10:29
that can sign 10 grand, you're very far from the person can
10:32
sign the $1,000,000 deal. And it was amazing like how well
10:36
that advice kind of in the web two world played segmenting and
10:40
just sort of understanding like where are you at in the sort of
10:42
like hierarchy of sort of decision makers inside a
10:44
company? And I think there's a bunch of
10:46
that, you know, that can be chosen the other way to sort of,
10:48
you know, self select who's the right one to get.
10:50
The other big, big news of the past couple of weeks, maybe past
10:54
month is this anthropic mythos model that came out right?
10:58
And they're like, it's too good. We can't give it to all of you
11:01
because bad actors are going to use it to hack every, you know,
11:03
it found risks and cybersecurity issues in every single major
11:07
software in the world. So now we're sharing it with
11:08
Apple and Google and the government, etcetera.
11:11
How much of that from your perspective, do you think is
11:13
real? How much of that is like we, our
11:15
service couldn't possibly handle this?
11:16
Like where? Where is the line there?
11:18
I think it's an incredible marketing moment, right?
11:21
Because like no one else is out there marketing, their AI is so
11:24
good that you can't even, you know, this is like talking.
11:26
About the club you can't even go into.
11:27
You can't, this is like, you can't even apply like there's no
11:30
application button. You have to go find somebody in
11:32
the club. Yeah.
11:33
I mean, if you're like, oh, I want to be in the security beta,
11:36
it's like cool, just go find the guy at Microsoft or at Cisco
11:39
that's going to like, you know, bring you in like you're not
11:41
getting in. And I think so I think it's an
11:43
incredible, like story and marketing message that
11:45
everybody's talking about. I also probably believe that
11:48
there's a deep capacity issue where that the capacity of that
11:52
that was required to roll. That model.
11:56
Out incredible, yeah, super big, yeah.
11:58
And so I think there's that as well.
12:00
And then it's also this just like incredible, you know, sort
12:03
of again, the government thing is also the marketing moment of
12:05
like how he's out there saying, hey look, you guys can't have
12:07
it, we can't give it to you. We we have to play, you know,
12:10
pick whose side we're on in terms of who we can offer this
12:12
to. It's it's like great marketing,
12:14
right? Opening eyes.
12:15
Not out there doing that today. That is not out there, Google's
12:18
not out there. It's like everybody's rallying
12:20
behind. They seem to be playing 40 chess
12:23
for sure. Yeah, I was talking to one of
12:25
these brokers that sells secondary shares.
12:27
You can buy open AI all day long on the open market.
12:29
You can't buy Anthropics literally not for sale right
12:32
now. Everybody's like waiting for the
12:33
next round and holding. So like where you normally can
12:35
trade it, you know, some percent above the last round and there's
12:38
people always willing to move shares.
12:40
Not right now. He's a impossible to buy
12:43
anthropic. You can go get some of the other
12:45
ones, but not anthropic. And so that's just, again, even
12:48
the internal folks are feeling this like sense of like, wow,
12:51
this thing's made a step function change.
12:53
And they're definitely like from a mind share to the one to beat
12:55
right now. And on the cybersecurity front,
12:57
I mean, like some of it has to be somewhat real in that like
13:00
these models are just going to get better and better at kind of
13:03
finding zero day vulnerabilities presumably.
13:06
Even just go run 4-6 or four seven against your code base and
13:08
it will find. A bunch of stuff, yeah.
13:10
And I think that maybe is the take away for counters that are
13:12
listening, right. It's like, if you haven't done
13:14
that, like let, let you know, let cloud code loose on your on
13:18
your code base. Then the other thing that I that
13:20
I wanted to make sure we talked about today was on the on kind
13:23
of AI in the enterprise, which I feel like is a lot of what you
13:26
and I talk about these days. And, and we're both having these
13:28
conversations with VC and PE firms.
13:31
How do we how do we bring the latest and greatest tools into
13:34
these kind of bureaucratic, slow moving and very security
13:39
conscious environments? Right.
13:40
There was a great Aaron Levy, you know, the founder and CEO
13:43
box tweet this past week. And you know, he's talking about
13:46
after going kind of on a listening tour and talking to,
13:49
you know, banks near retail, healthcare, consulting, tech,
13:51
sports. And like there's just a bunch of
13:53
stuff that he found that is like the through line, you know,
13:58
change management being really hard, token maxing that we just
14:01
talked about fragmented legacy systems.
