0:00

On today's show, we're covering all the

0:02

AI questions you've been too afraid to

0:05

ask. What we're doing is breaking down

0:07

with Meyer Gupta, the CMO at Cracken,

0:09

one of the best crypto exchanges in

0:11

the world. We're going to go through,

0:13

how do you use AI? Which model

0:15

is best for which task? And what's

0:18

the best strategy for AI adoption in

0:20

your company? You're going to leave this

0:22

episode with a whole new framing for

0:24

how to move forward with AI. Let's

0:26

get to today's show. We right back to the

0:28

show. We right back to the show. But first,

0:31

a quick word from our sponsor. Remember when

0:33

marketing was fun. When you had

0:35

time to be creative and connect

0:37

with your customers, with Hub Spot

0:39

marketing can be fun again. Turn one

0:42

piece of content into everything you

0:57

know, enjoy

1:00

marketing

1:05

again.

1:08

Visit

1:12

hubspot.com

1:16

to get

1:20

started

1:22

for free.

1:26

being here. So uh

1:28

kind of reintroduced you to our audience. I

1:30

think you were our very first guest, so

1:32

you've had a pretty incredible career. You're currently

1:34

the CMO and Chief Growth Officer over at

1:36

Cracking, one of the most impressive companies in

1:38

the crypto space, but just in a very

1:40

impressive company in general. And we wanted to

1:42

get you on because you are a forward

1:44

thinker, you were hitting up growth in Spotify,

1:46

you've led marketing and growth in freshly, just

1:48

a ton of incredible roles, usually in companies

1:50

that are actually trailbrasers in their industry. So

1:52

of course we wanted to get you on

1:54

to get you on to talk about to

1:56

talk about to talk about. But yeah, because

1:58

we want to know how. smart leaders and

2:00

just smart people in general are using

2:03

AI. So maybe we could just kick

2:05

off, would you give in our audience

2:07

a little bit of like context on

2:09

where you are seeing AI get used

2:11

across your market and growth teams. What

2:13

are some of the use cases that

2:15

you particularly find pretty interesting? Yeah, well,

2:17

let me, maybe if it's helpful, just

2:19

give context on what the growth theme

2:21

a crack and looks like. We are

2:23

obviously like many teams still experimenting. Some

2:25

areas are more mature in using AI,

2:27

some we are still learning, but The

2:29

growth theme at Cracken now in his

2:31

current stage, which I call our third

2:33

era in a way, is pretty much

2:35

end-to-end growth. So it has marketing. We

2:37

have a pretty decent-sized growth analysis and

2:40

research team, product design, product engineering, obviously

2:42

product-led growth as well. And to be

2:44

honest, the two areas where we started

2:46

using AI and experimenting was around creative.

2:48

both from a brand and storytelling standpoint

2:50

as well as performance. A lot easier

2:52

on performance of course because that's where

2:54

you need a lot of velocity and

2:56

variety and you know AI does a

2:58

great job in giving you those variations.

3:00

We're also a global organization you know

3:02

huge footprint in Europe where there are

3:04

so many different languages so you know

3:06

translation has been a huge area for

3:08

us both in product and off product.

3:10

But I've been really inspired by our

3:12

research team which is a pretty small

3:14

unit but they've been leveraging. AI quite

3:17

a bit for both call and conned,

3:19

you know, using platforms like bolt AI

3:21

for some of the call work. They've

3:23

been using pole fish. Again, I personally

3:25

haven't dug in a lot on these

3:27

platforms yet, but I know there's a

3:29

lot of internal inertia to see how

3:31

do we bring more efficiency, more velocity

3:33

in doing stuff that was taking much

3:35

longer in the past and also increase

3:37

the level of quality that we get

3:39

in some of these functions. What are

3:41

you using it for actually on brand?

3:43

Because we do get a lot of

3:45

folks on here when Kipp and I

3:47

have guests on. It is a lot

3:49

of performance marketing. It is a lot

3:51

of like informational content. I would love

3:53

to hear because crack and do a

3:56

ton of smart stuff on brand campaigns.

3:58

Thank you guys. have a formula one

4:00

or part of the formula one as

4:02

well so like what you what are

4:04

some smart things you do on brand?

4:06

Yeah it's harder you know what we

4:08

feel and we have this discussion at

4:10

least three or four times a week

4:12

we have a channel on slack where

4:14

we are brainstorming looking at all the

4:16

new tools that are coming in and

4:18

brand I think it always is where

4:20

AI is helping us get smarter and

4:22

come up with more ideas idea generation

4:24

versus actually coming up with the end

4:26

asset also because a lot of a

4:28

marketing right now has pivoted to being

4:30

very product focus. We are highlighting the

4:33

value proposition and the RTBs for our

4:35

core product, the interfaces, you know, we're

4:37

trying to show how customizable and flexible

4:39

our dynamic interfaces are for our crack

4:41

and pro product, for instance, which is

4:43

one of the best in the category.

4:45

And purely with AI, it's very hard

4:47

to focus in on a graph, focus

4:49

in on dashboards. So what we are

4:51

learning is. It's great to give us

4:53

ideas, it's great to give us different

4:55

variations, but when we are zooming in

4:57

in certain angles, that's where you still

4:59

need your more traditional way of creating

5:01

content. And the other place where the

5:03

team is really moving fast is sometimes

5:05

you have long form content, like we

5:07

have a lot of influences and KOLs

5:10

that we work with, but the team

5:12

is using different types of AI platforms

5:14

to create. shorter versions of that content

5:16

to drive faster distribution across social so

5:18

a lot of clip anything type of

5:20

you know platforms that you use to

5:22

create you know 100 different assets from

5:24

one long form asset that you may

5:26

have with an influencer which then you

5:28

can distribute it much higher velocity. Meyer

5:30

I like to add in on that

5:32

I think what's missed in the brand

5:34

creative product marketing side of things is

5:36

that historically there's just been a lot

5:38

of time and money spent trying to

5:40

guess is this idea match what the

5:42

person who I'm trying to communicate with,

5:44

like, does it match what they actually

5:47

want? And I think what you're saying

5:49

and what we found at Hub Spot

5:51

is like, AI is a fast and

5:53

cheap shortcut to that problem. We use

5:55

Claude internally on the Hub Spot marketing

5:57

team. Everybody has a license to Claude.

