HomeVideos

Trump's First 100 Days, Tariffs Impact Trade, AI Agents, Amazon Backs Down

Now Playing

Trump's First 100 Days, Tariffs Impact Trade, AI Agents, Amazon Backs Down

Transcript

2695 segments

0:00

I gotta wrap, guys. I got to catch a

0:01

flight to Miami. Let me do a closing

0:03

here. If you want to keep going, you're

0:04

welcome. Two. Three. Two. The plane.

0:05

Just wait. Just text the pilot and just

0:07

tell them you're all right. Listen. I'm

0:09

not burning all the allin credits, so to

0:12

speak, and all of our tokens. I'm

0:14

kidding. I'm kidding. I'm kidding. I'm

0:16

not flying private to everything and

0:19

then putting it on the all-in budget.

0:21

The rest of us are flying Southwest

0:23

for

0:24

your dictator Jim Pol. It's a strange

0:28

concept. Yeah. David S. I What does that

0:30

mean, Dave? When's the last time you

0:31

flew a commercial? Clinton, I haven't

0:34

missed a flight in about 15 years.

0:37

[Music]

0:39

Let your winners ride.

0:42

[Music]

0:46

We open sourced it to the fans and

0:48

they've just gone crazy with it.

0:54

All right, everybody. Welcome back to

0:55

the number one podcast in the world.

0:58

We're back. We're back and what an

1:00

amazing panel we have today with us.

1:03

Ryan Peterson, friend of the pod, is

1:05

back on the show. He's the CEO of

1:07

Flexport. How are you doing, Ryan? Did

1:09

you get any skiing in this year? I know

1:11

you like to ski in the deep powder like

1:13

I tried, man, but it was a busy year for

1:16

work and I got two little kids. I I did

1:18

a few days. Okay. So, you're Oh, yes. We

1:20

all forgot you gave control of your

1:23

company to somebody. It got a little uh

1:26

shaky, got a little contentious, and

1:28

then you took the reigns back. How's it

1:30

been being back in the pilot seat? Oh,

1:33

that was a year and a half ago. So, it's

1:35

a distant memory for in in flexport

1:37

time. That's like a decade. We've uh

1:39

yeah, really had an amazing run.

1:41

Although, these tariffs, I mean, I guess

1:42

that's why you guys invited me on. These

1:44

tariffs have kind of made a lot of

1:45

created a lot of new uncertainty in the

1:47

flex sports world. Okay, so we'll

1:49

definitely get into that. and of course

1:51

fan favorite back for his fourth

1:53

appearance on the pod. I I I was so

1:56

first of all I saw the comment I I saw

1:57

the comments last time I was on. I'm I'm

1:59

officially not a fan favorite but uh

2:01

glad to be back on and I will be I will

2:03

be representing free markets uh in uh in

2:06

this uh uh in this version. What do the

2:09

comments say about you? It was like uh

2:11

you know like uh loves Biden uh you know

2:16

to totally beta you know all the uh

2:19

lover beat soy boy you're filling in for

2:23

me though. Um I think so. Yes. I was

2:25

trying to represent uh I was trying to

2:27

represent libertarian values at the

2:28

time. But uh but I love this that the

2:30

leftists are embracing Milton Freeman. I

2:33

think it's all worth it if that's what

2:34

comes out of all of AOC AOC is going to

2:37

be a complete free market uh person

2:39

soon. Yeah, free market months are

2:41

coming soon. Embracing free market

2:42

values and the stock market, right? Yes.

2:44

Exactly. Because any decline in the

2:46

stock market is Trump's fault. So now

2:48

they're they're embracing the stock

2:49

market. Well, unfortunately

2:50

unfortunately the one day that he said

2:52

it's Biden's market, it was a it was a

2:53

green day. So that uh that didn't help

2:55

the case. I mean gosh uh well listen,

2:57

it's another massive green day already.

3:02

All that matters to me is that Uber is

3:04

the anti-tariff stock. It just does

3:07

great. It's not impacted by tariffs. So

3:08

here we go. with Chimoth. It started

3:10

already. We have TDS on both sides.

3:12

We've got Trump derangement syndrome

3:14

from Aaron. Wait, no, no, no, no. I want

3:16

to be defender syndrome from Sachs.

3:18

We've got both. Defender and

3:20

derangement. Here we go. Syndrome. Trump

3:23

bias syndrome on your part. Who? Me? Me?

3:25

I call balls and strikes. What are you

3:27

talking about? Let's get started. It's

3:30

starting already, folks. It's going to

3:31

be a great episode. Lots of excitement

3:34

with us again. Jason has the uh the rain

3:37

self-sabotage. Find every way to not get

3:39

rich syndrome. I do. What are you

3:42

talking about? You guys said you're buy

3:43

me out of this thing and I can get the

3:45

hell out of here. You know how much my

3:46

shares are in all in worth? For the love

3:49

of God, write a check.

3:52

Get me the hell out of here. I just may.

3:55

Oh god. I mean, I'm going to be a

3:57

terror. If Uber breaks 88, that's my

4:00

number. 88 is the number. You're all

4:03

when that happens. Uh, and we're getting

4:04

close. All right, let's get started

4:07

here. We have so many topics to get

4:10

through with us again. David Saxs. Hey

4:13

David, you're doing uh more episodes

4:15

now. The audience wants to know. I don't

4:18

know if we're allowed to make any

4:19

initial announcements, but people are

4:20

asking me on the streets, in the

4:22

airports, in the comment threads. Is

4:24

Sachs back?

4:26

Well, the ratings are back ever since I

4:28

came back to the show. That's for sure.

4:31

The ratings are back, show, but is

4:34

measurable back? Is Sachs back?

4:37

Well, I'm back as much as I can. Mhm.

4:40

And you are a partial employee of the

4:42

government. You can do 130 days a year

4:44

or something. Is that still the status?

4:46

Yeah, it's roughly half after work days.

4:48

Got it. And so what do you do? You have

4:49

a you have a punch clock there. When you

4:51

get to the White House, you punch in,

4:52

you punch out like Fred Flintstone or

4:54

what? Are you keeping track of these

4:55

days? How do you do it? I know why you

4:57

don't know this because you have yet to

4:58

be invited to the White House. But

4:59

that's interesting. I got not how it

5:02

works. Normal people just people just

5:04

badge in and badge out like that. Badge

5:06

it and badge. It's a natural place,

5:07

Jason. I I mean, literally, it's

5:09

interesting. There's a new private club.

5:11

It's incredible that you have thoroughly

5:14

prepared for this week, just like

5:15

always. I am always prepared.

5:17

Interestingly, I don't know if you

5:18

gentlemen know this, Ryan and Aaron.

5:20

There's a new private club in DC that uh

5:23

Don Jr. is doing and Sax is a member.

5:25

Chimat's a member. And I just checked my

5:27

Gmail. I checked all three of my Gmail

5:29

accounts, everything. No invite. You

5:31

must have gotten lost again.

5:34

Did you send a paper one? Was it like

5:35

you sent a goal card or something? Sax,

5:37

how do I get invited to this private

5:39

club? What is this private club?

5:40

Everybody wants to know. Well, we'll be

5:42

happy to have you as a guest. Okay. Do I

5:45

have to wear a MAGA hat and have the

5:46

courtesy MAGA hats at the door? If you

5:48

want to be a member, obviously there are

5:50

dues and a membership fee, and Okay. I

5:53

just didn't want to waste your time with

5:55

an offer that I knew you wouldn't uh be

5:58

willing to accept. It's only $500,000 is

6:01

what I read. Is that true?

6:03

That's true for for founding members who

6:05

have additional benefits, but there's

6:06

also a lower level that's the more

6:08

reasonable membership level. So, I think

6:11

people are getting a little bit carried

6:12

away with that number. Got it. Okay.

6:13

That's why I wanted to clarify. Yeah.

6:15

Yeah. There's like 10 founding members

6:17

who have that level and then there's a

6:19

lower level for more average member.

6:22

Chimatha, are you one of those 10? Yes.

6:25

Do you pay more if you have TDS or how

6:27

does that work? TDS premium. What do you

6:29

talking about? Jal specifically or what

6:30

are we talking about? the TDS search

6:33

charge. Asking for a friend. It's a TDS

6:35

search charge. You put the tariff search

6:38

we just we want a place to hang out in

6:41

DC. All of us have been to clubs like

6:43

the Battery or I don't know if you go to

6:45

LA like the I think places. There's

6:48

Malibu Beach House. There's Bird Street

6:50

Clubs. There are places in Palm Beach

6:52

that are really cool. In any event, we

6:54

wanted a place to hang out. And the the

6:56

clubs that exist in Washington today

7:00

have been around for decades. They're

7:01

kind of old and stuffy. To the extent

7:03

there are Republican clubs, they tend to

7:05

be like more Bush era Republicans as

7:08

opposed to Trump era Republicans. So, we

7:11

wanted to create something new, hipper,

7:13

and Trump aligned. Since I'm in the

7:16

government, I can't be an owner, but I

7:18

told him I'd be happy to be member

7:20

number one. And so, I, you know, said,

7:22

"Great, let's let's do it." And so,

7:24

we're creating a place for us to hang

7:25

out. That's basically it. We want a

7:27

place to go where you don't have to

7:28

worry that the next person over at the

7:30

bar is a fake news reporter or even a

7:33

lobbyist or something like that who we

7:35

don't know and we don't trust. Got it.

7:38

So, it's like any private club. You want

7:40

to go somewhere that's highly curated.

7:45

This private club uh movement is

7:46

happening all over the country, not just

7:48

Washington. But we're creating something

7:49

that didn't exist before in DC, which

7:52

again is younger, hip, Trump aligned,

7:54

Republican. We're uh I I actually

7:57

started a Kamala club in um in the Bay

7:59

Area. So um so we're Yep. I don't think

8:03

anyone would pay to join that though is

8:05

the problem, right? I mean, it's an open

8:08

bar, that's for sure.

8:10

Where where do where do you guys meet

8:12

up? In like Redwood City. We actually

8:13

meet up at the uh at the at the trade

8:15

ports. All right. was we're 100 days

8:18

into Trump 2.0. It's just a random 100

8:22

day thing, but everybody's talking about

8:24

everybody's hand ringing. What has it

8:26

been like for this first 100 days? How

8:28

does it compare to Biden? How does it

8:29

compare to Trump 1.0? 143 executive

8:32

orders, the most ever in the first 100

8:35

days. And they're moving obviously at a

8:37

at a different pace to uh be generous.

8:41

Major indices are down 7 to 10%.

8:43

Obviously, this trade war and tariffs,

8:47

the yield on the the yield on the

8:49

10-year, it's down about 40 basis

8:51

points. There's a lot going on. Let's go

8:53

around the horn. Ryan Aaron, you're our

8:55

guest. What's your take on the first 100

8:57

days? Is it what you expected, good,

9:00

bad, and otherwise, wins and fails,

9:02

everything. I'll go first. I think it's

9:04

a whirlwind. I mean, if you look at the

9:07

uh the John Boyd, the fighter pilot, has

9:09

this concept of the UDA loop, which is

9:11

observe, orient, decide, and act. And

9:14

the concept is that if you're in dog

9:15

fighting, if you're able to maneuver

9:18

through those UDA loops at a faster pace

9:20

than your than your competition, they

9:22

get disoriented and don't know what to

9:23

do. And I I think that that's got to be

9:26

how Democrats in Washington and maybe

9:28

mainstream Republicans in Washington.

9:30

Certainly journalists are all feeling

9:32

this like there's that the Trump Trump

9:34

administration takes action and before

9:37

anybody can respond to that they have

9:39

already done like four more things and

9:40

you're like wait I forgot to actually

9:42

follow up on the other thing that they

9:43

did that I didn't like. Uh, and so it's

9:45

yeah, it's pretty disorienting if you're

9:47

if you're trying to they they can't find

9:49

a line to fall back to and go, "Hey,

9:50

we're going to push back against this

9:51

policy because they're already moving on

9:53

to the next one, the next one." Um, so

9:55

that that's like my high level

9:56

interpretation. Obviously, I come at it

9:58

from a trade angle. I think everybody

10:00

knew that Trump was going to be he he

10:02

told us during the campaign that the the

10:04

most beautiful word in the English

10:05

language is

10:06

tariff. Don't tell them it's an Arabic

10:08

word, but the most beautiful word in the

10:11

English language. And so we knew that

10:12

was coming. I think that the the the

10:16

suddenness of it all caught people by

10:18

surprise. I mean, they told us April

10:20

1st, April 2nd would be Liberation Day.

10:23

They didn't tell us that it would go

10:25

live the next week, you know, and effect

10:27

you've already ordered these goods. So,

10:28

that's one aspect that people are kind

10:31

of disoriented about. And we're gonna

10:33

unpack. Yeah, we're going to unpack

10:35

that. Aaron, your thoughts on the first

10:36

100 days? Obviously you are a Democrat

10:40

and uh you were pretty vocally not in

10:44

support of Trump. So what's your take on

10:46

the first 100 days? Any any bright spots

10:48

for you things you you know support?

10:50

Actually Sax's world I'd say has has

10:53

been a bright spot. So especially I mean

10:55

I think we have a very clear message on

10:57

AI and uh and that that that is that's

11:00

been I think a huge net positive is um

11:03

you know if you look at the the past you

11:05

know few months uh out of all the the AI

11:08

push from the administration it's

11:09

unmistakably you know pro open source

11:11

you know pro you know bring as much AI

11:14

innovation you know to the US obviously

11:16

that the tariffs you know add a little

11:17

bit of a headwind to that. I have some

11:19

very strong asks, you know, around high

11:21

skilled immigration because I think

11:22

that, you know, AI talent is going to be

11:25

super critical to to actually win the AI

11:27

war. So, so I'd say that that

11:28

directionally has has had some positive

11:30

momentum. You know, from my perspective,

11:32

this is kind of playing out almost

11:35

exactly how I thought it would 6 months

11:36

ago. And then 3 months ago, I I think

11:40

there was some signs that maybe maybe,

11:42

you know, it wouldn't play out this way.

