0:00

Is there a better way I think

0:02

there is? Every single interface that

0:05

I interact with, every single problem

0:07

space that I'm trying to solve,

0:09

are going to be made easier

0:11

by virtue of this new technology.

0:13

If you were starting from scratch

0:16

today, you probably wouldn't build this

0:18

app-centric world. You can imagine a

0:20

post-phone world. The past 20 years

0:22

of consumer technology have been a

0:25

story of apps, of touch screens,

0:27

and of smartphones. These form factors

0:29

seemingly appeared out of nowhere. It

0:32

may be replaced just as quickly

0:34

as they were ushered in. Perhaps

0:36

by a new AI-enabled stack, a

0:39

new computing experience that is more

0:41

agentic, more adaptive, and more immersive.

0:43

Now in today's episode, A16C's growth

0:46

general partner, David George, discusses this

0:48

feature with arguably one of the

0:51

most influential builders of this era.

0:53

That is, Menace CTO, Andrew Boz

0:55

Boz Boz Bozworth. who has spent

0:57

nearly two decades at the company,

0:59

shaping consumer interaction from the Facebook

1:01

newsfeed all the way through to

1:03

their work on smart glasses and

1:05

AR headsets. Here, Boz explores the

1:07

art of translating emerging technologies into

1:10

real products that people use and

1:12

love. Plus, how breakthroughs in AI

1:14

and hardware could turn the existing

1:16

app model on its head. In

1:18

this world, what new interfaces and

1:20

marketplaces need to be developed? What

1:22

competitive dynamics hold strong and which fall

1:25

by the wayside? For example, will brand

1:27

still be a moat? And if we

1:29

get it right, BOSS says the next

1:32

wave of consumer tech won't run on

1:34

taps and swipes, it'll run on intent.

1:36

So is the post-Mobile phone era upon

1:39

us listening to find out? Oh, and

1:41

if you do like this episode, it

1:43

comes straight from our AI revolution series.

1:46

And if you miss previous episodes of

1:48

the series with guests like AMDA CEO,

1:50

Lisa Sue, Anthropic co-founder Dario Amade

1:53

and the founders behind companies

1:55

like Databricks, Waymo, Figma, and

1:57

more head on over to a16z.com/AI

1:59

Revolution. As a reminder, the content

2:02

here is for informational purposes only,

2:04

should not be taken as legal,

2:06

business, tax, or investment advice, or

2:09

be used to evaluate any investment

2:11

or security, and is not directed

2:13

at any investors or potential investors

2:16

in any A16Z fund. Please note

2:18

that A16Z and its affiliates may

2:20

also maintain investments in the companies

2:22

discussed in this podcast. For more

2:24

details, including a link to our

2:27

investments, please see a16z.com/disclosures.

2:33

Boss, thanks for being here. Thanks

2:35

for having me. Appreciate it. Okay,

2:37

I want to jump right in. How are we

2:39

all going to be consuming content

2:41

five years from now and ten years

2:43

from now? Ten years I feel pretty confident

2:46

that we will have a lot more ways

2:48

to bring content into our viewshed than just

2:50

taking out our fun. I think August

2:52

reality glasses, obviously, are a real possibility. I'm

2:55

also hoping that we can do better for

2:57

really engaging in immersive things. Right now you

2:59

have to travel. I'm also hoping that we

3:01

can do better for really engaging in immersive

3:04

things. Right now you have to travel to

3:06

like the sphere. Which is great. There is

3:08

a real possibility. I'm also hoping that we

3:11

can do better for really engaging in immersive.

3:13

Sure, we can go and pay a lot for

3:15

tickets. Is there a better way I think there

3:17

is? So 10 years I feel really good about

3:19

all these alternative content delivery vehicles. Five

3:21

years is trickier. For example, I think the

3:24

glasses, the smart glasses, the AI glasses, the

3:26

display glasses that we'll have in five years

3:28

will be good. Some of them will be

3:30

super high-end and pretty exceptional. Some of them

3:33

will be like actually little and like not

3:35

even tremendously high resolution displays, but they will

3:37

be like always available and on your face.