5:59

We have clawed projects and so we

6:01

have a whole project just all around

6:03

our core buyer persona. And so any

6:05

time we're writing a product page or

6:07

building a brand campaign doesn't matter what

6:09

it is. We can just ask basically

6:11

a fictional version of our customer what

6:13

they think. Give us feedback. What resonates,

6:15

what doesn't. And you have to spend a

6:17

lot of time and money on focus groups

6:20

and market research and stuff. And now the

6:22

cycle times can just be so much faster it

6:24

feels like. Are you guys seeing that? Yes, yes,

6:26

you know, obviously this conversation is lighting

6:28

me up to think about what other

6:30

things we could actually do in our

6:32

world almost right away. You know, one

6:35

element of what you mentioned, Kip, which

6:37

I think is different that AI brings,

6:39

different from a traditional model, is, see,

6:41

when you're trying to figure out how

6:43

your customer will respond, a creative team

6:45

or one of your subgrowth teams, they're

6:47

engaging with an internal research team to

6:49

understand, okay, tell me how, you know, what the

6:51

feedback could be. they're going to market, they

6:54

run the study, they run call in coin

6:56

and they come back. Not only speed and

6:58

velocity, but this is a direct hook

7:00

that now the growth teams have where

7:02

they don't have a dependency on another

7:04

layer. So it also helps them get

7:07

a deeper understanding of what the customers

7:09

may respond to and also bringing more

7:11

agility because you are now doing that

7:13

not once you've had an MVP of

7:15

an asset, but you're actually doing

7:17

it while you're in accepting in much early

7:19

on in the phase. Yeah, I think research

7:22

in general, I think you mentioned that

7:24

there is like one of the most

7:26

interesting use cases. The latest releases from

7:29

Deep Think, we covered 03, their

7:31

research capabilities. They actually are

7:33

displaced in most research roles, but we

7:35

have covered it before that there is

7:37

a really interesting way for these AI

7:39

models to replicate your customer, which we're

7:42

talking about, like how it can be

7:44

the voice of your customer, because it's

7:46

basically trained on the data of the internet,

7:48

but some of the recent agents like the

7:51

operator agent or another one that we're going

7:53

to cover soon on this channel called proxy

7:55

can actually help you laser in on parts

7:57

of the internet. So you could say like

7:59

go to G2. cried, go to these other

8:01

things, and use that data to replicate

8:03

my customer. And so then you're able

8:05

to actually have a back and forth.

8:07

There's a company that I'm an investor

8:09

in called Hyperband, and they actually do

8:11

this for companies where they'll build an

8:13

agent to replicate your customer in a

8:15

bunch of different scenarios. And then ramp

8:17

and sales reps can call that customer

8:19

and have a sales conversation and train.

8:21

on calling that customer. So I do

8:23

think it's a pretty interesting use case

8:25

in the future, which is, you know,

8:27

we've had multiple shows where today you

8:29

have this billion dollar, I think it's

8:31

multi-billion dollar industry, I don't know if

8:33

you know it as, because it's like

8:35

three billion dollar, this market research industry.

8:37

We looked it up before, it's more

8:39

than that, it's like eight to ten

8:41

billion, it's massive, massive, the market research

8:43

business. And like can you pay a

8:45

thousand people to like come back in

8:47

your data to like come back in

8:49

your data? ability for AI to just

8:52

do that en masse I think is

8:54

going to displace a lot of that

8:56

need as well and so I do

8:58

think there's a future where you'll be

9:00

able to test your creative test your

9:02

brand campaigns and get pretty instantaneous feedback

9:04

but you'll be able to pair that

9:06

with like internal data as well to

9:08

like be able to make sure that

9:10

a marketer can get things right more

9:12

than they get things wrong. Yes I

9:14

think On the research side, we're absolutely

9:16

seeing that. The pace at which we're

9:18

getting feedback, you know, is dramatically different

9:20

from when we are using a traditional

9:22

model. Now, one area where, you know,

9:24

we've taken a certain path where we've

9:26

created local instances, let's say, of chat

9:28

GPD, because there's confidential data, you know,

9:30

we are obviously for the nature of

9:32

the business that we are in. How

9:34

much are you guys seeing that being

9:36

done versus most brands and businesses leading

9:38

on all the data being in the

9:40

cloud? So one of the things that

9:42

we are doing a crack in is

9:44

we've created our local instances where all

9:46

the unstructured data is actually being pushed

9:48

into that local instance and is being

9:50

trained. And then, you know, you're just

9:52

asking a lot of questions and queries,

9:54

even just to understand user behavior, because

9:56

over a period of time, historically, these

9:58

have been PDFs, decks and docs with

10:00

tons of data, but I think creating

10:02

a local incentive just give that access

10:04

to thousands of crack nights pretty instantly.

10:06

Yeah, this is actually a super important

10:08

point. I think Kipp and I have

10:10

talked with this. The number one way

10:12

to make this very impactful within your

10:14

company is to be able to provide

10:16

the ability to collapse all the unstructured

10:18

data into a repository, that then you

10:20

can just have the employees easily access,

10:22

like so when they build things, it

10:24

can access that unstructured data. Obviously, Kep

10:26

and I are kind of maniacs, and

10:29

so we have... That's the nice way

10:31

to say it. We have... That's the

10:33

nice way to say it. We have

10:35

personal pay plans, I think, to every

10:37

AI tool, but we do have local

10:39

instances as well, very similar to the

10:41

way that you guys sound like you

10:43

guys sound like you're building. The unstructured

10:45

data unlock is crazy, right? And if

10:47

you look at some of the reports

10:49

with Google 2.0 flash, its ability to

10:51

extract data and insights from PDFs for

10:53

essentially 40, 50 cents is like. unreal

10:55

and kind of unmatched. There's two things

10:57

we've decided recently to do in our

10:59

team, Karen, right, which is one, we're

11:01

in the process of having technical writers

11:03

document all of the workflow so that

11:05

we can then be very clear on

11:07

what we're going to automate and what

11:09

we're going to automate now, what we're

11:11

then going to automate in the future.

11:13

And then the second thing is every

11:15

meeting's recorded. Yeah. Every meeting's recorded. And

11:17

we are creating like a whole nomenclature

11:19

for how those recordings and transcripts and

11:21

transcripts are stored. everything's recorded. We are

11:23

just going to unlock the creation of

11:25

unstructured data. And it was sweet, like

11:27

I had a big offsite last week

11:29

with a few people here and you

11:31

were there for a little bit and

11:33

we recorded everything. Then you dumped all

11:35

that in Claude and we could go

11:37

and basically build follow-ups and everything from

11:39

that. It's really great. So I think

11:41

those are two things that we're doing

11:43

here. And I think the other thing

11:45

we could talk about for everyone that

11:47

I don't think has been talked about

11:49

yet that Meyer kind of brings up.