11:44

um just based on some of the some of the

11:46

you know kind of early groups that were

11:48

coming to the White House the the the

11:49

the sort of deep business you know kind

11:52

of centricity of the White House you

11:54

know I think it was day one or two that

11:56

Stargate was announced you know at the

11:58

White House we're going to go build

11:59

massive infrastructure the case I'd like

12:01

to make you know once we talk about

12:02

tariffs is is I think there's an

12:03

alternative universe where you just lean

12:05

into acceleration as opposed to adding

12:07

headwinds but but so that would be that

12:10

would be the case of what what maybe

12:11

could have been you know very different

12:12

is we just keep double down on doubling

12:14

down on what's working while fixing the

12:16

parts that aren't working. But uh but

12:19

that would be, you know, my my uh my

12:20

judgment so far. Chimath, I mean, you've

12:23

been talking about it here every week.

12:26

You and I have uh been talking about it

12:28

pretty consistently, so I don't think

12:29

there'll be many surprises here, but

12:31

take a second and maybe assess what you

12:33

think if you had to pick a singular

12:35

thing that's gone really well and a

12:37

singular thing you think could be

12:38

improved. What What do you got? Let me

12:40

give you my overall grade.

12:44

And then I'll tell you how I get to

12:45

that. I think the first 100 days have

12:49

been a B+.

12:54

And here's how I get to that

12:56

score. There have been two things where

12:58

I think Trump

13:01

has frankly hit a home run. The first is

13:06

all of the direct

13:08

investment and specifically the foreign

13:10

direct investment into the United

13:13

States. I think it's approaching if not

13:15

it has already exceeded a trillion

13:17

dollars from corporations and

13:20

organizations and individuals from

13:22

around the world who have committed to

13:25

bringing money into the United States.

13:27

And I think strategically that's a

13:29

legacy that will live past him. So, I

13:32

think that's been an

13:33

A+. The second is we had a very unsafe

13:38

border

13:39

situation and he ran on shutting it

13:43

down. I'm not talking about the

13:45

execution of the deportations. I'm just

13:47

saying getting the illegal crossings to

13:50

zero and he's done that. So, that's been

13:53

an A+.

13:55

I think what's going to be more

13:57

controversial are these next three

13:58

things though. But in my interpretation,

14:01

I think the tariffs have been an

14:04

A and I think that the market reaction,

14:07

the stock market is only down 4%. And

14:10

the interest rate markets are, you know,

14:12

4 and a/4%. I think those have been an

14:14

A. Now the reason I think tariffs have

14:18

been an A is because it is uncovered in

14:21

my opinion how beholden we are to a

14:25

brittle supply chain and specifically to

14:28

China who is a friend but who's also an

14:30

enemy and I think that that's going to

14:32

really severely complicate

14:35

our flexibility and optionality in the

14:38

future as they do what is in their best

14:40

interests. Okay. So where have they then

14:43

not done so well? I think the documents

14:47

have been frankly a D. We were supposed

14:50

to get the Epstein files. We haven't

14:52

yet. We were supposed to get the Martin

14:55

Luther King files. We haven't. We did

14:57

get the redacted JFK files. I don't

14:59

think there's been very good

15:00

communication about why it's taking so

15:02

long. So I think it's a very small

15:04

narrow thing, but I think it had a lot

15:06

of attention on the way in. I think

15:09

the communications of the

15:12

tariffs and the back and forth have been

15:14

a C. I think the markets were not

15:19

led in enough of a way where they could

15:22

absorb the

15:25

volatility. But if you take it all in

15:27

its totality, I would give it a B+. I

15:30

think it's been a very productive 100

15:32

days. And when you look back, I think

15:34

in, you know, 3 years, four years, 5

15:36

years. Okay, we've made some important

15:39

progress. Saxs, obviously you're part of

15:42

the administration, so I'm not sure

15:44

exactly how to ask you this, but you s

15:45

you heard some nice compliments about AI

15:47

from Aaron. I I happen to agree with

15:49

those. I actually agree with uh a good

15:51

portion of the crypto stuff, too. I

15:52

think actually getting those uh

15:54

tightened up, which are your two zones

15:56

of excellence and your area that you're

15:59

focused on. I think you've done a great

16:00

job there. So, just bestie to bestie,

16:02

great job there. What's your Thank you.

16:04

What's your take overall? You know, it's

16:06

kind of hard, I guess, to ask somebody

16:08

in the administration to criticize the

16:09

administration, but hearing everybody

16:10

else's take, what's your response,

16:12

maybe? Well, I would I would highlight

16:14

three main areas that I think are big

16:17

accomplishments for the Trump

16:18

administration in the first 100 days.

16:19

So, so number one has to be the border.

16:21

Like Jamas said, I think you have to

16:23

give the administration an A+ on this.

16:25

They've completely stopped the border

16:26

crisis. I think we all knew that Trump

16:29

would take action on this because it's

16:30

one of the main issues he campaigned on.

16:32

I think if you had asked any of us, you

16:34

know, 4 months ago, would this problem

16:36

be completely solved? Meaning, border

16:39

apprehensions completely stopped, border

16:41

completely sealed within the first 100

16:43

days, I don't think we would have

16:45

believed necessarily that it would get

16:46

done so quickly, but it has. Uh, recall

16:50

that for 4 years during the Biden years,

16:52

we were told for the first 3 years that

16:55

the problem didn't even exist. Whenever

16:57

the videos were published of caravans

17:00

coming or throngs of people running

17:02

across the border, we were told that

17:04

these were cherrypicked videos on Fox

17:06

News. It wasn't real. Finally, in the

17:09

last year of the Biden administration,

17:11

they said, "Okay, we're finally going to

17:12

do something about it." They took some

17:13

limited actions and they said that doing

17:15

more than that would require new

17:18

legislation. Well, all of that was just

17:20

gaslighting. It turns out Trump came in,

17:22

he restored remain in Mexico and other

17:24

policies, completely stopped it. He had

17:26

this line at the state of the union

17:27

which I think is exactly right which is

17:28

we didn't need a new law we just needed

17:30

a new president. So I think that's area

17:32

number one. Area number two I would say

17:35

would be the vibe shift in the culture

17:38

around wokeism and DEI. You know how

17:41

quickly we forget about this but wokeism

17:44

has completely collapsed. Uh I don't

17:46

know that anyone is endorsing in a

17:49

fullthroated way. Moreover, beyond just

17:51

sort of the cultural aspect of it, I

17:54

think we've had significant policy

17:55

changes on DEI. Trump has basically

17:58

ended DEI at the government level. He

18:01

also signed an executive order ending

18:04

the use of disparit impact for

18:06

affirmative action. This is the policy

18:08

that said that even if you have a policy

18:11

that's applied in a completely neutral

18:13

and objective way, if it results in a

18:16

disparate impact where different groups

18:17

are represented in a different way in

18:19

the outcomes, then somehow that must be

18:22

racist. And that led to essentially

18:25

engineering the results of various

18:28

populations to basically fit quotas. And

18:32

I think all of that now has fallen by

18:34

the wayside. And I think that

18:35

meritocracy and colorblindness are back.

18:37

The only hold out really has been these

18:40

universities where Trump is now taking

18:42

action against Harvard and I think that

18:45

ultimately we will win that battle. You

18:47

see that even in relatively liberal

18:50

companies the DI departments have been

18:52

cancelled and they're moving back

18:53

towards more of a meritocracy. So I

18:56

would say that that's like big shift

18:57

number two. And I think if any of us had

18:59

tried to predict that 100 days ago, we

19:02

would have thought yes, Trump will do

19:03

something about it. But I don't think we

19:04

would have predicted the total collapse

19:07

of wokeism and DEI so quickly. And then

19:10

I'd say the third area which is still in

19:13

flight is the rep privatization of the

19:15

economy. That's a term that Scott Besson

19:17

used. I think that the Trump

19:19

administration needs to reprivatize the

19:21

economy. And I like that framing of it.

19:24

And there's a bunch of different pieces

19:26

under that. I'd say number one is

19:28

Doge again ending this hogw spending. I

19:33

do think that Trump has come into office

19:36

inheriting a very weak Biden economy

19:38

that was being propped up by massive

19:40

amounts of government spending that was

19:42

not only stimulating the public sector,

19:44

but it was also gooseing the employment

19:46

numbers as well. And we knew that that

19:49

spending was unsustainable. We have to

19:51

do something about it. So, I think for

19:53

the first time in decades, we've

19:54

actually started to make real cuts in

19:57

government, real cuts in the federal

19:59

workforce. And look, we'd like to do

20:01

more, but that is a huge shift in the

20:03

conversation. There's other pieces of it

20:06

as well. I mean, President Trump has

20:08

signed a significant number of executive

20:10

orders on deregulation. There's also

20:13

been unleashing energy. He ended Biden's

20:15

EV mandate and a lot of these like green

20:18

new scam projects, offshore wind, and

20:21

he's been encouraging oil and gas

20:23

exploration. So, I think there's that.

20:25

And then I appreciate what Aaron said

20:27

about tech innovation. We did repeal

20:30

Biden's exec order on AI, which was, you

20:32

know, 100 pages of unnecessary

20:34

regulation on AI. We've ended the war on

20:36

crypto, and I think we're trying to stop

20:39

the regulatory capture that benefits

20:40

large incumbents. So, you have all these

20:43

things, and there's been other things

20:44

that that have been done on the economy

20:45

as well, but I I do think that this sets

20:47

us up for a Trump boom in the future.

20:51

It's just that a lot of these changes

20:53

take time to to play out. Okay, great.

20:56

Well done. And well, I think I think I

20:59

think we knew Sax would be very pro.

21:01

Chamas uh Chamas seems really pro other

21:03

than he wants like the alien conspiracy

21:05

files released which we'll get soon.

21:08

What is the what is the view from Jay

21:10

Cal when the where you're the the the

21:13

leftleaning guy in the in the room? Uh

21:16

you know I'm kind of independent but

21:17

yeah social liberal. You know I I look

21:19

at what all Americans believe and and

21:22

try to build some consensus here. It's

21:24

one of the things I've been trying to do

21:25

on the pod is look for where we

21:27

agree. Americans universally want the

21:30

border secured. They don't want illegal

21:33

immigration and they don't want

21:34

fentinel. So this is the biggest win I

21:36

think for Trump which I think everybody

21:37

on the panel pointed out and Sachs you

21:40

were dead right like when we were seeing

21:41

those videos some of them were 5 years

21:42

old some of them were recent. Biden

21:45

really covered up what was going on in

21:46

the border and it took years to figure

21:48

out what was exactly going on there. So

21:50

that's the biggest win possible. I I

21:52

give overall just to be brief a B

21:55

for this first 100 days and I give you

21:58

know Biden like a C minus. The second

22:01

thing that everybody agrees on is they

22:02

want to downsize the government. They

22:03

don't want waste and fraud. So I think

22:04

Doge is the other huge win. The things I

22:08

think that could be improved really just

22:10

three simple things. The economic

22:12

uncertainty is really terrible for

22:14

running a business. I'm seeing a lot of

22:15

folks in my circle on my podcasts this

22:18

week in Startups and here telling me,

22:20

"Oh, I don't know how to plan for the

22:22

future." And we're going to get into

22:23

that with this tariff stuff and the

22:24

trade war. And so, I think economic

22:27

uncertainty, we have to sort of slow

22:30

down and maybe make it easier for people

22:32

to understand what the administration is

22:34

trying to do. I think rule of law really

22:36

matters to people. People didn't like

22:38

Biden's pardons. They didn't like

22:40

covering up his mental acuity. And I

22:42

don't think people like the deportations

22:44

without due process. We talked about

22:45

that on a previous episode.

22:47

Overwhelmingly, people want Trump and

22:50

the administration to obey what the

22:52

Supreme Court says. They really want

22:54

rule of law, the third term talk. Like

22:57

eight out of 10 Americans don't like

22:58

that kind of talk. Um, and then

23:01

conflicts of interest. Obviously, people

23:02

hated the Hunter Biden stuff. They hate

23:04

the memecoin stuff. And so that's where

23:07

it could improve. crisper

23:08

communications, more thoughtful

23:10

execution, maybe less trolling. I don't

23:12

like the White House Twitter account

23:14

trolling. And then focus on what got

23:16

Trump here. You know, you all said the

23:17

same thing. What got Trump in here was

23:19

the economy. And one thing that wasn't

23:22

mentioned by everybody is the peace

23:24

dividend. And Trump is making massive

23:26

progress in Ukraine, apparently. I don't

23:29

know if it's on the docket today or not,

23:30

but stopping the wars and making the

23:33

economy boom, those are the two most

23:34

important things that he could do. Build

23:36

on that. I I totally missed that. You're

23:38

absolutely right. That's another one

23:39

where I would give Trump an A+. Nat and

23:41

I had dinner with POTUS two weeks ago.

23:45

And wait, you had dinner with Trump?

23:47

This is breaking news. Well, okay,

23:49

whatever. Yes. Well, I think it's I

23:50

think it's remarkable how much of a

23:52

Putin apologist Jake House's become. I

23:54

mean, you want to end the war in Ukraine

23:55

now? Well, you're going to you're going

23:56

to give it You're going to give it to

23:58

Putin? You're not going to stop Putin.