14:04
You know, people, it's not about replacing humans with agents,
14:07
but kind of up up leveling and giving agents things to do that
14:11
people in the company were able to do Headless software.
14:13
There's the Benioff announcement this this past week of of
14:16
headless 360, the new sales force thing.
14:18
Like where, where are you doing all of that journey and what are
14:21
you thinking about there? Yeah.
14:22
I mean, a lot of Healthcare is arguably, even though a lot of
14:25
them are smaller from like it's not enterprise by size in many
14:28
cases like terms of employees, you're not thinking like 10,020
14:31
thousand. I mean, the hospitals are that
14:32
way, but like most of the like private practices or hundreds of
14:35
employees typically, but they act like enterprises in a lot of
14:38
the way, right? They're not thinking about
14:40
replacing jobs. It's like, what are all the
14:42
things that I could do if I had infinite amount of time, Like,
14:45
oh, I'm not calling all these patients back or I'm not
14:48
processing this sort of paperwork or this request at the
14:51
rate that I would want to, or I'm just ignoring unpaid bills
14:54
of this size. But if I had an infinite amount
14:56
of time or a very low cost way to do that, I would.
14:59
I think the worry that a lot of them have, and I think where
15:01
they're stuck is that, you know, you talk to these CIOs and CIOs
15:04
and CE OS to some extent, they're getting overwhelmed and
15:07
just diluted with people telling them they're going to add AI.
15:10
And it's not just all the new vendors that are coming into the
15:12
space. It's all the current vendors
15:13
that are like, Oh, I'm your phone company.
15:15
I'm going to add AI agents to my phone.
15:17
I'm your Internet provider. I'm going to add AI agents.
15:20
You know, they're just everybody's selling an AI agent
15:22
for something. And I think that is deeply,
15:25
deeply confusing for people that are like not living and
15:28
breathing this every day. They're like, who do I trust?
15:30
And so part of it's like a trust thing.
15:31
How do you, how do you sort of like crack through that and then
15:34
show them something that gets their confidence?
15:37
We're like, Hey, this is going to work for me.
15:38
I'm going to get out of this kind of like, you know, weird
15:41
question stance, but also like can I get a pilot that actually
15:43
works that I can deploy and run it?
15:45
So I don't know. I mean, from the marketing side.
15:46
I'm curious though, as you guys sort of approach, you work with
15:49
a lot of big enterprises on the marketing side, like how do they
15:52
think about AI? Yeah, it's interesting.
15:54
I keep hearing these like CM OS say, oh, we're going to do this
15:57
like AI week, we're going to do an AI boot camp, we're going to
15:59
get people using Cloud Co work. We're going to kind of learn the
16:02
tools and I think that's great. Like I'm not, it's not a bad
16:06
thing that trying to get folks that are that are not using them
16:09
today. But I think that kind of misses
16:11
the, the meta point, which is AI is not going anywhere.
16:16
It's not about like spending one week trying to learn tools.
16:18
It's like, how do we integrate all this stuff into our workflow
16:22
permanently? And then how do we RIP and
16:24
replace huge pieces of the kind of marketing technology sack and
16:27
you know, our tech stack in general with AI native AI first
16:31
tools. And then how do we go higher AI
16:33
native AI first people. And I think that's where the
16:35
folks that are like really winning, like I keep, I know we
16:38
keep talking about Anthropic, but like for I think 8 or 10
16:40
months, they had one growth marketing guy on the team for
16:44
all of cloud and he was using cloud code to just build
16:46
everything. And it's like, if they can get
16:48
that much leverage at a company that's now worth, I think 800
16:51
plus billion dollars that, you know, what the what legacy
16:54
enterprise could do if they were to embrace this technology
16:57
broadly and like put everyone through like I remember back in
17:00
the day and I think it just got acquired by mural, but like
17:02
Brian Belfer and reforge, those guys.
17:04
Like that was the growth boot camp.
17:07
Like that was the thing that people would go to.
17:08
I don't know if there's an equivalent yet, but like there
17:10
needs to be the AI boot camp that is the equivalent of forge
17:14
or General Assembly or one of these where you go in and maybe
17:17
it exists and I just haven't heard about it yet.
17:19
But like you get, you get re educated and reoriented to
17:22
become AI native. And like your first thought, and
17:25
you know, I talked a lot about this is your when, when you have
17:27
a thing to do, the first question you ask yourself is
17:29
like, how can I use AI to do this?