11:51

of one of these frontier models? When

11:53

do you baby build a app on

11:55

top of their APIs that's specific to

11:57

your company? when you're trying to stand

11:59

up AI use cases within your company,

12:01

do you build that custom? Do you

12:04

have some opinions on that idea? I

12:06

want to kind of hear what you

12:08

think and hear what Meyer thinks there.

12:10

Okay, that is actually a pretty good

12:12

question, right? The one that I thought

12:14

about much more deeply is, I will

12:16

make sure, is this the same thing,

12:18

which is, when you're trying to stand

12:20

up AI use cases within your company,

12:22

do you build that custom? Do you

12:24

use a vendor? open source and non-open

12:26

source, like do you commit to a

12:28

certain model? I think that's one of

12:30

the harder things to figure out right

12:32

now, because that's like one of the

12:34

things that I'm trying to figure out

12:36

all the time across our go-to-market, which

12:38

is when you were trying to stand

12:40

up things in AI, the most important

12:42

thing is to get signal as quick

12:44

as possible, because it's really hard to

12:46

know what the capabilities are of the

12:48

model at scale. And so there's two

12:50

things there. I still think in a

12:52

lot of cases you do have to

12:54

customize it a lot to your needs

12:56

because I think AI works best when

12:58

it replicates workflows that are unique to

13:00

how you do things, right? The best

13:02

AI companies actually are shipping daily because

13:04

they're sitting with design partners looking to

13:06

see how these people work and then

13:08

ship into their needs to integrate AI

13:10

seamlessly into the workflow. And so you

13:12

actually in a lot of cases have

13:14

to build custom off-the-shelf software is even

13:16

really hard to get signal. I would

13:18

still always want to start with off-to-shelf

13:20

because I would want to try to

13:22

prove the use case as quick as

13:24

possible versus committing to building something internally.

13:26

The open source one and the closed

13:28

source one is actually a really interesting

13:30

question. So I've always believed open source

13:32

is going to be a really interesting

13:34

question. So I've always believed open source

13:36

is going to be a big part

13:38

of how companies use, based upon the

13:41

use case you want to do. So

13:43

you can imagine you're trying to. do

13:45

something in the future and you're using

13:47

some sort of interface and that interface

13:49

basically is able to capture the intent

13:51

like the use case you want to

13:53

do and then in behind the scenes

13:55

it can orchestrate and send to the

13:57

right model dependent upon your needs, right?

13:59

And that does not exist. Like I

14:01

actually try to invest in some companies

14:03

doing that, but it is actually a

14:05

way harder problem to solve than I

14:07

first thought. So I don't admire if

14:09

you have opinions on that, but that

14:11

there would be my like off-the-top opinions.

14:13

Yeah, this is a great point. I

14:15

think it's a multi-dimensional framework to be

14:17

honest, because there are so many different

14:19

layers to figure out what your strategy

14:21

should be. On one hand, I think

14:23

it's a call walk and run. I

14:25

think the lowest hanging fruit for any

14:27

business in any team is leverage a

14:29

SAS platform and experiment and prove incrementality

14:31

and then you figure out the path

14:33

forward. I think the path forward is

14:35

determined I feel based on a couple

14:37

of things. One is that core to

14:39

your core product offering. If it is

14:41

core to your core product offering you

14:43

want to own the IP, you want

14:45

to build it. Yeah. And that is

14:47

core within the product. So that plays

14:49

a key role. Second is, you know,

14:51

the security and confidentiality of that data.

14:53

And if you're playing around with that,

14:55

no-brainer, I think most of the SAS

14:57

platforms, most of the cloud-based solutions today

14:59

will fulfill most of your needs. And

15:01

if there is a very unique edge

15:03

case, then it's a big question, Kiran,

15:05

I do feel you have to ask

15:07

is, hey, is the cost-benefit analysis strong

15:09

enough where you go down that path

15:11

where you customize or you build an

15:13

adaptive layer on top of it? Or

15:15

is it reasonable enough that it can

15:18

solve 80% of your efficiency? and then

15:20

20% are edge cases. But the local

15:22

instance point that I made earlier, it

15:24

is because there is a certain type

15:26

of data that we just would not

15:28

want to leave our four boundaries, and

15:30

we felt that, okay, we can create

15:32

a local instance, continue to train it

15:34

over time with all the unstructured data.

15:36

So I think there are three or

15:38

four different layers, and I feel either

15:40

which way you got to start with

15:42

experimenting with what's out there, and absolutely

15:44

see some of the early signals. Yeah,

15:46

I agree. Look, I think to chime

15:48

in to try to help everybody watching

15:50

and listening to today's show, what's gonna

15:52

happen is a lot of companies are

15:54

gonna be like, oh, I'm behind, I

15:56

wanna get into this AI game. And

15:58

they're gonna go by business seats of.

16:00

GBT, of Claude, of Jim and I,

16:02

pick a core frontier model. And that

16:04

is a great place to start. Really,

16:06

really good place to start. There's nothing

16:08

wrong with doing that. What you will find

16:10

is, if you have very specific

16:12

use cases, building small applications on

16:15

top of the API, is like

16:17

orders of magnitude cheaper. Yeah. Right, like way

16:19

way way way way cheaper. And so if you are very focused

16:21

in what you're trying to do, I think you're

16:23

going to be better off to build custom. If

16:25

you are general use cases trying to figure that

16:28

out, I think probably going and buying those seats

16:30

is probably not a bad place to start. When

16:32

you buy the seats, you get a bunch of

16:34

features and those features continue to improve as they

16:37

roll out and make the product. better, but if

16:39

you have like very focused use cases where there's

16:41

a couple of very specific things you want to

16:43

do and use AI for you are far better

16:45

to build off of an API like a basic

16:48

web app or agent because it's going to be

16:50

way way way cheaper and then I suspect

16:52

if you're hosting things locally and you

16:54

have specific use cases or data privacy

16:56

concerns, that's probably where open source is

16:58

going to come in. You're going to

17:00

take an open source model, you're going

17:02

to run it locally, you're probably going

17:04

to pick one that's best for the

17:06

very specific problem or small set of

17:08

problems you have, right? And that's I

17:10

think probably the three sets of choices

17:13

people are making today. Right. Yeah. I

17:15

do have a question from you guys

17:17

because you of course have been digging

17:19

so deep into AI and talking to

17:21

so many leaders. From an operationalational

17:23

being most effective when AI is

17:26

assessed and adopted and evaluated within

17:28

each team? Or are you saying,

17:30

okay, actually find a person across

17:32

the board, across teams, because look,

17:34

there's so much innovation happening,

17:36

it's also noise, right, because you're

17:39

also running your core business, you're

17:41

trying to hit those KPI, you're

17:43

trying to hit outcomes, but this

17:46

is all relevant. So operationally, what

17:48

have you seen being most effective

17:50

in terms of experimenting and identifying?