24:01

I'm totally in favor of what Trump's

24:03

doing in in negotiating a deal to get

24:05

more money. Oh, you want to talk to

24:06

Putin now? I've always wanted to talk to

24:08

Putin. I just don't trust him. But you

24:10

you can trust him. Let me tell you what

24:12

Trump said. So, there we go. There was a

24:14

handful of us at dinner and then he got

24:18

up to say a few words at the end and he

24:21

reminded me why I was so inclined to

24:27

vote for him, which is he talked about

24:29

his uncle

24:31

and he talked about how his uncle taught

24:33

him about the severity of nuclear war

24:36

and how people don't understand how

24:39

intense and how destructive it is and

24:42

the power of these weapons and he left

24:47

that speech at the end saying and this

24:49

is why I'm so fundamentally against this

24:51

thing and it reminded me to your point

24:54

Jason it is so easy to forget that

24:57

there's only one existential risk

25:00

save like aliens coming from the

25:02

heavens, right? There's only one

25:04

existential risk where all these issues

25:06

become fringe issues. You know, you

25:07

mentioned rule of law, border security,

25:09

foreign direct investment, tariffs, it

25:12

all goes out the window in a nuclear

25:14

war. And I was like, I am so glad this

25:17

guy's in charge because this one issue,

25:20

he never waivers.

25:23

Yeah. And I think there's all kinds of

25:25

complicated moments that could make this

25:29

an issue. And this was where my biggest

25:31

issue with Biden was was I did not know

25:33

who was in control. And I think that

25:35

Trump in the first 100 days, to your

25:37

point, I think has completely reinforced

25:39

that there are no conditions under which

25:42

he'll go to war. He has time and time

25:43

again showed find the off-ramp. And I

25:46

think that that's really healthy for

25:47

Americans to see. Yeah. And let me build

25:49

on that point with respect to to Ukraine

25:51

is we were on a glide path before the

25:54

Trump presidency that Biden had put us

25:57

on a certain path. Kla Harris gave every

26:00

indication she would have continued it.

26:02

What was that path? It was a path of

26:04

continued escalation and doubling down

26:06

in Ukraine. Recall that it was Biden

26:08

himself at the beginning of the war who

26:10

said that if we give Ukraine Abrams

26:13

tanks and F-16s or attackums or high

26:18

Mars or if we allow them to hit targets

26:20

inside of Russia, it would lead to World

26:23

War II. He actually used the word

26:25

Armageddon. So at the beginning of that

26:27

administration, they were very concerned

26:30

about how an escalatory path could lead

26:32

us into direct conflict with Russia and

26:35

World War II. And yet, despite that, at

26:38

every fork in the road where they had a

26:39

choice, they ended up doubling down.

26:41

They gave the Abrams tanks. They gave

26:42

the F-16s. They gave the High Mars. They

26:44

gave the attacks. And finally, when

26:47

Biden was a lame duck in his last couple

26:48

months in office, they did the most

26:50

reckless and irresponsible thing, which

26:52

is allow American weapons to be used to

26:55

strike targets on Russian soil. Not just

26:57

fighting in Ukraine, but on Russian

26:59

soil. Moreover, we now know from a New

27:01

York Times article that just came out in

27:03

the last few weeks that it was American

27:05

generals and American intelligence who

27:06

are planning this war. So, when you're

27:08

talking about striking Russian targets

27:10

on Russian soil, it's not just the

27:12

Ukrainians using our weapons. They're

27:14

using our targeting, they're using our

27:16

guidance, they're using our satellites.

27:18

I mean, we are deeply integrated in the

27:19

kill chain. This is the United States

27:21

being a co-elligerent in the war,

27:23

hitting Russian soil. That is incredibly

27:25

reckless and dangerous. I have no doubt

27:27

that if the Democrats were still in

27:30

office, we would be in an escalatory

27:32

spiral right now with the destination

27:34

being World War II. And I do think that

27:36

Trump has pulled us back from the brink

27:38

there. There's obviously still more work

27:40

to do. But I really appreciate the

27:42

efforts that Steve Wickoff has

27:44

undertaken where for the first time in 3

27:46

years, we've at least had direct

27:48

diplomacy with the Russians. We weren't

27:50

even talking before. We weren't even

27:52

talking before. I mean, talking is a

27:53

great thing and and and apparently we're

27:55

going to keep supplying with them with

27:57

weapons as long as they pay for them.

27:59

So, it's going to be very interesting to

28:00

see how this all hashes out over the

28:02

next 100 days or so. Let's keep moving.

28:05

I don't think we know that yet. Let's

28:07

let's wait and see on that. Okay. Yeah.

28:08

I mean, I think that's Yeah. Uh what was

28:11

reported, but you're right. We should

28:12

wait and see. Okay. Downstream tariff

28:15

impacts. We got to talk about this, and

28:16

this is why we have you here, Ryan,

28:18

since you're in the thick of it. you

28:20

tweeted a thread last week about the lag

28:23

time uh of shipments from China and when

28:26

you were on I guess during COVID you

28:28

really educated us to how the supply

28:30

chain how the supply chain

28:32

works and according to the thread that

28:36

you shared somewhere around early June

28:38

we're going to expect warehouses

28:40

trucking the entire supply chain maybe

28:43

to start to seize up or layoffs I don't

28:47

know how you would frame it Ryan but are

28:49

we asked the point of no return with

28:52

regard to the supply chain. Is there an

28:55

off-ramp for this tariff conflict war

28:58

negotiation with China in your mind?

29:00

What are you seeing on the streets and

29:03

in the purchase orders and the invoices

29:06

at Flexport? Definitely not past the

29:08

point of no return. I think we're still

29:10

right in the middle of the don't judge

29:12

the cook while he's cooking is one, you

29:13

know, like let's see what the we'll see

29:14

what it tastes like at the end is I

29:16

think a starting point here and we're

29:17

still they're still in active

29:18

negotiations. So I don't think today's

29:20

it's not static. Now the world does want

29:23

a lot more certainty and that's a big

29:25

cause of what's happened here and what

29:27

has happened is a 60% decline in

29:29

bookings of ocean freight from China to

29:31

the US. I mean so that's really really

29:33

pretty dramatic like probably exceeding

29:36

what was expected. I don't think, you

29:39

know, when they issued when they rolled

29:40

out the initial reciprocal tariff plans

29:42

on on April 2nd, it was meant to be a

29:44

54% tariff on China. Then, you know,

29:47

there's multiple cycles of escalation.

29:49

We ended up at what's now 154% tariff.

29:53

So, this is this is a lot higher than

29:56

anybody planned for. And so therefore, I

29:58

don't think anyone's planning for a 60%

30:00

decline in ocean freight. Ryan, let me

30:03

ask you a question about that. Are

30:05

people actually paying that 154%?

30:07

There's been this discussion online and

30:09

it's it's sort of unclear from the

30:11

administration and from

30:14

retailers stuff that's landing that

30:16

people ordered before April 2nd. Are

30:20

they actually paying the 154% on top of

30:22

what's landing? It's it's live now. Um

30:24

it is it was based on departure date. So

30:26

goods that departed China after midnight

30:30

Eastern time on April 9th are subject to

30:33

the tariffs upon arrival. And so now

30:35

enough time has passed that pretty much

30:36

all the ships that are arriving now left

30:38

China after April 9th when that started.

30:40

Um so yes. So what happens? People are

30:42

paying it or are people saying I won't

30:43

take delivery because it's Jason you

30:45

have you have to pay it Ryan correct me

30:47

if I'm wrong but you have to pay it at

30:49

the dock in order to get the goods

30:50

released. More or less more or less

30:52

that's true. They they allow you have a

30:55

bond in place so you can pull the goods

30:56

out before you pay but it the money's

30:58

owed at that time and then you get you

31:00

get like a two week time frame to

31:01

actually make the payment. But there are

31:03

strategies here. a lot of people are

31:04

doing that you can use what's called a

31:06

bonded warehouse and move cargo into

31:08

this warehouse uh and then you only owe

31:09

the duties when the cargo leaves that's

31:12

what I was asking like is there a hack

31:14

here to it's not that that lets you

31:16

defer things and it's very very common

31:17

right now people are searching

31:19

everywhere for bonded warehouse capacity

31:20

because in a bonded warehouse not only

31:22

you defer payment to when the cargo

31:24

leaves the warehouse but you only owe

31:26

the duty amount based on at that date so

31:30

if the duties come back down which a lot

31:32

of people are betting they will on the

31:33

China specific speific duties, you'll

31:35

actually lower your tariff burden. And

31:37

then there's another hack for this,

31:38

which is effect use a Mexican or

31:40

Canadian bonded warehouse. So you move

31:42

the goods into Mexico and then you

31:44

actually only technically import them

31:46

into the US at a future date when

31:48

tariffs are lower. So I understand a lot

31:49

of a lot of companies are doing that

31:51

right now, too. Um we're helping some

31:53

people with that type of strategy,

31:55

but yeah. Is that Sorry, Ryan. Do you

31:57

think that the government will they view

31:59

that okay that kind of hack and or like

32:03

you know like if you look at the GDP

32:04

numbers one of the craziest things was

32:06

the inventory pull forward that people

32:08

did to your point like trying to get as

32:10

much stuff into the United States before

32:13

April 9th as an example. Yeah. I mean

32:16

it's not a hack. It's a bonded

32:17

warehouses are been around for decades

32:19

and they're they're very commonly used.

32:21

I don't know that it'll be that material

32:22

in the scheme of things that it would,

32:25

you know, cause a change in the law

32:27

around bonded warehouse. So, you don't

32:28

think, for example, the Department of

32:31

Commerce will have an issue with the

32:33

strategy of sending inventory into

32:36

Mexico that essentially you're

32:37

essentially like, isn't it, it's a work?

32:41

Like, instead of paying the China

32:42

tariff, now you pay a Mexico tariff,

32:44

which should be less. Is that the idea?

32:46

Well, you can move it into a bonded

32:47

warehouse in Mexico even and not pay

32:49

Mexican tariffs either. and you just

32:51

wait until it imports. But I mean,

32:52

what's the Department of Commerce or the

32:55

customs to do? It's sort of you just

32:57

delayed importing the goods. You've

32:58

imported them in the future and you

33:00

know, it doesn't I I wouldn't even call

33:02

it a hack. It's just sort of like people

33:04

people are going to get creative here.

33:05

You know what I mean? Like that's the

33:06

job. Actually, the government should set

33:08

the rules and the rest of us got to

33:10

figure out, all right, how are we going

33:11

to compete and make money in this

33:12

environment that they've created? Ryan,

33:14

in that tweet you redid, which was a

33:16

pretty dramatic tweet painting a very

33:18

like I don't know, you know, like a

33:21

pretty dire situation. Where are we at

33:24

in terms of how uh dire this will get or

33:28

resolvable? paint us the the the best

33:31

case scenario and what you expect could

33:33

happen in that case or if this gets

33:35

extended are we going to see as you know

33:37

people are hand ringing empty store

33:40

shelves Christmas gets ruined and all

33:43

these layoffs start happening in the

33:45

supply chain take us through the two

33:46

scenarios that people are debating yeah

33:48

I mean the the the bleak scenario which

33:50

is I don't really think it's going to

33:52

happen I think that the administration

33:53

doesn't want this to be their legacy

33:54

that they like created a policy that

33:56

just like kind of tanked small business

33:58

and supply chain. So, I don't I don't

33:59

actually think this is going to happen,

34:00

but the bleak scenario is tariffs stay

34:03

at this level for 145% on China. The 10%

34:06

goes way back up to what it was

34:08

originally announced in reciprocal

34:09

tariffs. So, there's no like safe haven

34:10

for tariffs and trade just falls off a

34:13

cliff and a lot of companies go bankrupt

34:14

in our in our especially small companies

34:16

are the ones that are importing from

34:17

China. Reality is like tariffs have been

34:20

high on China for a long time. Labor

34:22

costs in China are not are not there for

34:24

cheap labor. You're there for quality

34:25

manufacturing at this point. like

34:27

there's much cheaper labor in Southeast

34:29

Asia, other parts of the world than

34:30

there is in China. So you're in China

34:32

because of the manufacturing

34:33

capabilities, the ecosystem, not just

34:35

for cheap labor. Uh and if you could

34:37

have moved, you would have already with

34:38

the 25% tariffs from the Trump's first

34:41

terms were pretty were high enough

34:43

incentive. And so that's the bleak

34:46

scenario is that small business starts

34:48

getting wiped out. The ones that are

34:49

buying from China and it's a lot of

34:51

brands like it's not just Amazon seller

34:53

selling stuff that you don't need. It's

34:56

like all the brands that you know are

34:58

like you know fashion brands, apparel

35:00

brands. I had cuts clothing on this

35:03

weekend startups last week and he said

35:05

there's going to be like if this doesn't

35:07

get revol resolved in like let's say two

35:09

to four weeks in his group chats people

35:11

are going to start layoffs and they they

35:13

can't

35:15

physically restart the supply chain in

35:18

Vietnam or wherever to make t-shirts. So

35:20

Aaron what's your thought on this as

35:22

well just bringing you in. Sure. Well,

35:24

well, first of all, I mean, Ryan has

35:26

supplied me with a high degree of doom

35:27

scrolling and uh it's just like a horror

35:30

show reading his tweets. First of all, I

35:32

I like I would feel better if the

35:34

messages out of the administration were

35:36

either more kind of consistent or that

35:39

there was a logical connection between

35:42

do we either want to raise the kind of

35:46

you know tariff revenue stream or do we

35:48

want free trade like like those things

35:50

are are working against each other

35:52

because like depending on who you talk

35:53

to they they say this is a a mechanism

35:55

to bring down income tax which obviously

35:57

then by definition means that they

35:59

expect the tariffs to sort of persist.