17:30
Like and like, it's not even just how can I use AI to help me
17:34
do this. It's like, can this task be
17:36
automated fully where I'm now the human in the loop and I'm
17:38
reviewing the output and then I'm going back and then doing
17:40
multiple turns. I just talked to you about a
17:42
kind of a deck that I produced the other day that was like
17:44
many, many turns using Claude, But ultimately I did not write a
17:48
single slide. I just told it what I was trying
17:50
to do. We worked on strategy together,
17:52
we worked on the contents, we worked on design and just went
17:54
back and forth turn, turn, turn. And so I think the, the TLDR for
17:57
this is like as a, as a growth guy and as a marketing guy, if
18:00
you're running an organization that's trying to, to go to
18:03
market faster and ship with the velocity that these new kind of
18:06
AI native companies are shipping, you have to completely
18:09
rewire the way you think and the way you operate.
18:10
And you have to use the latest and greatest tools and it's,
18:13
it's chaining them together and, and, and kind of automating
18:16
workflows so that your team can level up in a huge, huge way.
18:19
Yeah, I know it makes sense. I think the boot camp things
18:21
interesting. One of my companies, we're
18:23
starting to do some of these boot camps and launch of a
18:25
conditional marketing and it'll be interesting to see how that
18:28
works. I think the first ones focused
18:29
on webinars and you know, just helping people market in like a
18:33
more sort of direct case because I think you've seen this, right,
18:35
like the old days of like just running organic SEO as your only
18:39
thing, it's not going to work. The old days of only running pay
18:41
per click, not going to work. So it's like, what is the AI
18:44
sort of modern Go market? The other thing that we talk
18:46
about a lot that came up in in Aaron's tweet was this kind of
18:49
unanimous sense that everyone is working more than ever before.
18:52
And I think you and I, you and I talk about this all the time.
18:54
It's like one more prompt, like just want, like just keep, let's
18:56
hit it one more time. What can we do to kind of up
18:59
level like Saturday and Sunday is, is, is for playing with our
19:01
third party harnesses and like creating these kind of new tools
19:05
and building stuff like you and I built and you built and I kind
19:08
of took over the, the, the thumbnail generator for the
19:10
podcast. It was like a 45 minute, you're
19:12
sitting on, I think your couch on your phone doing it.
19:15
Like, it's wild what's possible here.
19:17
And it's like, we can do so much more now.
19:20
And so it's not about just the volume of work, but like what is
19:23
actually worth spending your human brain and leverage on, and
19:26
how can you point your tools at the most important next thing?
19:29
Yeah, yeah, I think. And the magic is like, you know,
19:32
one way was like it's like a video game and you're like level
19:34
almost to one more level. I want to go fight one more
19:36
boss. I want to go somewhere.
19:37
It's like there's just this like desire and urge to kind of click
19:40
one more time, tweak one more time.
19:42
And it's like insane, right. But but also like kind of fun,
19:45
right in the same way where it's like it doesn't feel like work.
19:48
And they always say like, I'll go find your passion or find
19:50
your, you know, your little mood muse or whatever that word is
19:53
like. But what they really want, it's
19:54
like, man, you just want to find something where you're like,
19:56
damn, this is like the best Co work teammate where you're like,
19:59
man, we're just jamming and we're like, whatever you're
20:01
talking about. Even if you're not fully
20:03
specific or fully confident in what you can go do.
20:07
You know, you always can just take the next step.
20:08
And it's like that ability to constantly keep popping it, you
20:12
know, makes it really hard to, to not stop, right?
20:15
So what's the take away for founders that are trying to
20:18
build for the Enterprise right now?
20:19
What do you think? Yeah.
20:21
I mean, I think for a founder building for these big companies
20:24
today, I think you need to find like something that's like your
20:27
wedge. And I think it needs to be
20:29
something that's sufficiently hard that they just can't do
20:33
that thing internally by just grabbing it and throwing it in
20:36
cloud code or throwing it in Chachi PT, right.
20:38
And I think the one, the example I always give in Healthcare is
20:41
like, you know, doing real time AI voice that's deeply
20:44
integrated into your EMR. So in our case, Athena one like that
20:48
is not a vibe codable project for a bunch of reasons, but
20:51
adjacent stuff is like you can go vibe code, some of the
20:54
website stuff you can go vibe code.
20:55
You know, faxing was a big thing that was in medical.
20:58
It's like it's pretty easy to take a fax and put that in
21:00
ChatGPT and be like extract all the fields and you know, give me
21:03
an XML file or a Jason file. And so my thing is like, pick
21:06
something sufficiently hard where you sort of get value for
21:10
doing something technically hard.