17:53

What are the right tools to play around for what

17:55

type of use cases? This is a great question.

17:57

Keep and I spend a bunch of time on this

17:59

last week. So this is really timely. I

18:01

kind of divided it up into three

18:03

parts in terms of AI adoption within

18:05

the company. So we have these like

18:07

top down goals from the company. And

18:09

it's basically just a stake in the

18:12

ground. It's saying like we believe these

18:14

places across the company should be transformed

18:16

by AI because we know where AI

18:18

capabilities are today. But the thing and

18:20

I kept kind of bill for is

18:22

we build the use cases that we

18:24

believe AI is going to be transformational

18:26

for. in the future, regardless of where

18:28

the model capabilities are today. And what

18:30

we mean by that is, I think

18:32

a year ago we talked about this,

18:34

which is the model capabilities is a

18:36

solved problem. It's time and money. Yeah,

18:38

the model capability, don't worry that it

18:40

cannot do what you want to do

18:42

today. Bill the infrastructure and the setup

18:44

to do that thing and the model

18:46

will sell itself. That was 12 months

18:48

ago. If you look what's happened in

18:51

the last five months, that was like

18:53

pretty accurate, right? The model capabilities have

18:55

like been transformative. It's speeding up, not

18:57

slowing down. So I think putting a

18:59

stake in the ground and having large

19:01

bets at the company level and then

19:03

having a large pod that can go

19:05

after those pod that can go after

19:07

those bets that can go after those

19:09

bets. think and what I've seen in

19:11

other companies is it should be run

19:13

like a growth project not an IT

19:15

project you have to have a growth

19:17

methodology to how you approach those because

19:19

AI in and of itself is an

19:21

iterative technology it is not a deploy

19:23

the technology like regular software and there

19:25

you go right it is a very

19:27

like learn as you go then the

19:30

second thing is team enablement and then

19:32

the third thing is employee in general

19:34

enablement so I'll give you somewhat take

19:36

on employee enablement that might not go

19:38

down well with folks right I think

19:40

on the employee side of things The

19:42

job of the team in the company

19:44

is to provide the tools and the

19:46

kind of permission to go and play

19:48

with AI, integrate AI into your workflow,

19:50

figure out what works, what doesn't work,

19:52

like play around with it, here's the

19:54

tools, here's some like courses that you

19:56

might want to go take, nothing is

19:58

mandated, and if you don't understand that

20:00

this is a paradigm shift in how

20:02

you do work, and you don't want

20:04

to integrate it into your work, and

20:07

you are not curious about it, that's

20:09

on you, most important things that you

20:11

can do. The second one is the

20:13

hardest one, actually, and that's the one

20:15

I want to, like, pitch over the

20:17

KIP, which is, what do you do

20:19

at the team level, right? So should

20:21

teams understand how AI can transform what

20:23

they do, or should there be a

20:25

central AI team that goes into those

20:27

teams? spends time with them, looks at

20:29

their workflows, and it is the AI

20:31

specialist team, and it understands how it

20:33

can transform what they do. And I

20:35

think actually either or or of those

20:37

could work. I don't know if there's

20:39

a right option. Here's my take, because

20:41

this is a really good discussion, and

20:43

a bunch of marketing leaders and VCs,

20:46

you know, listen to this on RSS,

20:48

and so I'm sure we'll kick off

20:50

some debate. I think you have to

20:52

have both of what Karen just said,

20:54

of what Karen just said, is the

20:56

what Karen just said, is the honest,

20:58

is the honest, is the honest, is

21:00

the honest, is the honest, is the

21:02

honest, is the to take off in

21:04

a team or within an organization is

21:06

to centralize some valuable unstructured data use

21:08

case. Like the second our product marketing

21:10

leader had a couple projects in Claude

21:12

that was like here's everything we know

21:14

about our persona and you can just

21:16

get feedback from that persona because I've

21:18

taken all the time to train it

21:20

upload all of our decks and docks

21:22

and everything there and I've tailored the

21:25

output and everything for you. She did

21:27

another one about like removing any like

21:29

business jargon and stuff from your copywriting

21:31

like basic things that like everybody can

21:33

use, that is how you get real

21:35

adoption. And so I think the individual

21:37

teams are required to be the experts

21:39

to find and gather the unstructured data,

21:41

put it in the frontier models, Claude,

21:43

ChatGPT, whatever you're using, and make that

21:45

accessible to their team and the broader

21:47

team. I think you then need a

21:49

team of specialists who are experts in

21:51

AI and automation to say, we can

21:53

now. automate 90% of localization. This is

21:55

a solved problem. We got deep amount,

21:57

we got a bunch of different tools

21:59

where we can make localization 10 times

22:01

better and we're gonna go run and

22:04

build very specific workflows, custom software, whatever

22:06

it may be to go and solve

22:08

that problem that like your core team

22:10

isn't going to have enough knowledge. and

22:12

expertise to go into. We can go

22:14

and train that team and do everything

22:16

we need, but I think both is

22:18

the only possible outcome. Karen. Yeah. Yeah.

22:20

Let me tell you about a great

22:22

podcast. It's called Creators of Brands. It's

22:24

hosted by Tom Boyd. It's brought to

22:26

you by the Hubspot Podcast Network. Creators

22:28

are Brands explores how storytellers are building

22:30

brands online, from the mindsets to the

22:32

tactics. They break down what's working so

22:34

you can apply that to your own

22:36

goals. Tom just did a great episode

22:38

about social media growth called 3K to

22:40

45K on Instagram in one year selling

22:43

digital products and quitting his job to

22:45

go full-time creator with Gan and Mayer.

22:47

Listen to creators or brands wherever you

22:49

get your podcast. I totally agree. I

22:51

think it's a hybrid model and I

22:53

was thinking through how are we running

22:55

it and what are the teams that

22:57

are running it successfully? The big outlier

22:59

or the big difference in the scenarios

23:01

is anytime the actual business team, and

23:03

that could be marketing growth product, anytime

23:05

they are the ones generating the inertia.