36:01

um which is totally different from let's

36:03

go negotiate deals that just allow for

36:05

the you know free trade to actually

36:07

increase and so are we worried about the

36:09

reciprocity or we worried about kind of

36:11

revenue stream so that's a whole whole

36:12

issue you also have this issue which is

36:15

the messaging from the government and

36:16

this is the meta point I'll make in a

36:18

second is about is about how we could

36:19

have actually accelerated into the

36:21

transformation of the economy but you

36:23

know you have folks like Lutnik etc you

36:25

know going on on TV talking about the

36:27

the end state of our economy which are

36:28

are actually probably you fine messages,

36:31

but but we haven't seen what that vision

36:34

looks like, you know? So, everybody is

36:35

kind of confused like does this mean

36:37

that we literally go into manufacturing

36:38

plants and we're like the ones literally

36:40

doing the screws on an iPhone or is it a

36:43

bunch of jobs which are next generation,

36:45

you know, jobs which is like we're

36:46

managing robots and and like shipping

36:48

and logistics grows as a result of this

36:50

and all of the surrounding kind of

36:52

supply chains, you know, start to grow.

36:53

So like like you know to like there's a

36:56

there's an underlying I mean you know mo

36:58

most people on this call have managed

37:00

teams like you do change management you

37:02

lead people to the end state that that

37:04

you want them to sort of see the

37:06

potential in and I think some something

37:08

that kind of gets missed is and the part

37:10

that kind of confuses me is like I don't

37:11

know if if to you know exactly to Ryan's

37:14

point like people are in China because

37:15

of the ecosystem of manufacturing yet

37:17

the messages you get out of the

37:18

administration are like oh you know

37:20

we're going to have fewer toys at

37:22

Christmas time. It's like, no, that's

37:23

not the big picture. Like, the big

37:25

picture is is like, like, you know, this

37:27

is supplying the parts that go into

37:29

building a manufacturing plant and

37:31

building a car that that allows us to

37:33

actually, you know, be even remotely

37:35

competitive in car manufacturing. So,

37:37

where should in your mind this all lead?

37:39

Because you have some thoughts on

37:41

American exceptionalism and maybe

37:43

skating to where the puck is going. So

37:45

for if you were to become an adviser

37:47

Yeah. on this as a technology expert and

37:50

and somebody who spent their whole

37:51

career in it, what would you advise them

37:53

to do? I'd get rid of Navaro immediately

37:55

and and you would basically say, you

37:58

know, Mia Kulpa, like oops. And like

38:00

obviously you need to like land that

38:02

with some really cool trade deals that

38:03

make everybody kind of feel happy. And

38:05

you basically say, you know what, like

38:07

let's go back to the first two days of

38:08

Trump, which is let's announce massive

38:11

deals. We're bringing manufacturing here

38:14

with Stargate. We're doing TSMC. We're

38:16

building Nvidia chips. We're going to do

38:18

a deal which is you get like 5% tax uh

38:21

rate if you build in America. And so you

38:23

just stimulate a a manufacturing boom in

38:26

in the country. You know, we

38:28

incentivize, you know, automation across

38:30

the manufacturing. We use that as a

38:32

competitive weapon to go and compete

38:34

with with the the sort of lowerc cost

38:36

labor that that that happens

38:37

internationally. You know, we we find

38:39

every incentive and tool we can. we

38:41

deregulate, you allow people to build

38:43

these plants and so you don't have to go

38:45

through the, you know, three-year EPA

38:46

process like like you you just

38:48

accelerate from from this position and

38:51

you see you see it all as upside and so

38:54

then business leaders, you know, if you

38:56

go talk to the Fortune 500 company that

38:58

actually has to build, you know,

38:59

anything right now, you give them a path

39:01

to say, listen, we're going to help you

39:03

transition away from your current supply

39:04

chain and we're going to make it even

39:06

more competitive and more compelling in

39:07

America to do that. You know, there's a

39:10

reason that Elon builds in America like

39:12

like he has he's actually made it be

39:14

more effective to to be able to to, you

39:16

know, bring automation to manufacturing,

39:18

to be able to to build locally, but he

39:20

wasn't forced to do that. And so so I

39:22

would just I would argue like you you

39:24

use as many carrots as possible in some

39:27

surgical areas. And Chimath, I've heard

39:28

your points about the like, you know,

39:30

chips, pharma, you know, AI, like in

39:32

those surgical areas, we get tough where

39:35

where necessary. Um, and and if we have

39:37

to do, you know, a couple sort of very

39:39

surgical tariffs, you know, to kind of

39:41

make make people move the the direction

39:43

that we want, that's totally fine. But I

39:45

I mean like like it's like the even

39:47

arguing the premise is hard because

39:48

because we act like we're like like it's

39:50

like countries that are screwing us, but

39:52

actually businesses are are

39:53

independently making decisions about

39:55

where they want their supply chain to

39:57

exist. They they have in a free market

39:59

they've made that decision. They don't

40:01

need the government to tell them where

40:03

where are they supposed to or where are

40:04

they allowed to to have their supply

40:06

chain operate. that that ends up with

40:08

just lots of economic distortions that

40:10

everybody on the the right would have

40:12

called, you know, the left socialists

40:14

for trying to kind of implement central

40:16

planning around supply chains. So that's

40:17

my piece. Well, no. What do you think,

40:19

Chimath, here of this sort of

40:21

reframing/offramp and sort of maybe the

40:23

positive spin on it, hey, if you want to

40:25

make t-shirts, you know, you want to

40:26

make commodity items, have at it. Free

40:29

trade, you know, reciprocal tariffs,

40:31

great checkbox there. But here is a

40:35

series of incentives and a path forward

40:37

to do the advanced stuff to do robotics

40:40

etc. Let me answer this in a different

40:42

way. Okay, a lot of those things he's

40:45

actually doing. I think this is where we

40:49

are is we're beyond

40:52

TDS. There's something that comes after

40:54

it. And I think that the mainstream

40:57

media has just lost their mind to a

41:01

degree that they hadn't even lost their

41:03

mind in Trump one. I'll give you a

41:05

couple of examples. Um well, one

41:08

example, by the way, just a shout out to

41:09

our friend, completely brazen,

41:12

ridiculous, shitty reporting by the Wall

41:14

Street Journal last night. When they

41:16

were told that this, you know, this

41:18

whole Tesla thing was a total farce,

41:20

they continued to publish it. Okay,

41:22

fine. They're they were referring to

41:24

Elon. The board Yeah. starting a search

41:28

to replace Elon and then the board said,

41:29

"Well, wait. We told you we weren't

41:30

doing that." And they didn't even

41:32

mention they communicated that directly

41:34

to the Wall Street Journal. The Wall

41:35

Street Journal said, "I don't care. I

41:37

have an axe to grind." Correct. Yeah. I

41:41

think that Trump has a strength, which

41:43

is he shapes these potholes for the

41:45

mainstream media to fall into. The

41:48

downside of that though is that that the

41:49

mainstream media then doesn't do the

41:51

other part of the job which is to tell

41:53

the things that are important. So for

41:55

example, we spent a lot of time

41:57

breathlessly talking about the MS-13

42:00

knuckles of the guy, right? Or then we

42:04

spent a bunch of time talking about how

42:06

MSNBC blurred out the names of the

42:09

placards on the lawn. Okay, but here's

42:12

the other part where then they get so

42:15

tilted. Here's what they don't report.

42:17

They didn't report that, for example,

42:19

when Trump took a shot at Harvard, he

42:23

also reinforced and strengthened

42:25

historically black colleges and

42:26

universities. Totally did not get

42:28

written. I'll give you another

42:31

example. This past

42:33

week, Bessant said that the tax bill

42:37

will allow you to fully deduct all the

42:41

PPE and all of the incidental costs of

42:45

building a factory. I heard that, I

42:48

immediately went to my wife. She runs a

42:50

pharma business. This is exactly what

42:52

she's trying to figure out. And we now

42:56

are like, how do you build a business

42:57

case if this actually gets effectuated?

43:00

The point is that

43:03

thing would create an absolute economic

43:07

bonanza if it were to get

43:10

passed. Other than people hearing it on

43:12

this pod or randomly maybe finding it on

43:15

a direct clip that Bessant puts out on

43:17

X, there has been zero coverage by the

43:20

mainstream media. Yeah. But but like

43:22

like the the uh first of all that Yeah.

43:25

Okay. MSM and whatever we want to call

43:27

it aside like the that is that is still

43:30

on the administration for driving a a

43:33

change management process that that

43:35

causes people to build on momentum and

43:37

not causes boards to basically say oh

43:40

are we going to pivot our entire supply

43:41

chain this week because because Trump

43:43

you know didn't get a call back from

43:45

from sheet like like that that is like

43:48

this is really not a a a TDS MSM issue

43:51

this if you talk to Fortune 500 CEOs did

43:54

you know about the PPE thing No, but

43:55

that's not but like I'm I don't need to

43:57

like the thing that I know but there are

43:59

many other CEOs that do they they're

44:01

controlling trillions of dollars of

44:02

capital allocation. It's an important

44:04

thing. If we had if we had Mary Bar on

44:06

this call and and we said Mary has has

44:10

you know Trump increased your ability to

44:12

execute and operate and accelerate the

44:13

the transition to the US or has he had

44:16

headwinds that make it tougher to

44:18

navigate right now? Which which way do

44:20

you think she'd go?

44:22

I think that she would give you a

44:23

calculated answer that neither is pro or

44:26

con. I think I think that answer changes

44:27

by the day. I think that if you talked

44:29

to her last week, she would say this has

44:30

been a major headwinds and then

44:32

yesterday to Chimal's point, they did

44:34

this thing where you can depreciate or

44:35

fully expense in year one capital

44:37

improvements or building out factories.

44:39

But also earlier this week, they made it

44:42

so that auto auto parts are not subject

44:44

to the tariffs. They created a huge

44:46

exemption that wasn't there everything

44:49

like they should have. By the way, This

44:51

all speaks if this was a bit planned, it

44:53

should have been there in the beginning

44:54

because these auto auto companies were

44:56

saying, "Hey, this is going to bankrupt

44:57

us if we have to pay taxes on where I

44:59

would take circling back to

45:01

communication and making a crisper and

45:03

clear, Erin, where you are right

45:06

expectation is it's my job to stay

45:08

informed." Okay. As a CEO of my company,

45:11

I try to stay informed and you're right.

45:13

It is hard because sometimes I find

45:15

myself hunting and pecking to find the

45:17

things that matter. But I do put a bunch

45:20

of that responsibility into the lap of

45:22

the people that are supposed to actually

45:24

report the facts. They can choose. They

45:27

didn't have to run that article about

45:29

Elon, which turned out to be total

45:31

and horseshit on the front page

45:33

of the Wall Street Journal. They could

45:34

have talked about what Ryan just

45:36

mentioned as the first article and said,

45:38

"Here completely changes your ro and roe

45:42

calculations for 90% of the S&P 500."

45:45

That was not the article they chose to

45:47

write and to publish. I but I also think

45:49

it comes back to my original point

45:50

around the UDOT loops that they're the

45:52

Trump is the administration is running

45:54

these very tight hey let's take an

45:56

action let's see what happens let's see

45:58

the reaction and then take another

46:00

action and Washington's used to doing

46:01

all these committees that plan

46:02

everything for 5 years or something or

46:04

whatever 18 months and then roll it out

46:06

slowly and they're going hey let's roll

46:08

it out oh crap we're about to cause this

46:10

huge problem in the auto manufacturers

46:12

and they're all telling us they're going

46:12

to go bankrupt okay 3 days later they

46:15

push an update and it's kind of it feels

46:17

chaotic But yeah, to summarize, Ryan and

46:20

Aaron, your position so we can keep

46:21

going through the docket. Hey, a little

46:23

less shock and awe, maybe a little more

46:25

predictability, a little crisper

46:27

communication, and Chimath, I think your

46:28

position is, hey, maybe the mainstream

46:30

media can play a better role here in

46:32

focusing us on what matters. That

46:34

wouldn't be my takeaway. So, yeah. Okay.

46:36

Um, my my What's your takeaway? I mean,

46:38

like like zero shaken like like not a

46:41

little less like like I like my strategy

46:44

would be 100% different. Actually, uh,

46:46

Scott Besson has an incredible podcast

46:48

from like September of last year, and he

46:50

basically said, you know, Bid Biden got

46:53

it all wrong. And I was like listening

46:54

to it. I was like, oh, okay, actually,

46:56

this is kind of cool. Like, he basically

46:58

says, deregulate the US, make it easier

47:01

to build manufacturing in the US,

47:03

increase the GDP, and then you'll be

47:05

able to take in less tax revenue, spend

47:06

less in the government. And it was like,

47:08

oh, this is actually like a glide path.

47:10

We could take we could take the fact

47:12

that we did a soft landing relative to

47:14

the rest of the globe. We're winning in

47:16

AI. We're winning in it in in you know,

47:19

you know, number of categories. We

47:20

obviously need more energy. We need we

47:22

need to bring in manufacturing into the

47:24

US. And and so you you have this great

47:26

momentum which is where we are the tech

47:28

leader, you know, in the world. Let's

47:29

like just pour fuel on that. And so to

47:32

pour fuel on that, you you do you just

47:34

do a series of carrots and and the and

47:36

the winds that build a flywheel of

47:38

positive energy. Like the reason why I

47:40

take I take a little bit of exception to

47:42

to Chimas's MSN point is that I think to

47:44

some extent Fortune 500 CEOs are not the

47:46

ones like oh my gosh like like Rachel

47:49

Matto said this like I'm going to go and

47:51

and you know worry about this topic now

47:52

like like the information coming at them

47:55

is is I'm not talking about the

47:56

information that's presented. I'm

47:58

talking about the information that's

47:59

excluded. How do you get the information

48:02

that's not published and shared broadly?