21:11
And then and then I think the piece that we took, I mean, a
21:14
good six months to learn was you have to find a way to get that
21:18
into their day-to-day workflow in the least amount of like
21:21
steps from their side. So you can't be like when the
21:24
old days it was like, OK, great. We were like, oh, we're going to
21:26
take all your call recordings, analyze it, build a full custom
21:29
prompt library and all these things.
21:30
And that worked. But it required so much work by
21:33
their part to like organize their data, give us all their
21:35
training guides, give us all these calls.
21:37
And we're like, man, we've got to simplify it.
21:38
And so we like, we look at your business and we're like, OK, we
21:41
think these are the five modules that we're going to go deploy.
21:43
Just give us two or three answers.
21:45
Like when somebody calls with a headache, is that urgent or not?
21:49
On certain departments, it is is because they just had an
21:51
epidural and that could be a serious side effect in primary
21:54
care. It's like, no, it's cool.
21:55
You can just take two Tylenol and come in in the morning and
21:57
we'll work out, you know, and I think understanding that as a
22:00
very simple example is how like you go from like it's a ton of
22:03
work to bring this new AI project into an enterprise for
22:05
it's like, no, no, it's actually going to be really easy.
22:07
We're going to actually take work off your plate.
22:09
We're going to adapt to the way you work today.
22:11
And so when you're presenting it, you're not selling them a
22:14
product or a company. You're like, hey, I'm selling
22:16
you a solution, I'm selling you a transformation.
22:19
I'm going to get you from where you are today to the next step.
22:22
And it's not piece that we've found now the most success and
22:26
sort of just getting in quickly and then keeping it small enough
22:28
where there's shared success. Like you don't want to be like,
22:30
oh, this AI thing's dragging on forever.
22:32
It's like, no, no, it's kind of working.
22:34
How do we do the next thing? Oh, let's turn it into a little
22:36
mini sprints and every, you know, three to four weeks we're
22:38
launching a new little workflow or something else.
22:40
That's yeah. I would, I would say it's like
22:42
pick a, pick a good wedge, find a hard problem and find
22:45
something simple to start with as like the summary of what you
22:48
just said. Kevin, that was a fun one.
22:49
We covered a lot, but the through line for me is that
22:52
compute scarcity isn't an accident.
22:54
It's it's actually strategy. The anthropic is picking
22:56
customers. The Mythos model is the best
22:58
marketing moment in AI right now and nobody else is even trying
23:00
to play that. Game.
23:01
Yeah, I mean, the piece that hit me is how it trickles down,
23:04
right? Compute gets rationed at the
23:05
top. You start making decisions about
23:07
what a token's worth and if you're hitting the limits on the
23:10
$20 plan, you just need to upgrade and you know, it's that
23:12
is your flow state. 200 is the cheapest higher you'll ever get.
23:16
You're not going to be able to find a human for 200 bucks a
23:18
month. Yeah.
23:19
And on the enterprise side, the this myth that the one AI week
23:22
at a company can fix anything, the companies actually went
23:25
rewiring themselves end to end. Anthropic had one growth
23:27
marketer and I'm Claude. That.
23:30
That's the bar, right? Yeah.
23:31
And if you're a founder trying to sell into enterprise, stop
23:33
selling the product. I mean cell transformation.
23:35
Pick a wedge that's sufficiently hard that they can't just vibe
23:38
code themselves, and then make it super simple for them to plug
23:41
it into their day-to-day. All right, Kevin, what's your
23:43
top five from this one? Yeah, I think with the first one
23:45
we said it computes the new scarcity.
23:47
I think the second, if you're worried about token costs,
23:50
you're probably not building something interesting enough.
23:52
And then, you know, artificial scarcity can be a marketing
23:54
strategy to work for clubs, so why not work for AI and tokens?
23:58
AI week just doesn't rewire the company.
24:00
And then finally, I mean, dude, for enterprise founders, you
24:02
know, it's got to pick a hard enough wedge.
24:04
If this episode made you rethink how you're thinking about
24:06
tokens, customers, or the way your company is approaching AI,
24:09
send it to your CIO, your CMO, or the founder you know who's
24:12
still waiting for permission to build?
24:14
Yeah, and if you're not subscribed to Founder Mode yet,
24:16
let's fix that. Leave us a review, share it with
24:18
your team, and come find us on the newsletter for more.
24:21
That's it for today, because the question isn't whether you're
24:24
spending enough on tokens, it's whether you're building
24:26
something worth spending them on.