23:07

They are the ones pushing the use

23:09

case. We are seeing better results and

23:11

outcome. Now, of course, in some cases,

23:13

they'll have a dependency on, you know,

23:15

somebody in a global data team or

23:17

engineering to figure out, you know, a

23:20

local instance of something or create customization.

23:22

But anytime they are the ones who

23:24

are driving and pushing for the outcome

23:26

because they are closest to the efficiency

23:28

opportunity in what they are doing, we

23:30

are actually seeing success. Yeah, I think

23:32

that makes a ton of sense. I

23:34

think curiosity has never been more valuable

23:36

than it is today. I was talking

23:38

to a friend about this earlier on.

23:40

Your time has never been more valuable

23:42

than it is today because the opportunity

23:44

to apply it to AI is so

23:46

huge. Like I think about this all

23:48

the time, which is... I have a

23:50

really high bar for what I should

23:52

spend my time on if it's not

23:54

AI. If I'm doing something during the

23:56

day and it's not AI... I'm like,

23:59

why am I doing this? Like it

24:01

has to be a really high bar

24:03

for this to be important. I don't

24:05

think that's healthy, by the way. I'm

24:07

not saying that's what we should do.

24:09

I actually don't think it's healthy because

24:11

it forces me into a bunch of

24:13

like tailspin. Well, the best quote I've

24:15

heard on this is from Dan Chipper,

24:17

founder of every friend of the pod.

24:19

He was reviewing open AI, many, many,

24:21

and he was like, these advanced reasoning

24:23

models are a bazuka for the curious.

24:25

just like you take out a whole

24:27

big problem really quickly and learn something

24:29

and I'm somebody who's for better words

24:31

addicted to learning and so like that

24:33

just becomes insatiable and you're right here

24:36

you just stop being like why am

24:38

I doing anything that is like a

24:40

bottom 20% yeah thing ever yeah that

24:42

is so spot on and I almost

24:44

feel it's like social you know when

24:46

social became a thing whatever 10 you

24:48

know 14 years back you could not

24:50

just assume that you would understand the

24:52

nuance of social and building communities if

24:54

you yourself were not in it. Exactly.

24:56

Yes. You have to live and breathe

24:58

it to then say, oh, I have

25:00

a chance at building a brand that

25:02

is distributed through social and build communities.

25:04

So, believe it or not, of course,

25:06

I followed both of you and a

25:08

real lot of your content. Late last

25:10

year, as I was getting into the

25:12

break. I actually dove really deep in

25:15

Claude myself with my financial data, personal

25:17

personal financial data. Me and my wife,

25:19

we always end up overspending and you

25:21

know, like it's a cluster F when

25:23

you actually look at your statements, there

25:25

are like thousands of rows and your

25:27

Excel eventually breaks. So you can't really

25:29

do anything. And then you look at

25:31

all kinds of pie charts that typical

25:33

financial systems give you, but it doesn't

25:35

really tell you anything. So I read

25:37

a lot about Claude and I saw

25:39

one of the post from Kieran where

25:41

you had shared or six platforms and

25:43

tools you using. You know, I used

25:45

three or four weeks to get into

25:47

Grock and Claude and Gemini. So what

25:49

I did was I cleaned up all

25:52

my data. I removed all the PI.

25:54

Man, I dumped all the CSP files

25:56

into Claude. And I went crazy asking.

25:58

kinds of questions because why are we

26:00

overspending? Where are we spending? You know,

26:02

and what are the repeatable spends? My

26:04

wife thought that she's overspending on Amazon

26:06

and I corrected that because, you know,

26:08

this clearly said, no, this is how

26:10

much you're spending on Amazon versus, you

26:12

know, all of the restaurants. And we

26:14

were surprised that despite having our own

26:16

car, we were spending so much an

26:18

Uber in a span of three months.

26:20

And then we, so I think the

26:22

only way you unlock AI. in your

26:24

professional life is when you're actually living

26:26

and breathing it and finding incrementality and

26:28

finding those use cases in your post

26:31

on life as well. 100% I think

26:33

personal life is a big one. What

26:35

I do now is everything I do,

26:37

I'll start with AI. And the financial

26:39

ones, that's a pretty interesting one. I

26:41

do something similar where I plugged in

26:43

my portfolio to open AI and then

26:45

asked it, how can I diversify this

26:47

more? I've made my first investment two

26:49

months ago in a clean energy fund.

26:51

I had no idea what this clean

26:53

energy fund was, nothing, like it was

26:55

all recommended by AI, broke down by

26:57

AI, added to my portfolio for diversification

26:59

in climate change, and I just said,

27:01

I'm going to like just invest in

27:03

this and see what happens, right? Because

27:05

I'm so committed to it, I'm like,

27:08

just invest in this and see what

27:10

happens, right? Because I'm so committed to

27:12

it. I'm like, literally putting my money

27:14

into, I've been fascinated by. is how

27:16

the reasoning models have become better with

27:18

strategy. While you're pulling that up, my

27:20

son has like medical stuff and like

27:22

if I get a complex medical report,

27:24

yeah, like I just upload that and

27:26

get like the full real like summary

27:28

of it. Yeah. Before like waiting for

27:30

a specialist is like magic. Exactly. The

27:32

health ones, I have all of my

27:34

health data now in a project. In

27:36

a project, right? Yeah, in one project.

27:38

That's super smart. Just to give an

27:40

example of how incredible AI is, not

27:42

that I was not listening to our

27:44

conversation was not dialed in, but I

27:47

wanted to set this up because I

27:49

wanted to show you. Keep in mind

27:51

when I'm going through this. I did

27:53

this whilst we were talking on the

27:55

pot. Wow. Okay. And the reason I'm

27:57

saying that is because I think it

27:59

would take a growth team a week

28:01

or so to be able to do

28:03

this. Now, one caveat is this is with

28:05

synthetic data because I can't show real data

28:07

on this show. I've been trying to get

28:10

good at trying to create synthetic data to

28:12

show use cases. The synthetic data is just

28:14

not complex. So what I've given it is

28:16

like a dashboard for a growing SAS company

28:19

at 30 million in ARR that wants to

28:21

double its AR over the coming years. And

28:23

so the first prompt is, and if you're

28:25

subscribing to the podcast and you're watching this,

28:28

we'll put the sheet for the prompts in

28:30

the YouTube comments. You don't have to try

28:32

to like look at the prompts. But the

28:34

first prompt is basically just saying, okay,

28:36

well, I need to get to that number in

28:39

12 months. Of all the metrics that you

28:41

see here, what are the three to five

28:43

I should focus on? And I'm going to

28:45

skip past this a little bit because it's

28:47

not that interesting because again, it's synthetic data.