48:03

No, no, but but come on. Like like

48:05

Goldman Sachs and JP Morgan are not

48:06

writing reports on the fact that we

48:08

might enter a recession because of of

48:10

MSNB's MSNBC's reporting on this topic.

48:13

Like like again, but again, that's not

48:15

what I'm talking about. I'm saying like

48:17

gladhanding some highlevel

48:19

prognostication which nobody ever gets

48:21

right is in my opinion worthless. What

48:24

I'm talking about is the details. So

48:26

when you talk about something as narrow

48:28

and specific as excluding PPE or

48:32

allowing you to double or triple

48:33

depreciate something in a given calendar

48:36

year, fantastic. That is narrow. It's

48:38

precise. It's specific. It's actionable.

48:40

And what I'm saying is if I surveyed the

48:43

500 CEO of the S&P 500, dollars to

48:46

donuts, the overwhelming majority would

48:48

not have known. And had they brushed up

48:50

against that somehow in their normal

48:52

media consumption to then ask their

48:54

teams, the odds of that would have been

48:56

zero as well. So I guess then who do we

48:59

who do we blame for this chim? Is it the

49:01

administration's job or mainstream

49:02

media? But I bet you everybody knows

49:04

about the blurring out of the stupid,

49:06

you know, pictures on the lawn and the

49:08

MS-13 knuckle tattoos. Yeah. Okay. So

49:10

let's uh wrap up on this just really

49:12

lightning round here. Amazon

49:14

flip-flopped on a new tariff

49:15

notification on their websites. Trump

49:19

said he had a great discussion with

49:20

Bezos. He solved the problem very

49:22

quickly. He did the right thing, good

49:24

guy, etc. If you haven't seen this, it's

49:28

something that Teimu is doing. Here's

49:29

what Timu does uh currently today. Nick,

49:33

you have that image. If you could pull

49:34

it up of just when there is a tariff,

49:37

they explain the tariff coming into the

49:39

country. They put it as like a line

49:41

item. I thought this was actually kind

49:44

of cool. I don't know why people take

49:45

offense, Brian. This is pretty standard

49:47

stuff. So Amazon's competitor Tim Teimu

49:49

is putting in the import charges. They

49:51

don't say tariffs. They don't say taxes.

49:53

Import charges. This is like a standard

49:54

thing. This happens in other countries

49:56

too. What is this? What is Teeu? This is

49:58

like a dollar store. It's basically like

50:00

a dollar. Is it this the last place you

50:01

would ever

50:02

buy jeans? Yeah, you can buy $12 jeans.

50:06

Basically, your left sock from Laura

50:08

Piana costs less than Timu's entire

50:12

inventory of jeans. The point being, um,

50:15

I thought this was actually a plus for I

50:19

think they totally misplayed this. They

50:21

they they did it, they rolled it out,

50:23

then they got criticized. I think they

50:24

were called a treasonous company from

50:25

the White House press uh, you know, by

50:27

the press secretary. Uh, they totally

50:29

misplayed this because they should have

50:30

gone and leaned into it and said, "Yeah,

50:32

we're showing you all these tariffs when

50:34

you buy from China. If you buy from

50:36

America, you don't have to pay any

50:37

tariff and look at all these other

50:39

products." Come on. Come on. That would

50:40

have lasted three and a half seconds.

50:41

This is exactly consistent with the

50:43

other issue which is they're playing

50:45

whack-a-ole. Okay, we're going to do

50:46

something with the automakers. We're

50:48

going to we're going to try and solve

50:49

some problem with Amazon. Like like this

50:52

is a sign that that it like it's not a

50:55

good strategy if you have to do this

50:56

much whack-a-ole. Like like they they're

50:58

not like they can't cover up what Amazon

51:01

is going to end up dealing with because

51:02

there's going to be 500 other retailers

51:04

that that don't get the call with Trump.

51:06

So, so this is like that that like to me

51:08

that's evidence of of clearly this they

51:11

didn't think through the entire

51:12

downstream set of of of conditions that

51:14

are going to change as a result of this.

51:16

Sure. Yeah. I I thought this was a big

51:19

win uh Chimoth because they could then

51:23

have Amazon Here's um a mockup somebody

51:25

made. I'll pull it up here. It was

51:27

interesting. They could to Aaron's point

51:30

just show hey here's a bunch of American

51:32

companies by American when you do a

51:33

search. Ryan's point. I'm sorry, Ryan's

51:36

point. Hey, here's what it might look

51:37

like. Pull pull that um the OralB

51:39

toothbrush one up, Nick, if you got it

51:41

right there. So, somebody mocked this

51:43

up. I think this could be the hugest

51:44

win. You could have the retailers do buy

51:48

American, buy it once, buy a high

51:50

quality product from America. If you

51:52

look here, we don't have the products.

51:54

It wouldn't work. We don't have the

51:56

products.

51:58

Well, I mean, we do have for some

51:59

products, you know, Americanmade

52:01

products. Uh, you know, I buy my boots

52:03

from Danner and those are all American.

52:05

Yeah. So, we should just go back to

52:07

communism and we're all going to make

52:08

our our shoes. Like, it's like that's

52:10

like we we're in a global market. Like,

52:12

we buy from everywhere. Chamop, any

52:14

thoughts on this? I mean, obviously

52:16

there's the whack-a-ole angle. There's

52:19

buy American and be proud of it. There's

52:22

communication. Here's the narrow

52:23

question. I I got a bunch of emails from

52:26

people and a bunch of them were Amazon

52:29

sellers and I don't know Nick if you can

52:32

find it but I posted their comments and

52:34

I reshared them just to kind of

52:36

highlight the issues that they were

52:38

going through and at the core of it was

52:41

a feeling by them that Amazon had

52:45

abandoned them as American purveyors and

52:48

sellers of goods and that Amazon on the

52:52

margins had attended

52:54

to

52:56

help competitors from abroad come stand

53:00

themselves up and compete and

53:03

essentially cannibalize on price and

53:04

margin. This is my view completely and

53:06

that this is the biggest opportunity

53:08

that I think the Trump administr

53:10

administration's flying at 40,000 ft

53:11

doing macrolevel negotiations and look

53:13

and failing to see some of these micro

53:15

optimizations that are really really

53:16

real. So you in the United States you

53:18

can import goods as a foreign company.

53:21

You don't not you do not have to create

53:22

an LLC or any sort of registered entity

53:25

in the United States to import goods.

53:27

Sometimes they say, "Oh, Americans pay

53:28

the tariff." Like that is not true. In

53:31

many many cases, the foreign company

53:32

just imports this stuff and they sell on

53:34

Amazon. And when they get caught

53:36

cheating, they can they can lie about

53:38

the valuation and pay a lower tariff.

53:40

They can change the classification and

53:41

pay a lower tariff. They can import

53:43

stuff that, you know, is harmful to

53:44

children, has lead, paint, whatever

53:46

else. There's no enforcement at all. You

53:48

can't. So you're saying Amazon third

53:50

party is like a bit of a backdoor to

53:52

abuse the system, right? I mean, Amazon

53:53

is sort of just playing the game that's

53:55

on the field, but they're this is legal

53:57

in the United States is these companies

53:58

import stuff and I think it's 60% of all

54:00

of all the sellers on Amazon are these

54:02

Chinese registered. They're not

54:03

registered in the United States at all.

54:04

Just Chinese companies. Which sounds

54:06

profoundly unfair in terms playing

54:09

field. Can we just take a step back and

54:10

also acknowledge that we are talking

54:12

about a level of detailed issues that we

54:16

would never have talked about 6 months

54:17

ago. That there was

54:19

no interest in even bringing this up.

54:22

Like if if Ryan wanted to bring up the

54:25

sort of hollowing out of

54:28

American salesmanship, let's say, if you

54:31

will, because of like this arbitrage

54:33

that Amazon does for GMV. That would

54:36

have been a snoozefest. except today it

54:39

can actually get a lot of awareness and

54:41

Erin mentioned this and I've mentioned

54:43

this before but like I think that there

54:45

are four things that really matter

54:46

batteries AI pharma APIs and rare earths

54:50

that now is on the agenda I think the

54:51

the positive way to look at this is the

54:54

American economy is too complicated if

54:56

you had waited Aaron for a study for all

54:59

of the implications we would have been

55:01

waiting forever and nothing would have

55:03

happened and I think that we've made

55:05

macrolevel moves you're right and now we

55:07

are finding what the implications are in

55:10

course correcting in real time. And I

55:12

hope what happens though is when we find

55:14

these big thorny issues, I think the

55:16

Amazon thing is a is a pretty

55:18

interesting issue actually about like

55:19

American competitiveness. Now the

55:21

question is do we follow through and get

55:23

to the root cause of it and and fix it.

55:26

And those feedback loops are there. I

55:27

mean the Trump administration is going

55:28

to act on this and if then there's an

55:30

act that's coming out of Congress as

55:31

well to shut down the foreign import of

55:33

records. So those feedback loops are

55:34

there in ways that I don't I don't know

55:36

if they were there in the past. Go

55:37

ahead, Erin. We can totally chaos monkey

55:38

the the economy and just see what

55:41

breaks. I think that the you know you

55:44

you you sort of phrase the we could, you

55:46

know, do the research paper and we could

55:48

do, you know, Aspen Institute as a bad

55:50

thing, but also it can be a bad thing if

55:53

you're the small business owner right

55:54

now who has 30 employees and you just

55:56

literally don't know what you're going

55:57

to do next month. And so, so that that's

56:00

that's sort of then the argument to not

56:02

counterbalance. That's like why you do

56:04

have some bureaucracy and and you don't

56:05

chaos monkey the economy and why Rand

56:07

Paul is literally saying we shouldn't

56:09

actually let you know have unilateral

56:11

you know control over tariffs. So you

56:13

know interesting that that like dynamic

56:16

there. All right, you want to uh wrap us

56:19

up here or you want to pass? Oh, am I

56:22

still on the pod? Yeah. Well, you turned

56:25

your camera off. So, look, I spent 80

56:28

minutes debating this topic with Larry

56:30

Summers three weeks ago.

56:32

The point I made then is that we had in

56:36

this city at for 25 years a globalist

56:38

consensus on trade that distorted a lot

56:42

of outcomes. And I don't need to rehash

56:44

that debate, but I'll just recall that

56:47

Larry Summer's main argument for why

56:50

this would not work out is that the

56:52

market was down. Do you remember that?

56:53

That was his evidence. Yeah. That this

56:56

wasn't going to work. Okay? And it was

56:58

all about the market not pricing in

57:00

lower expectations. Well, guess what?

57:02

The market is actually up since

57:04

Liberation Day on April 2nd. So, what

57:07

happened 3 weeks ago was basically a

57:10

panic in the market over this policy and

57:13

the media has been trying to fuel that

57:15

panic. Now, what I said as well is we do

57:18

have to stick the landing on this. I

57:20

mean, President Trump shifted the

57:21

conversation away from this globalist

57:24

consensus and he's now redefined the

57:27

debate, but it is now up to Scott

57:29

Bessant, the Treasury Secretary, Howard

57:31

Lutnik, the commerce secretary, Jameson

57:33

Greer, the US trade rep, and so on, the

57:35

Trump trade team, to now negotiate these

57:38

deals, stick the landing. And I agree

57:40

with you to the extent that the sooner

57:42

that is done, the better because it is

57:44

good to provide business certainty. But

57:47

the idea that so far this hasn't worked.

57:50

I think again the main argument against

57:51

that was the market reaction that now

57:54

the market's not positive. So I think my

57:56

point is just we need to give this time

57:59

to work. I think it's too soon to be

58:01

judging this policy as if it hasn't

58:02

worked yet. It needs to be executed

58:04

properly. And quite frankly, Ryan, I

58:06

mean, I remember the last time you were

58:07

on this pod, you were coming on about,

58:11

wasn't there like some union deal that

58:13

was supposed to shut down all the ports

58:14

and all the shelves would be empty? That

58:16

never happened either. It did happen. It

58:18

did. They shut down for 3 days. I don't

58:20

Okay. I don't remember the the shelves

58:21

being empty, which is now the new panic

58:24

the media is trying to create. So, look,

58:26

there's a lot of pants wedding that's

58:28

occurring here that's being fueled by

58:29

the media. Well, well, I I do want I

58:31

want I do want to I just want to

58:32

bookmark one one thing. The only thing I

58:33

actually um I was more frustrated

58:35

listening to the Larry Summers and and

58:37

you conversation because I was like, why

58:38

Larry make this point like come on.

58:40

Like, what? Don't go down the WTO rat

58:42

hole. Like, that's that's not relevant.

58:43

So, here here's the other thing. Of

58:45

course, it's relevant. It's how we got

58:46

here. No, no. Like like as in like

58:49

that's 25 years ago. Like let's worry

58:50

about literally today and like what what

58:52

do we do going forward today? And the

58:55

only thing I just want to say because I

58:57

because I I do think that that you know

58:59

I appreciate your point about hey

59:01

there's like you know everybody's

59:02

freaking out whatever but to be totally

59:04

fair that some of that freak out whether

59:07

we can decide how emotional it needs to

59:08

be is the reason that then Trump walks

59:10

back the things that then cause the

59:12

market to correct. So, so we can't just

59:14

say the market's back and and like see

59:16

we didn't need to freak out cuz it was

59:18

literally I said we have to make the

59:19

deals. We have to stick to the landing.

59:21

But look, China over the last 25 years

59:23

has been able to strategically

59:25

annihilate our rare earth processing

59:28

capability and our ability to cast rare

59:30

earth magnets. We just sat back and

59:32

watched as the market basically went to

59:35

the lowest bidder which was being

59:36

subsidized by the Chinese government

59:38

which the WTO allowed them to do. And

59:40

now we have a critical dependency. Yep.