28:49

So the data was pretty obvious that it

28:51

picks out churn rate, gives me examples of

28:53

why it picked out that tactics and tradeoffs,

28:56

picked out ARPO, picked that conversion rate

28:58

free to pay it. And this is

29:00

a PLG company. So it has premium

29:02

tier started tier, pro tier enterprise, activation

29:04

rate, and then referral rate, referral rate.

29:06

Cool. And so then I asked it

29:08

to put together this. And this is

29:11

where it started to get much better.

29:13

Now this is 01. This is not03.

29:15

The reason it's not03. I don't think

29:17

that's a European thing. I think this

29:20

is just an everyone thing. I can't

29:22

upload files or even graphics.

29:24

I can't upload anything. I can't

29:26

upload files or even graphics. I

29:28

can't upload anything to03. So then

29:31

it will take those metrics, right?

29:33

It does its own hypothesis. It

29:35

starts to give you pretty great experiments. When I

29:37

first looked at this, which is just like as

29:40

we were going through things in the podcast, I

29:42

think it's as good as an average growth team,

29:44

if not like a pretty competent growth team, where

29:46

it will give you the kind of experiments you

29:48

can run to improve churn rate. There's one here

29:50

that is actually pretty interesting. So it started to

29:52

do things now that I found it never used

29:54

to do, which is give me something that I

29:57

hadn't thought of. So this one here, early warning

29:59

warning outreach outreach. us to like basically trigger

30:01

usage signals and so when you see someone

30:03

drop in daily active users you take your

30:05

daily active user cohort and then you trigger

30:07

emails to cohorts where you see the daily

30:10

active usage drop off over time yeah which

30:12

is actually a pretty smart thing to do

30:14

one question yeah my mind's buzzing two questions

30:16

actually one is are you also uploading your

30:18

UX and design flow for it to understand

30:21

where the opportunities could be? That's a good

30:23

question. That's one question. Let's start there because

30:25

I'm very, very curious about how to leverage

30:27

this. This was going to be my end

30:29

and point, but I'm glad you brought this

30:32

up, which is, if I actually went through

30:34

this, I actually went through this, and so

30:36

I actually went through this, and so I'll

30:38

go away through this, and so I'll go

30:40

away through this, because I want to cover

30:43

your point, because I was going to think

30:45

more deeply about these problems. and do another

30:47

version you'll get better results better results better

30:49

results each time now there's been a release

30:51

lately on how deep-secret others and these reasonable

30:54

models were managing to get better results and

30:56

one of the key things was just every

30:58

time it thought it was finished giving you

31:00

the answer they would just give it the

31:02

word wait hmm that's it wait and they're

31:05

doing that to say think more and what

31:07

it's doing in the background is it's looking

31:09

through all of the things it's giving you

31:11

and stack ranking them and trying to give

31:13

a better one a better one and better

31:16

one so that's the reason the reason of

31:18

models in the background are trying to reason

31:20

out what is the best answer they can

31:22

give you not too similar from search but

31:24

like I think somewhat more sophisticated and then

31:27

it's giving me a big swing but your

31:29

point is really important because I'm pretty sure

31:31

the more context you give it, the better.

31:33

So the way to actually make this incredible

31:35

is you give it your actual real data,

31:38

which for cracking, I assume, is very complex,

31:40

right? Which would be better because then the

31:42

reason the model would be able to decipher

31:44

the first question of like really tell me

31:47

that metrics that matter in this complex model,

31:49

that is actually really important. The ones that's

31:51

picked here are like pretty self-explanatory. They're just

31:53

like best in class PLG metrics. it would

31:55

be interesting for a business like Hub Spot

31:58

or Cracken, it would pick more interesting things.

32:00

The second point is, okay, we'll turn that.

32:02

into a growth plan, but actually, here's all

32:04

of the experiments we've run. So you have

32:06

a library of all previous experiments, which again,

32:09

speaks to the fact, the most important thing

32:11

to do for AI to be impactful is

32:13

documentation. Born thing you get everything of. It's

32:15

really documentation, and you load in all your

32:17

experiments, and then to your point, which is

32:20

the next version I want to do internally,

32:22

which I think is a great idea. We

32:24

do have a ton of wire mapping and

32:26

flows. from our customer journey. And so I

32:28

would actually just add all of that into

32:31

the context window as well. And I suspect

32:33

you're going to get a first version of

32:35

a growth plan that needs to be edited,

32:37

but does not in any way need to

32:39

be rebuilt. Yeah. Okay. Here's what I would

32:42

love to do. Man, I would love to

32:44

come back in four weeks because I'm not

32:46

giving you guys much other than some abstract

32:48

shit. But all I have to figure out

32:50

today and I'm going to tell you the

32:53

type of growth questions. We are trying to

32:55

answer that any business that is global is

32:57

trying to answer. And I want to bring

32:59

some of those use cases back. But the

33:01

only caveat is I need to figure out

33:04

the local incident because we won't push that

33:06

into cloud. I think that is too scary

33:08

for me to even think that I'm going

33:10

to do it. Doesn't matter what type of

33:12

enterprise. But I get it. Here are some

33:15

no freaking brainer questions that I can assure

33:17

you. This guy. or any of these platforms

33:19

will help us get much faster. One is

33:21

in a global company I'm always looking at

33:23

growth rates difference between geographies. Yes. No-brainer just

33:26

tell me hands down why is activation rate

33:28

stronger here and worse there. Now they're all

33:30

just looking at all the variables all the

33:32

inputs in the activation rate and telling me

33:34

what the differences are. Two we always look

33:37

at okay who's a high value cohort and

33:39

who's a lowest value cohort but trying to

33:41

figure out session analysis to understand Okay, what

33:43

is it the high value cohort doing beyond

33:45

the aha moment? You know, that is driving

33:48

maximize LGBT. Now, all that is doing is

33:50

data crunching. It's just trying to figure out

33:52

patterns, which is what AI can do a

33:54

lot faster than a data scientist would. running

33:56

different queries. And then obviously we focus so

33:59

much on payback period because we challenge ourselves

34:01

to have very strong fully loaded payback periods

34:03

and then starting to look at even just

34:05

basic stuff okay what is the IRR on

34:07

your CAC? You know stuff where we are

34:10

spending a lot of energy which is rather

34:12

mechanical or you are searching for nuggets if

34:14

I can go back and figure out a

34:16

way to dump all my raw data I'll

34:18

come back with some very interesting insights that

34:21

we are generally trying to solve every single

34:23

day, but it's taking us long. Yeah, and

34:25

I'll just give one quick tip there. When

34:27

you get the internal data and you have

34:30

those internal trends, and you can anonymize it

34:32

because I'm going to say use deep research

34:34

here, and you might not be able to

34:36

use that locally, I don't know, but you

34:38

can anonymize it to make you feel better,

34:41

just give it the trends. Parrot with external

34:43

trends in those geographies. Is this exactly what

34:45

I was going to say? Yeah, okay, okay.

34:47

It's deep research plus the similar web API.

34:49

Yes, yeah, exactly. What happens is you get

34:52

all this data internally in any kind of

34:54

company of scale. If you're a 10 person

34:56

or 100 person or 1,000 first company, you

34:58

get some data and you're trying to contextualize

35:00

it. largely on like anecdotal information. And once

35:03

AI helps you synthesize it and find the

35:05

key trends and timing on those trends, you

35:07

can just basically bring in deep research, similar

35:09

with API, and a couple external sources, and

35:11

like, it will correlate very perfectly for you.

35:14

But promise you. You guys have totally changed

35:16

my mindset in 30 minutes. That's what we're

35:18

trying to do. I mean, look, obviously I've

35:20

always had this hurdle to think about how

35:22

am I going to apply it. It is

35:25

always in the back of the back of

35:27

the mind. We are spending so much energy

35:29

right now, we know exactly what questions we're

35:31

trying to answer. We know where we are,

35:33

we know how we can drive growth, but

35:36

the latency in getting those insights is just

35:38

insane right now. What I want to go

35:40

back and do is pick somebody from my

35:42

team to figure out how we're going to

35:44

launch, and then maybe you're right, you know,

35:47

there is elements of anonymized data, which we

35:49

can actually... upload and put patents. An interesting

35:51

thing is in crypto, market behavior is easily

35:53

captureable because the biggest market driver is a

35:55

global Bitcoin price and we have a lot

35:58

of price indexes. So that can eliminate that

36:00

variable and really understand what else is happening

36:02

in a particular geography different from another one.

36:04

Yeah, I would say you have actually incredible

36:06

external trends as well. like I would say

36:09

there's some like really interesting patterns across geographies

36:11

in your external trends because you know it's

36:13

such a vibrant and popular space. Yeah. Yeah.

36:15

I think that's a great place to leave

36:17

it. The last one I'll leave you with

36:20

that I think is like pretty interesting is

36:22

I have all these different project assistants. I

36:24

did one for growth. Now a project assistant

36:26

is basically we have an EA, she's incredible,

36:28

but like to be across every single thing

36:31

you do. that is hard to scale. Is

36:33

it possible? Yeah, I've started building project access

36:35

and for different projects and there is a

36:37

good one for growth, which is if you

36:39

have a Google Jam, Open AI, custom chat,

36:42

TV, or cloud projects, you can just. Literally

36:44

not care about the structure of anything that

36:46

people are giving you as long as it's

36:48

specific to this project Just say like hey

36:50

put all the things that are related to

36:53

this project your updates your experiments Whatever they

36:55

may be in a single decks doesn't matter

36:57

everything in a single folder right named project

36:59

Which is usually a goal like increased activation

37:01

rate by 50% in 2025 everything goes in

37:04

a folder goes into project assistant. It is

37:06

pretty incredible. You can get everything you need

37:08

if you're saying like tell me about the

37:10

three experiments in January that we did the

37:12

one that basically led to the biggest impact

37:15

and why the other two failed. You never

37:17

have to go on this like conundrum of

37:19

slacks in emails to try to figure out

37:21

stuff. And again, these mundane use cases are

37:24

the ones where there's huge upside because if

37:26

you're managing hundreds of people, all of that

37:28

stuff takes hours and hours of your day.

37:30

And that one I have found to be

37:32

incredibly valuable. And the last thing I was

37:35

showing kept earlier is like, it can build

37:37

out interfaces dynamically for you in the actual

37:39

console. example would be,

37:41

anytime I do a

37:43

meeting, I get the

37:46

meeting transcript and I

37:48

said, add all the

37:50

follow -ups into a table.

37:52

And so it will

37:54

keep updating the table

37:57

in the AI instance.

37:59

And I say, remove

38:01

this. And now at

38:03

these ones, I've done

38:05

those. It's just for

38:08

me, you can't share

38:10

it with people easily

38:12

because it doesn't write

38:14

back into a file.

38:17

But it's unbelievable. Like I can't describe

38:19

how much of a game changer these

38:21

assistants are. The key is to attach

38:23

them to a singular project. Yes. And

38:25

that's your repository for the data for

38:27

them. And then they are your assistant for

38:29

a specific project. And then you don't have to

38:31

care about all of this crazy structure. You just

38:34

say, please put all of your updates into this

38:36

folder. Do you like why do I need project

38:38

management software? Yeah, I don't think you do in

38:40

the future. You definitely don't in the future. It's

38:42

kind of wild. Yeah. And actually, even

38:44

if you do, even if you're using it,

38:46

it's not effective. Because as a human brain,

38:48

you're not actually mapping out all the dependencies.

38:50

Yeah, you're not thinking through. And also the

38:52

worst thing, the point you made about, okay,

38:54

what are the last three experiments you did? And

38:57

which one worked and why the one

38:59

that didn't work? The challenge we have

39:01

today in the mechanical operations is we are

39:03

not going back. We are not looking

39:05

at what we learn from those experiments because

39:07

it's too hard. It's too hard. It's

39:10

too hard. It's so hard. Exactly. Yes. So

39:12

it's not only that we are inefficient

39:14

because it's so manual. But then we

39:16

become ineffective because we are not technically applying

39:18

the learning from experiments that were done.

39:20

And imagine if you were not the one

39:22

who did it. Exactly. Then it's even

39:24

harder. Right. Literally, one of the best things

39:26

Kieran, you and I did in the

39:28

last six months was like, we were going

39:30

through planning at the end of last

39:32

year. And we just took some of the

39:34

raw decks and data from just planning.