59:42

In our supply chain on China for

59:45

basically every electric motor in every

59:47

product, including cars. That was crazy.

59:49

We should not have allowed that to

59:50

happen. How are you going to change

59:52

that? So, we needed to shift the

59:54

political conversation to recognize the

59:57

ways in which free trade led to unfair

60:00

trade and created unacceptable

60:02

dependencies on the for the American

60:04

economy. Wait, wait, just I just want to

60:07

make one one point. It's more than that.

60:09

point and then I'll but this is the

60:11

national security of the United States

60:13

that that's at stake. Let's go take your

60:16

favorite pet issue. China invades

60:18

Taiwan. Okay. And we have to take a

60:21

side. Jason, my pet is hold on. Let me

60:24

just finish. And the and the and the

60:27

Chinese say here are the implications of

60:30

supporting Taiwan on this A, B, C, and

60:33

D. You don't get any pharma APIs. You

60:35

don't get any rare earths. You don't get

60:36

any batteries.

60:38

Okay, it'll it'll send life back 50

60:41

years. Or let's say China and India get

60:44

into a fight and we're forced to pick a

60:46

side. Same situation. The point is

60:48

there's all these scenarios that we

60:50

never even

60:52

considered us being able to have

60:55

strategic optionality to make the

60:56

decision that's morally and ethically

60:58

right for the United States. And I think

61:00

that we have learned through this lens

61:03

that these are huge issues. The thing

61:05

that the Chinese did that was so

61:06

brilliant, which we still don't have an

61:08

answer for is they have these national

61:10

champions. And being a national champion

61:13

allows you, and we'll talk about this in

61:14

AI, it allows you to blur the lines

61:17

between the public and private

61:18

partnership. It allows you to blur the

61:20

law. It allows you to blur capital. And

61:22

I'm not saying we have to do that, but

61:25

what I am saying is we need to have our

61:27

own answer to it. And that was never on

61:30

the table until April 9th.

61:33

All right. So yeah, but but like 100%

61:36

like do that strategy and then and then

61:38

and then don't have a mad rush. What is

61:40

the strategy? No, no, no. Because

61:41

because that's a you if you have a, you

61:43

know, Ryan, what what is the number? I

61:44

don't know, trillion of imports or

61:45

whatever, like you don't need everybody

61:47

then than then jamming the system to

61:49

build their supply chain, you know, in

61:51

the US to solve that problem

61:52

immediately. If we're sitting here in 9

61:54

months and you're saying this and there

61:56

are no deals, I would say that you're

61:58

right. what Howard Lutnik said last week

62:01

and again we may have all gotten caught

62:03

up in the knuckle tattoos and we missed

62:04

this but he was very clear we have a

62:07

country a deal is already done we're

62:09

convening parliament it's going to be

62:11

the first of many so for all we know

62:14

there's like 30 deals that are waiting

62:16

in the wings and the first one will set

62:18

the tone and and I think that Sax is

62:21

right here which is it's way too early

62:23

to declare defeat and that it was quote

62:25

unquote chaos I think if we're sitting

62:27

here in 9 months and foreign direct

62:29

investment has shriveled up and domestic

62:32

investment has shriveled up because

62:34

there is just no continuity. You have a

62:36

claim. But that's No, no, because I I I

62:38

don't I don't think that'll happen. I

62:40

don't I don't think that'll happen. I

62:41

think we will end up in like I'm I'm

62:42

with Ryan like we will end up in a good

62:44

spot because we'll iterate through this.

62:45

My my only point is there's an

62:46

alternative path that that could have

62:48

occurred. It could have been done in a

62:50

more thoughtful well-communicated

62:52

pattern instead of hey let's do barrel

62:53

rolls with the airplane. I don't

62:55

disagree with you, Aaron, and I can tell

62:56

you we've already started to see layoffs

62:58

and nobody wanted to even initiate the

63:00

barrel roll, guys. Yeah, listen, we get

63:03

it. It's like, hey, I don't want

63:05

anything to change. I think we agree to

63:07

disagree on this. We got to move on to

63:09

the next one. Hold on. This one last

63:11

point. Excuse me. Excuse me. You didn't

63:13

call me for 40 minutes. I just want to

63:14

make one final point. Eron, here he

63:16

goes. Erin, where were you with this

63:18

perfect plan? Yeah. Where were you with

63:21

this perfect plan before Liberation Day?

63:23

I was telling Kla about it. You were

63:25

telling Kamla. Okay, great. They were

63:27

having nobody nobody ties and they were

63:31

talking about this specific issue. All

63:33

the people who suddenly know what the

63:35

perfect plan is and how to perfectly

63:37

execute it. No barrel rolls had nothing

63:39

to say about this topic for 25 years.

63:41

Now all of a sudden they've come forward

63:43

with their perfect plans. I would say

63:46

that's victory for Trump. Finish. It's

63:47

like look at this. The best thing of all

63:49

of this is you've got the Liberals

63:50

embracing Milton Friedman and their

63:51

backgrounds on there. I love it. Yes. Uh

63:54

All right. Listen, we're going to agree

63:55

to disagree. We're going to agree to

63:56

disagree on this one. You know, the

63:58

Liberals love the stock market. Listen,

64:00

Sachs Kla is coming on next week. We're

64:02

going to make some cocktails. It's going

64:03

to be wonderful. You we'll ask her some

64:06

direct questions about it. But I want to

64:08

talk about AI agents. 2025 shaping up to

64:11

be the year of AI agents. Tons to talk

64:14

about here. Open AI is planning to

64:16

charge between two and 20K a month for

64:18

different levels of AI agencies would be

64:21

basically cron jobs they would run in

64:22

the background and do things for your

64:24

company that uh humans are doing right

64:26

now. You may have heard of this uh

64:28

agentic tool. Again agentic is just a

64:30

fancy word for agent which is a fancy ro

64:32

word for like a cron job that just runs

64:35

uh perpetually. Yep. And uh Manis is uh

64:39

the company in China that started this

64:41

weirdly benchmark invested in it that's

64:43

created a whole bluff on the side and

64:46

Manis' website has a really good

64:47

visualization of what these agents would

64:49

look like. So first of all I think

64:50

you're giving a little too much credit

64:51

to Manis. They didn't come up with the

64:53

agents but I do think that they have a

64:55

very good demo and it's hard to know

64:57

exactly how real it is because not

64:59

everyone's used it and it's from China.

65:01

It's from China. I'll get to that in a

65:02

second. If you go to their website, you

65:05

can see a bunch of their demos. And I do

65:08

think that what they deserve credit for

65:09

is advancing the ball on the UI

65:13

paradigm. And it's not that other people

65:16

weren't doing this. I mean, I think

65:17

notably Anthropic was doing this with

65:18

its operator product, but the basic idea

65:21

is that you've got this two-pane view

65:23

and in one window, you've got the

65:25

standard chatbot interface and then in

65:27

the other view, you can see what the

65:29

agent is doing. And that agent has the

65:32

ability to toggle between it currently

65:34

four apps. There's search, browser,

65:37

code, terminal, and document editor. And

65:40

so when you give Manis a task, the first

65:42

thing it does is create a to-do list in

65:44

the document editor. You can kind of see

65:46

it there. And then it works sequentially

65:48

to achieve each of those tasks and then

65:49

puts an X on them there. And you can

65:52

kind of see it working. And I think

65:53

what's cool about the demo is just the

65:55

way that it seamlessly toggles between

65:57

those four apps. and you can see what

65:59

the AI agent is doing, you know, and

66:01

it's browsing the internet, it's

66:02

searching for things, it's writing

66:04

documents, it's crossing things off its

66:06

to-do list. Now, I think it's pretty

66:08

easy to imagine where this goes, which

66:10

is you'll be able to connect an agent to

66:13

all of your SAS apps. So, it won't just

66:15

be four applications. It'll now be

66:17

connected to dozens of applications,

66:19

including ones that already have your

66:21

data. And it's going to know what

66:23

actions it's possible to take in those

66:24

apps. So when it creates its to-do list,

66:26

there's a much wider range of things

66:28

that it can accomplish. And in fact,

66:30

there's a new standard called MCP which

66:33

is taking off like wildfire which is

66:35

built specifically to enable agents to

66:37

connect with applications and understand

66:39

the data and understand the actions that

66:40

are possible in those SAS applications.

66:43

So look, Manis is just at the tip of the

66:45

iceberg here. I think this will become a

66:46

very standard UI paradigm. That's the

66:48

reason why I mention it. Not because I

66:50

am predeclaring them to be the winner in

66:52

the space, but just because I think

66:54

there's a lot of talk about agents and I

66:56

think it's hard to conceptualize what

66:59

that means without just seeing it

67:00

visually. A great great summary there,

67:02

Saxs. And Aaron, I want to get your

67:04

thoughts on it because obviously you're

67:06

running Box and and you have your your

67:08

finger on the pulse of this. We actually

67:10

started building one of these in our

67:11

venture firm. We have 20,000

67:14

applications a year and we have updates

67:16

coming in from investments. We are now

67:18

taking those sacks Aaron and we are

67:20

having an agent sort them and then look

67:23

for competitors and compare them to the

67:25

last update and we're looking into our

67:27

notion our kod and saying what else have

67:30

what other communications have we had

67:32

what questions should we ask about the

67:33

startup and about their strategy and

67:35

then we're presenting that in Slack to

67:37

our team. So this is coming fast and

67:39

furious and we spend I don't know

67:42

probably 15 minutes on each of those

67:43

incoming applications. You start doing

67:45

the math on that. Talk about 5,000 hours

67:47

of work. Aaron, what are you seeing on

67:49

the street? What are you doing at Box in

67:51

terms of agents landing right now in Q2

67:55

of 2025? Uh, yeah. I mean, I think I

67:58

think Sax represented it well, which is

68:00

which is, you know, you have to now

68:02

think about AI as as effectively being

68:04

able to do anything on a on a computer

68:06

or another piece of software as as a

68:08

human can do. And the little distraction

68:11

that that I think happened two years ago

68:13

after the chatbt moment was we sort of

68:15

thought about that as oh we're just

68:17

going to now you know do like typing

68:18

information retrieval and that's a new

68:20

paradigm for user interfaces let's say

68:22

so you just like talk to your software

68:24

and you like search Zillow via chat that

68:27

was sort of a little bit of a

68:28

distraction that that's super helpful

68:29

like when you want basic information

68:31

lookup or whatnot the big breakthrough

68:32

was starting to think through these

68:34

things as as full you know effectively

68:37

uh uh you know agentic systems that that

68:40

operate on any amount of data, any

68:42

amount of tools for as long as you want

68:44

to complete any task that you want. And

68:46

this is sort of the big year where

68:48

agents are starting to, you know, enter

68:50

the vocabulary of enterprises, of IT

68:53

people, of, you know, larger and and

68:55

certainly small organizations. Um, and

68:58

and it kind of requires you to have a

68:59

little bit of a of a reset moment on how

69:01

you think about AI, which is which is

69:03

it's not just now a kind of a co-pilot

69:05

that you talk back and forth with. it's

69:07

actually something running behind the

69:09

scenes that's now actually starting to

69:11

deliver, you know, real automated, you

69:14

know, kind of work for you. And so lot

69:16

lots of implications like, you know,

69:18

massive implications to what the

69:19

software business model is in the

69:20

future. You know, I I I would argue, you

69:22

know, strongly that's a massive TAM

69:23

increase um because now software starts

69:26

to go after labor spend. It completely

69:28

changes the dynamics of then, you know,

69:29

how do you build a moat in a world of AI

69:31

agents? Um uh but but I think unpack

69:34

that piece there Erin. You said

69:35

something very interesting how you how

69:37

software then is going to go over human

69:40

spend. Yes. Explain that concept. Unpack

69:43

it for a second. David and I, you know,

69:46

we we we go back way back in SAS land,

69:48

but like you used to basically just, you

69:50

know, you built you built a piece of of

69:52

software and you sell it for the number

69:54

of of people in the organization. And

69:56

so, you know, company has 500 employees

69:59

and uh and you sell that thing for,

70:01

let's say, $10, you know, a user a

70:03

month, you know, 120 bucks a year and

70:05

you make 60,000 bucks. Um, and so so

70:07

that that's kind of the business model.

70:09

Uh, now when your software actually

70:11

brings the underlying workflow to the

70:14

customer or the underlying outcome to

70:16

the customer, so you know that that

70:18

company might have 10 lawyers and so

70:20

previously if you were selling software

70:21

for lawyers, you had a maximum amount of

70:23

10 seats that you could sell. Now all of

70:25

a sudden if your AI agents are doing the

70:27

equivalent of let's say parallegal work

70:29

or some form of professional services

70:31

all of a sudden you might be able to

70:32

sell a multiple of the initial kind of

70:34

10 seats that you would have sold

70:35

previously. So you see it as a huge

70:37

opportunity because now you're not

70:39

enabling a human to be 5% more

70:41

productive. You're replacing a human or

70:42

you're replacing one out of 10. Yeah.

70:44

And and actually I'm going to take a

70:45

massive I'm going to do an underscore on

70:47

this point though. I I don't I don't

70:48

like the word replace because I I think

70:50

actually most of the upside is actually

70:52

going to be for companies that now

70:53

deploy labor at things that they

70:55

wouldn't have deployed labor at before.

70:57

And and you know, maybe I'm biased from

70:58

the view we have, but most of our

71:00

conversations with customers are when

71:02

they have AI agents, it they can

71:04

actually now go and actually deliver

71:06

work in areas that would have been

71:08

unaffordable previously. So they

71:09

actually they weren't doing the work.

71:12

True in logistics. So, like we we're

71:14

making thousands of phone calls a day

71:15

using AI, calling truck drivers, and we

71:18

if we have a load, we've got 400,000

71:20

truck drivers using the mobile app, the

71:22

Flexport mobile app. I don't have enough

71:24

loads to keep them all checking it every

71:26

day to see if there's a load that

71:28

matches them. And if they don't check

71:29

it, I they're they're useless to me now.

71:32

And it was too expensive to call the

71:33

truck driver and have a human call and

71:35

talk to them, even if it's a human in a

71:36

call center in the Philippines. Whereas

71:38

with AI, it's almost free. I'm calling

71:40

thousands of them a day going, "Hey,

71:42

this load looks like it's a good match

71:43

for you. Are you interested?" And and

71:45

then we activate them on the platform.

71:47

Um, that's new work that wasn't going to

71:49

happen before, not just a replacement.

71:51

Yeah, I I think that I think probably I

71:53

think like we we have, you know, in the

71:55

valley, unfortunately, we've been

71:56

co-opted a little bit somewhat with with

71:58

a little bit of a doomer, you know,

72:00

mindset in some areas and and we think

72:02

of then AI is, okay, like like it's all

72:04

fixed pie, it's going to replace things.

72:06

And on the ground with large

72:08

enterprises, the vast majority of the

72:10

use cases are are it's it's the you

72:13

know, it's the ability to finally review

72:14

the contracts that we never got around

72:16

to reviewing. It's the ability to

72:17

finally automate an invoice process that

72:19

we never did. It's the ability to go in

72:21

and just create marketing campaigns in

72:23

every language that we never got around

72:25

to. And so so I think that'll probably

72:27

be actually like 90% of the usage of AI

72:29

in the future will be things that if we

72:31

look back and we snap the line right now

72:32

and we said this is what knowledge work

72:34

is today. 90% of AI usage will be things

72:37

that we don't do today. 10% will will

72:39

replace you know what we're what we're

72:40

doing in some areas. I think that's the

72:42

right take cuz Chimath I can tell you in

72:45

our firm we would never have associates

72:46

or researchers or analysts sax you also

72:49

were in this line of work as well

72:51

venture capital you would never have

72:52

them review legal documents. That's

72:54

something lawyers would do in the legal

72:55

department. But now because of AI, we

72:58

have them say, "Here's the safe. Here's

72:59

the term sheet. Here's the edited

73:01

version, dump it all in, find out what

73:04

the changes are, what are the deltas

73:06

here, and and then let's have a

73:07

discussion about what the founder

73:08

changed in a Sander document, and we

73:11

don't have to bother with an attorney,

73:13

and maybe you wouldn't have even checked

73:14

those documents if you were, you know, a

73:16

seed fund or an angel fund. You would

73:18

just go along for the ride because

73:19

you're the 10th person signing the

73:21

documents." So what what do you think

73:22

Chimoth here in terms of the premise

73:24

that maybe it's 10% replacing work

73:27

that's happening but this is blue ocean

73:29

and we're going to do 90% of like new

73:30

stuff that we just never got to. Yeah, I

73:33

tend to I tend to believe that's true. I

73:34

think the customers that we sell into at

73:38

8090 are largely large enterprises as

73:41

well. So not dissimilar to Aeron's

73:42

customer base. What I would say is that

73:45

what they are encountering is the trough

73:49

of disillusionment.

73:52

And I don't know if Erin, you're seeing

73:53

this as well, but every single, you

73:56

know, CIO ran

73:58

around signing up some sort of AI

74:01

product in large part because their CEO

74:04

would say to them, hey, what's your AI

74:07

strategy? And the reason the CEO asked

74:09

them that is that at some point somebody

74:10

on the board said, what are we doing

74:12

about AI? So that's the the

74:15

cascade that that that we went through

74:17

in the last two years. And I think what

74:20

has happened now is people have spent

74:22

billions and billions of dollars. I

74:24

think you can see it in the revenue

74:25

traction of the AI

74:28

companies. But I think where we are

74:30

today is that there are some real

74:33

technical complexities that have not

74:35

been solved. I'll give you an example.

74:37

We have a lot of customers in regulated

74:39

industries which is to say that if you

74:42

make a

74:43

mistake you will get fined or you will

74:46

get shut down. Life sciences,

74:48

healthcare, financial services are three

74:52

examples. People still don't seem to

74:54

appreciate that when you replace

74:56

software that is deterministic with

74:59

software that is probabilistic, meaning

75:01

software that somebody wrote for

75:04

you, do A then do B, then do

75:08

C with an LLM that can

75:11

hallucinate, you'll have

75:14

errors. So what used to be a throwaway

75:17

thing, which is quality assurance and

75:18

QA, right? unit testing, integration

75:21

testing is now the only thing that

75:22

matters. Why? Because if you're a

75:24

financial services institution and

75:26

you're supposed to do KYC and hit BORS

75:28

and now all of a sudden you send a wire

75:30

somewhere in Syria, guess what? You're

75:32

in trouble. Bueno. Yeah. If you're a

75:34

healthcare company and you're supposed

75:36

to do some clinical diagnosis to send

75:38

out a drug on time and you don't do that

75:41

because the the model

75:42

hallucinates, that's a real problem. And

75:45

I am guaranteeing you, we have not seen

75:49

the class action lawsuits that will come

75:52

when those errors will eventually be

75:54

made. They're guaranteed to be made. We

75:57

just don't know the scope and the scale

75:58

of them. So that's why I'm sort of of

76:01

this posture where I think we've sold in

76:06

a ton of promise. I think the reality is

76:09

much more tactical. It's a little bit

76:11

more benal. I think we're sorting

76:13

through the exact use cases where you

76:15

can put guard rails around these error

76:18

rates where it's okay and tolerable.

76:19

Like Ryan will probably tell you there's

76:22

some number of phone calls that just

76:23

sound totally fcocked but he's okay with

76:27

that because the broader thing is okay

76:29

and Aaron will so I don't know. So I

76:32

don't want to talk to my customers

76:33

though. I have it calling truck drivers

76:35

to offer them you know offer them loads

76:37

but I'm not having to talk to my

76:38

customer. Sorry, I I meant your truck

76:40

drivers, but my my my point just is that

76:43

I think agents are

76:45

real, but I think that we are far away

76:48

from that because we're still at the

76:49

phase of how do you build reliable

76:52

software in production for an enterprise

76:55

versus the toy apps that you see on the

76:58

internet which is like let me vibe code

77:00

something. I think these things are

77:02

worlds apart still. Okay, so let me get

77:03

saxed in on here and just to inform the

77:05

audience you heard tri of

77:06

disillusionment. This comes from the

77:08

hype cycle. This is something Gartner

77:10

has been talked about for a long time.

77:11

So in case you're taking it for granted

77:13

if you're watching, you have some sort

77:15

of technology trigger like agents. You

77:16

have this like peak of inflated

77:18

expectations. Now we're in the trial of

77:20

disillusionment. Hey, this stuff doesn't

77:21

work. It's hallucinating. But we're kind

77:23

of going up that works. It just doesn't

77:26

say we're on the slope of I don't see

77:28

the disillusionment. I don't know where

77:29

this is coming from. I don't even think

77:30

we're at the peak yet. Oh, okay. So you

77:33

think we're still going up? cuz I I a

77:35

lot of people to Shimat's point were

77:37

buying stuff and saying, "Hey, it

77:38

doesn't work, you know, and now we're in

77:40

the mess." Let me say let me say it

77:42

differently, Sax. I think we have not

77:44

yet figured out how to move the budgets

77:46

from experimentation to mainline

77:49

production. Meaning where large chunks

77:51

of the US economy are comfortable enough

77:55

with the ways in which hallucinations

77:57

are managed such that they will replace

78:00

legacy deterministic code with this new

78:03

probabilistic model generated code

78:06

meaning model enabled code. Let's just

78:08

put it that way.

78:11

Where are we on the slope here? Yeah.

78:12

Look, I would I would separate change

78:15

management issues, which are always

78:16

going to be important and there's always

78:18

going to be big ones whenever there's a

78:19

big disruption, especially in enterprise

78:21

and especially around compliance and

78:22

legal and all that kind of stuff. I

78:23

would separate that from the impact of

78:25

the underlying technology trend. And I

78:28

don't think the impact has come anywhere

78:30

close to peaking yet. And in fact, I

78:33

would say the rate of progress is

78:35

exponential right now on at least three

78:38

key dimensions. So number one is the

78:40

algorithms themselves. The models are

78:42

improving at a rate of I don't know

78:43

three to four times a year. They're not

78:45

just getting faster and and better, but

78:49

qualitatively they're different.

78:50

Remember, we started with pure LLM chat

78:52

bots. Then we went to reasoning models.

78:55

And the difference there is with a

78:56

chatbot, I just it's like kind of a

78:58

smart PhD or college student giving you

79:00

an answer off the top of their heads.

79:02

The reasoning models, it's more like the

79:04

PhD saying, "Okay, let me go off and

79:06

think about that. Let me do a project on

79:07

that." And it could work for 30 seconds

79:10

or a couple of minutes. I mean, as much

79:11

compute as you want to throw at it and

79:13

it will break down your complicated

79:15

question into a bunch of sub questions

79:16

and then it'll try different approaches

79:18

and it can validate some of those

79:19

approaches and come back to you with a

79:21

much more impressive answer. And if

79:22

you've been using like the Gro 3 deep

79:25

research or the new Chad

79:29

GBT3 to do these types of new reasoning

79:32

models, it's pretty mind-blowing what

79:34

they're capable of. Have we even come

79:36

close to figuring out how to tap the

79:38

potential there, especially in an

79:39

enterprise context? No, but my point is

79:42

that the rate of progress on the

79:43

algorithms is again three or four

79:47

times

79:49

here. Okay, go finish about sex and then

79:52

I'll I'll take it and pass it. Go ahead.

79:53

Well, I was trying to lay out the

79:54

dimensions of which progress is

79:56

proceeding exponentially. Okay, so one

79:58

is the algorithms. Okay, which is not

80:00

just quantitative, it's also

80:01

qualitative. We didn't even get to the

80:03

agents part of it yet, but that's the

80:05

next big leap after reasoning models.

80:06

We're just starting to scratch the

80:08

surface there. Then you've got the

80:10

chips. I mean, the chips are getting

80:12

better at, I don't know, three to 4x a

80:14

year. We've gone from, you know, the

80:15

H100 to the H200. Now we're on the

80:18

GB200. We'll be a GB300 soon. We'll be

80:22

on to three times better per year or

80:24

they get better three times per year.

80:26

No, no, no. They're getting the chips

80:27

themselves, depending on how you measure

80:29

it. Each generation of chips is probably

80:31

three or four times better than the

80:32

last. Okay. and Nvidia is back to

80:36

rolling out new chip, new generation of

80:38

products roughly annually and I'm just

80:40

using them as one example. Obviously

80:42

there are other companies as well. So

80:44

basically the lead from Hopper to

80:46

Blackwell to got it Reuben I guess will

80:48

be in next year and and then I think

80:50

Fman's coming after that. I mean really

80:52

an astounding rate of progress. It's not

80:54

just the individual chips are getting

80:55

better. They're figuring out how to

80:56

network them together like with NVL72.

80:58

It's like a rack system to create much

81:01

better performance at the data center

81:03

level. And that would be like the the

81:04

third area where you're seeing basically

81:06

exponential progress. Just look at the

81:09

number of GPUs are being deployed in

81:11

data centers. So when Elon first started

81:13

training Grock, I think they had maybe

81:15

100,000 GPUs. Colossus was 100,000.

81:17

Correct. Right now they're up to

81:18

300,000. They're on the way to a

81:20

million. Same thing with OpenAI's data

81:23

center, Stargate. And within a couple

81:25

years they'll be at I don't know 5

81:27

million GPUs, 10 million GPUs. So and

81:29

you see that on the power side, right?

81:31

You're going from 100 megawatt data

81:32

centers to 300 megawatts to we're just

81:35

starting to now see the first gigawatt

81:37

power data centers. I don't even think

81:39

they're live yet, but this is where

81:40

they're trying to get to. And I don't

81:42

think it's beyond the real possibility

81:44

that we could be at 5 or 10 gawatt data

81:46

centers in the next I don't know several

81:48

years. So, so my point is just look, the

81:50

algorithms, the chips, and the data

81:53

centers are all improving or scaling at

81:56

a rate of, I don't know, 3 to 4x a year.

81:58

That's 10x every 2 years. Okay? Where

82:01

people don't understand exponential

82:02

progress is that if you're getting

82:04

better at 10x every 2 years, that

82:06

doesn't mean you'll be at 20x in four

82:08

years. It means you'll be at 100x. 100x.

82:11

So the models, the chips, and the data

82:12

centers will all be 100 times more

82:14

powerful in four years, let's say at the

82:16

end of of this presidential term. So you

82:18

multiply those things together, the

82:21

algorithms, the chips, and then the raw

82:23

compute that's available, you're talking

82:25

about a millionx increase, some of which

82:28

will be captured in price reductions,

82:30

some of it will be in the performance

82:31

ceiling, and then some of it will just

82:33

be in the overall amount of of AI

82:37

compute that's available to the economy.