39:36

And we were just put them in chat

39:38

to you. And we're just like, what

39:40

are all the dependencies we're not accounting for? What's

39:43

the stuff that's going to go wrong that

39:45

we're not thinking about? And you just

39:47

like, that's just been really, really hard. And

39:49

we got like, a great readout, we

39:51

changed a couple things was awesome. And you're

39:53

like, it took like 20 minutes

39:55

and you moved on. It was great. Yeah. You know,

39:57

with all the content that you guys are sharing Like

40:00

what's the best way to get the

40:02

portfolio of the platforms or different type

40:04

of use cases? It's like how do

40:06

you feel if the team's now going

40:08

into the second gear, for example, the

40:10

gear one is okay. Hey, we've tasted

40:12

blood. We've seen it works. Now it's about

40:14

expanding the horizons and really pushing the

40:16

limits. Now, do you guys have a

40:18

bit of a gear to playbook? Like exactly

40:20

what we just discussed. There are a lot

40:23

of names. I was trying to capture some

40:25

of them. And I'll dig in. Any thoughts

40:27

on that? Is that something that you guys

40:29

are helping guiding the industry with? Like going

40:31

from AI foundational use cases to then really scale

40:34

the AI use cases? Is that? Or like what

40:36

model to use for what thing? Is that what

40:38

you're asking? Yes. Yes. What's it what model to

40:40

use for what thing? And that's what's tricky. Yeah.

40:42

I think Claude for anything. Right. If you're marketing,

40:44

just buy Claude seats for everybody. Yeah. Right now.

40:46

Any write-in is like, Claude, even internal stuff for

40:48

execs and stuff, I still run through Claude. Claude

40:50

is a better creative model than open AI and

40:53

Google. Yeah. And a very good coder still, by

40:55

the way. From what everyone tells me, because I

40:57

can't delineate between all three, they all seem pretty

40:59

great at coding, Claude is like the preferred model

41:01

for codeine, and people don't know why. I don't know

41:03

if you've seen that, but people. I don't know why

41:05

it's so good, but it's so good, but it's so

41:07

good, but it's just like, but it's just like, like,

41:09

like, like, like, like, like, like, like, like, like, like,

41:11

like, like, like, like, like, like, like, like, like, like,

41:13

like, like, like, like, like, like, like, like, like, like,

41:15

like, We had Scott on who leads product for Entropic

41:17

and he was asking like, why do you love Clot

41:20

so much? I was like, well, it's like, why do

41:22

I like one friend better than the other? I don't

41:24

know. I just like gel better with Clot, right? It

41:26

just gets me, it understands me, it just knows me.

41:28

The O3 Open AI models are definitely better for strategy.

41:30

And so if you take a growth, that example, I

41:32

showed a growth dashboard and took the

41:34

data and then asked for a strategic

41:36

doc, I still find Open AI in

41:38

particular the recent models and you can

41:40

see on the benchmarks, it's just better

41:42

at strategy. Google Gemini is not far

41:44

behind and I need to actually get

41:46

deeper into the Gemini tutored releases. The

41:48

beauty of Gemini is just the integration

41:50

with Google's. platforms. It's huge. Yeah. Yeah.

41:52

That is their key advantage. And I

41:54

think the thing that could still mean

41:57

that they win because just the integration

41:59

into G drive. is you have all

42:01

the context, right? I got a

42:03

rough draft of everybody's performance review

42:05

just from all the decks in G

42:07

drive and Gemini. It's so much faster,

42:10

so much better. Oh man, yeah. This is

42:12

so amazing, man. I mean, obviously, if you

42:14

think AI at different places, but I'll

42:16

have so much more to share in

42:18

four weeks because I'm taking a lot

42:20

of energy away from it. Sweet, let's do it.

42:23

Yeah, I think this is a great episode because we

42:25

got some real insights into... we're a company like cracking

42:27

and you are how you think about AI. And then

42:29

just riffing on these things of like what are things

42:31

you're thinking about in the future and it'll be interesting

42:33

for you to come back on and tell us like,

42:35

do you do you find it good? Did you not

42:37

find it good? Has it been impact or not find it

42:39

good? Has it been impactful or not? Because A because AI,

42:41

because AI is such a broad thing, it's such a broad,

42:43

has it been impact or not? Because it has it has

42:45

been impact or not, has it, has it, has it, has

42:48

been, has been, has been, has been, has been, has been,

42:50

has been, has been, has been, has been, has been, has

42:52

been, has been, has been, has been, has been, has been,

42:54

has been, has been, has been, has been, has been, has

42:56

been, has been, has been, has been, has been, been, been,

42:58

been, been, been, been, been, That's the biggest part that comes

43:00

back to that operational question but I think this is

43:02

where leaders have to just take it on

43:04

upon themselves and just get the feed in

43:06

hand and I think the key is what

43:09

strategy you use to to filter out from

43:11

the clutter right exactly it is very easy

43:13

to get distracted exactly so that's why your

43:15

point at the end is important look we've

43:18

shared a lot of names a lot of

43:20

platforms but just focus yourself on two

43:22

or three. That's it. You know, and

43:24

those are the big ones and start

43:26

there. Then you can start to become

43:28

niche. Oh, here is this one thing.

43:30

But otherwise, it can get really noisy

43:33

and then you get overboard and

43:35

then you go back to your

43:37

old habits. Yeah, yeah. Yeah, I

43:39

actually think this is the most

43:41

interesting test of how quickly users

43:43

change behavior of all time. Yeah. Because

43:45

to your point, the natural inclination will

43:47

always be just, you know, doing the

43:50

thing I used to do. Well the other thing

43:52

here and you and I have talked about a lot is

43:54

that like I truly believe that like the what

43:56

model is best for what thing question is probably

43:58

not the right question. I think the right

44:00

thing to say is all of these things

44:03

are transformatively powerful and we are

44:05

underusing them. Yeah. And so if

44:07

I just took one, yeah, and

44:10

just obsessed about becoming deeper in

44:12

my adoption, fluency, expertise, and just

44:14

one, I'm probably far better than everybody

44:16

else. Yeah. Right? Because it's like there's

44:18

so much of all of these models

44:20

that we're underusing, because we are flipping

44:22

back and forth between all of them.

44:25

Yes. Yeah. Cool. All right. I think

44:27

this was a great episode. We really

44:29

appreciate you coming on. As always, you're

44:31

an incredible guest. And I think your

44:33

insights here are great for our audience.

44:35

So I appreciate it. And I think

44:37

we just were trying to go through

44:39

the process everybody has been going through.

44:42

Right. And like everybody is kind of living

44:44

in the same world over the last six

44:46

to 12 months. And I hopefully we shared

44:48

some perspective that's helpful. And then Meyer's going

44:50

to come back and give us around too,

44:52

which will be awesome. Which will be awesome.