82:40

But the impact of this thing is going to

82:42

be absolutely massive and I think people

82:43

still don't even appreciate that fact

82:45

because they don't understand

82:45

exponential progress. Yeah. And I think

82:47

maybe just to square the circle the the

82:49

because because everything is that that

82:52

you just said tax is what I think is

82:54

propelling the industry and then the

82:55

reality on Jama's side like like just

82:57

just to connect the dots. So uh we have

83:00

an eval test that we do uh where we run

83:03

enterprise data through every model to

83:05

to kind of figure out its accuracy rate

83:07

and and you know how much it's not even

83:09

hallucination but just literally how

83:10

much data does it miss when we ask for

83:12

facts. The best model in the world um

83:14

actually interestingly we was gro three

83:16

on on this particular test. We send it

83:18

500 documents and we ask for 40 data

83:21

fields back from the documents and so it

83:23

has to get every single data field

83:24

correct and we only do a single pass. So

83:27

we send the document to to the to the

83:29

model, we get a single pass back. Right

83:31

now the the best score is about 90%. Um

83:34

and so you can imagine a number of

83:35

industries where you can't have 90%

83:37

accuracy, you know, if you give

83:39

something, you know, a question on 40

83:41

data fields. Now there's ways to solve

83:42

it is you rerun it multiple times or you

83:45

chunk up the document into smaller parts

83:47

and so it doesn't get confused by the

83:49

large context window. But a lot of the

83:51

people that were deploying AI a year ago

83:53

or a year and a half ago weren't doing

83:55

that. And so they they you know they did

83:57

have a a kind of a a pilot run of

83:59

something and it it kind of worked okay.

84:01

And what they have to realize back to

84:03

your point sax is like this space is

84:05

literally exponentially you know

84:06

changing and so if you don't use the the

84:08

latest methods of okay you have to

84:10

actually like run the data through the

84:12

model multiple times um and you have to

84:14

chunk up the data into smaller parts and

84:16

you have to use a reasoning model and

84:17

you have to make sure that your your

84:18

prompt is like hyper tuned for the

84:20

particular use case. If you haven't done

84:21

those four things, then you probably

84:23

will actually end up with a project that

84:25

fails. And and even so, even when you do

84:27

all those things for even, you know,

84:29

harder problems, you know, you're still

84:30

going to run into issues. So, I think I

84:32

think the challenge is that everybody's

84:33

running a million miles an hour right

84:34

now and they're trying a lot of things.

84:36

Some work, some don't work at the same

84:38

time that the space is actually, you

84:40

know, you know, changing at a at a

84:41

pretty uh, you know, kind of crazy rate.

84:43

and let's take a look at our partner

84:45

Poly Market and uh which company they

84:47

think will have the best AI model by the

84:49

end of 2025 and get feedback from our

84:51

panel on if you think this is accurate

84:54

and who you would pick here. Looks like

84:57

Google is in the lead here. 41% of

85:01

people believe that they will have the

85:05

best AI model by the end of the

85:06

question. What is what is the dimension

85:08

like you know we use Gemini so for many

85:11

tasks at 8090 we use Gemini it's

85:13

incredible but for most of our codegen

85:15

we use anthropic and claude kicks ass

85:17

it's it's exceptional this is uh based

85:20

on the best scores in the chatbot arena

85:23

which just became a fourth company so

85:26

that is slightly different because

85:27

people have

85:28

gamed tests so that is a rub there

85:31

people are now building their a model

85:33

for the AVAL unit right all the models

85:34

are way overfitted for these eval so but

85:36

if you had to pick who's your I mean so

85:38

I guess Chimath you're saying you have

85:39

to take a task by task it's what depends

85:41

on task I agree what Zach said is right

85:43

so it's kind of like what what problem

85:45

are you trying to solve and then you

85:46

have to ride this technology wave that

85:48

is compounding very quickly all I was

85:50

just trying to get across is that the

85:51

error rates have been diminishing but

85:54

not nearly as fast as you need for some

85:55

sectors of the economy so you can use a

85:58

model to generate deterministic code

86:00

that's great and as long as you unit

86:02

test it and integration test it'll be

86:04

fine but I'm saying if you're going to

86:05

use a model in production

86:07

in an environment where if there are

86:10

consequences

86:12

we're not there

86:15

yet but you could use it for writing or

86:18

writing jokes or

86:20

maybe but that's that's too binary like

86:23

it's already used right now in

86:24

healthcare but it's just the doctor's

86:26

meeting notes that that would normally

86:28

take 30 minutes to go and transcribe

86:30

yeah so so it's we you can't be too too

86:32

black and white on that one yeah what's

86:33

happening right now the reason why the

86:35

the progress is so rapid in coding

86:37

assistance and I I think you know you're

86:39

right that enthropic with was it claw

86:43

3.7 37 yeah yeah I think they're the

86:45

leader and in fact I think the manis

86:47

demo that we showed it's not entirely a

86:49

rapper on clawed because they actually

86:51

they do a number of different things but

86:52

I think they are significantly using

86:54

anthropic for the code assistant part of

86:56

it in any event the reason why the

86:59

progress is so rapid with coding is

87:01

because code compiles and you can

87:04

determine objectively whether it works

87:06

or not, you can validate it. And so that

87:09

makes it a perfect area for AI to get

87:13

better at through reinforcement learning

87:14

and test time compute is AI tries a

87:16

bunch of things. It sees what works. It

87:18

sees what compiles, sees what the user

87:20

then accepts, and then be able to learn

87:22

and iterate based on that. That's why

87:24

coding right now is really the big

87:26

breakthrough application and use case.

87:28

But it's not going to be the only one. I

87:29

mean, math is another good area where I

87:31

think AI is improving rapidly again

87:33

because math you have proofs and you can

87:37

look at the results and see if it

87:39

validates. Now, I think one of the big

87:41

questions in terms of AI progress is how

87:44

extensible is the progress to other

87:46

areas that don't easily validate that

87:47

way. So, for example, legal work is I

87:52

think a really good area for AI, but how

87:54

do you validate that it's correct? You

87:56

would have to go to a court, right? the

87:57

court is the compiler like a lawsuit is

87:59

the compiler or maybe the laws or you

88:02

could hire like a thousand lawyers or

88:04

experts in an area to basically do you

88:07

know reinforcement people are doing

88:09

people people are doing but it's not

88:10

like a compiler to your point it's not

88:12

like a it's not the progress isn't going

88:13

to be as rapid because it's harder to

88:15

validate but absolutely but my guess is

88:17

that once they figure out how to nail

88:20

coding math and the things that are

88:22

easily validated they can move to the

88:23

things that are harder to validate and

88:26

but I I think this is one of the big

88:28

questions is whether I think people just

88:30

kind of assume that AI progress will be

88:32

you know equally fast in all areas and I

88:34

think it's possible that AI gets really

88:36

good in some areas better than human but

88:38

it's sort of childlike in other areas

88:40

and um narrow possible outcome is make

88:43

your point right like with a finite

88:45

answer or an answer we know is the

88:47

definitive well this is but but this is

88:49

the the important thing about the agent

88:52

kind of you know let's just say uh

88:54

framework or architecture uh you

88:56

momentum was was instead of just saying

88:59

okay we're going to do a single pass

89:00

through the model and then whatever it

89:02

comes back with is is we're going to be

89:03

satisfied um like you know the legal

89:06

work might be might be reviewed by

89:08

another agent whose job is to review

89:10

legal work and so we we can just throw

89:12

more and more compute at the problem uh

89:14

and we're just early in figuring out how

89:16

to architect those or multiple models

89:18

right you could

89:20

have you have enthropic checkp check

89:23

Gemini yeah when you have some anomaly

89:25

it spits back out to the user. So human

89:27

in the loop still matters in this type

89:28

of process. In the early days of OCR,

89:31

you would have a computer say here's the

89:33

characters in this legal document. Then

89:34

you'd have two humans type it in and

89:36

then you would get a certain level of

89:37

certainty. You'll quickly find that when

89:39

you layer these models on top of each

89:41

other, the test time compute costs are

89:43

astronomical. Y and Aaron's probably

89:46

dealt with this like it's like I get a

89:47

bill from AWS and it's like oh wait,

89:51

hold on a second. I just, you know, per

89:54

100,000 this month. what's going on. So

89:56

we have to get to the bottom. That that

89:58

by the way is another major trend line

90:00

which is that the new applications that

90:01

we talked about are all much more token

90:03

intensive. So we went from basic LLMs

90:06

totally you know which don't require

90:08

that many tokens to give you an answer

90:09

to the reasoning model where you can

90:11

spend a thousand times more tokens just

90:13

getting one answer to a question and now

90:15

the agents are going to be even more

90:17

token intensive than that. So the amount

90:19

of compute required to serve all these

90:21

new applications is going to be massive

90:23

which is why I think the capex buildout

90:25

actually makes sense when you do a deep

90:27

research to your point David you're

90:29

firing off maybe 200 queries and it's

90:31

asking them the AI is saying hey what

90:33

query should I ask on behalf of the user

90:35

and then you go down that rabbit hole

90:36

you it's basically like doing 200 of

90:38

them at once Ryan uh your thoughts here

90:40

on AI first companies and agentic

90:42

computing well the one I really wanted

90:44

to tie back to was actually our earlier

90:45

conversation on tariffs and there's a

90:47

very real use case of of LLMs is how do

90:50

you classify a product and you'd like to

90:52

get to what we see today when we did our

90:55

first machine learning based natural

90:57

language classification of a product you

90:59

take a product URL listing page a

91:01

Shopify page or Amazon page and say hey

91:03

what what classification code is this

91:05

what duty is owed six years ago in a

91:08

hackathon we got to like 70% accuracy

91:11

we're now in high 90s accuracy versus

91:14

what a human trained expert will get to

91:16

but you actually get to which is not

91:18

good enough. You know, you're wrong 3%

91:20

of the time. You know, you might have

91:22

committed a violation of the law for

91:23

sure. But actually, what is truth in

91:26

that regard? It's there's a lot of gray

91:27

area in this. And truth ultimately is

91:29

what does customs say? What does the CBP

91:32

determine is correct? And those guys are

91:35

using software, right? That's pretty

91:38

that's a very simple algorithm and it's

91:39

a decision tree that's going, okay, is

91:41

it a shoe? Yes. Is the top made of

91:43

leather? Yes. You know, is the bottom

91:45

made of rubber? and they just go through

91:47

a very simple and that outputs it. And

91:49

so on some level, if you convince the

91:51

government to use your LLM, it becomes

91:53

true whether it's true or not. I think

91:55

there's going to be some interesting

91:56

cases like that that we haven't really

91:57

thought through of like when does the

91:59

government adopt these to be the source

92:01

of truth. Okay. All of this speaks to

92:03

this thing that's going to sound totally

92:05

esoteric, but like we all used to on QA,

92:10

right? Like the least talented engineers

92:12

were allocated to QA. I think in the

92:15

world of AI it's it'll end up being the

92:17

most talented. You know we internally at

92:19

8090 we call it improvement engineering

92:22

and it's a total specialty. It's similar

92:24

to when I kind of coined the growth team

92:27

at Facebook. I feel it's the same kind

92:29

of moment where improvement engineering

92:31

is really the skill that translates toy

92:35

apps and vibe coding into something

92:37

that's very practical and real. And we

92:40

like and my team and the leader of this

92:42

team, he's like steeped in things like

92:44

Japanese kata management from like

92:46

Toyota and quality systems and these are

92:48

all the things that matter when you're

92:50

trying to just shrink the error rate

92:52

down to zero so that you can use it in a

92:54

reliable way and also to document it so

92:57

that if people want to question what

93:00

happened or have you know recompense or

93:03

some some way to come back and say hey

93:05

that that really harmed me, how do you

93:07

even do that like these are all very

93:10

complicated issues that that will get

93:12

sorted up. Super I think interesting.

93:14

Okay, four. I got to wrap guys. I got to

93:17

catch a flight to Miami. Let me do a

93:19

closing here. If you want to keep going,

93:20

you're welcome to. Three, two, the plane

93:21

just waits. Just text the pilot and just

93:23

tell them you're all right. Listen, I'm

93:25

not burning all the allin credits, so to

93:29

speak, and all of our tokens. I'm

93:30

kidding. I'm kidding. I'm kidding.

93:34

private to everything and then putting

93:35

it on the allin budget and the rest of

93:37

us are flying southwest

93:39

for

93:41

your dictator Jim Paul. I miss a flight.

93:44

It's a strange concept. Yeah, David s.

93:46

What does that mean? David, when's the

93:47

last time you flew commercial? Clinton,

93:49

I haven't missed a flight in about 15

93:51

years

93:53

for Ryan Peterson from Flexport. Aaron

93:56

Lee from the amazing box. Chim pitia.

93:59

David Saxs your chairman dictator Zar. I

94:02

am the world's greatest moderator. Love

94:04

you, boys.

94:07

We'll let your winners ride.

94:09

Rainman David S.

94:14

We open sourced it to the fans and

94:16

they've just gone crazy with it.

94:19

Queen of

94:21

[Music]

94:27

besties

94:27

[Music]

94:29

are my dog taking notice your driveways.

94:34

Oh man, my habitasher will meet up. We

94:37

should all just get a room and just have

94:39

one big huge orgy cuz they're all just

94:40

useless. It's like this like sexual

94:42

tension that we just need to release

94:43

somehow.

94:45

Wet your feet. Wet your feet. Your feet.

94:50

That's going to be good. We need to get

94:51

merch. I'm going all

94:55

[Music]

94:59

in. I'm going all in.

Interactive Summary

The podcast features a panel discussion with Ryan Peterson (Flexport CEO), Aaron Lee (Box), Chimath, and David Saxs, covering several key topics. They first discuss Trump's first 100 days, with varied assessments on border security, economic policies, and the impact of tariffs, alongside a mention of a new private club in DC. A significant portion is dedicated to the immediate and potential downstream effects of tariffs on the global supply chain, with Flexport CEO Ryan Peterson detailing declining freight bookings and strategies companies employ to mitigate costs. The conversation then shifts to the rapidly evolving field of AI agents. The panel explores their functionality, potential to transform various industries by automating tasks (and creating new work), the challenges of hallucinations and quality assurance in enterprise adoption, and the exponential advancements in AI algorithms, chips, and data center infrastructure.

Suggested questions

8 ready-made prompts