3:40

I wouldn't be doing work there. But like

3:42

if I'm just trying to grab simple

3:44

content in moments between, it's pretty good

3:46

for that. So I think what we

3:48

are seeing is, as you'd expect, we're

3:50

the very beginning now of a spectrum

3:52

of super high-end, but probably very expensive

3:54

experiences that will not be evenly distributed

3:56

across the population. A much more broadly

3:58

available set of experiences. that are not

4:00

really rich enough to replace like

4:02

the devices that we have today.

4:04

And then hopefully a continually growing

4:07

number of people who are having

4:09

experiences that really could not be had

4:11

any other way today. You're thinking about

4:13

what you could do with mixed reality

4:15

and virtual reality. Yeah, we're going to

4:17

build up to a lot of that

4:19

stuff. So throughout your career, I would

4:21

say one of the observations I would

4:23

have is you've been uniquely good at

4:25

piecing together various big technology shifts into...

4:27

new product experiences. So in the

4:29

case of Facebook early days for

4:32

you, obviously you famously were part

4:34

of the team that created the

4:36

news feed, and that's a combination

4:38

of social media, a mobile experience,

4:40

and applying your like old school

4:42

AI to create to it, to

4:44

create to it, to create it,

4:46

to it, to create it. Exactly.

4:48

But that's pretty cool. And like

4:50

a lot of times these trends,

4:52

they come in bunches. And that's

4:54

what creates the breakthrough products. So

4:56

maybe take that. If there was a

4:58

thing that, not me specifically, but I think

5:00

me and my cohorts at meta were really

5:03

good at, was like, we really immersed in

5:05

what the problem was. What were people trying

5:07

to do? What did they want to do? And

5:09

when you do that, you are going to reach

5:11

for whatever tool is available to accomplish that

5:13

goal. That allows you to be really honest

5:16

about what tools are available and see trends.

5:18

I think the more oriented you

5:20

are towards the technology side. you get

5:22

caught in a wave of technology and

5:24

you don't want to admit when that

5:26

wave is over and you don't want

5:28

to embrace the next wave. And you're

5:31

building technology for technology sake. So like

5:33

solving a product problem. But if you're

5:35

embracing like what are the issues that

5:37

people are really going through in their

5:39

life and they don't have to be

5:41

profound. I bring that up just because

5:43

I think we're in this interesting moment

5:45

where a lot of people wanted a

5:47

new wave to becoming because it would

5:49

have been advantageous to them. The AI

5:51

revolution that's happening right now is it really

5:54

feels tangible. These are real problems

5:56

that are being solved and it's

5:58

not solving every problem. It creates

6:00

new problems, it's fine. So it feels like

6:02

a substantial, real, new capability that we have.

6:04

And what's unusual about it is how broad-based

6:07

it can be applied. And while it has

6:09

these interesting downsides today on factuality and certainly

6:11

compute and cost and inference, those types of

6:13

tradeoffs feel really solvable and the domains that

6:15

it applies to are really broad. And it

6:17

says that's pretty unusual. Certainly in my career.

6:20

You know, almost always when these technological breakthroughs

6:22

happen, they're almost always very domain specific. It's

6:24

like, cool, like, this is going to get

6:26

faster, or that's going to get cheaper, or

6:28

that's now possible. This kind of feels like,

6:31

oh, everything's going to get better. Every single

6:33

interface that I interact with, every single interface

6:35

that I interact with, every single problem space

6:37

that I'm trying to solve, are going to

6:39

be made easier by virtue of this new

6:42

technology. That's pretty rare. revolution in computing interfaces.

6:44

And we really started to feel 10 years

6:46

ago, like the mobile phone form factor, as

6:48

amazing as it was, this is 2015, was

6:50

like already saturated. That was what it was

6:52

going to be. And once you get past

6:55

the mobile phone, which is again, the greatest

6:57

computing device that any of us have ever

6:59

used to this point, of course, it's like,

7:01

okay, well, it has to be more natural

7:03

in terms of how you're... Getting information into

7:06

your body, which is obviously, I do usually

7:08

through our eyes and ears, and how we're

7:10

getting our intentions expressed back to the machine.

7:12

You no longer have a touchscreen, you no

7:14

longer have a keyboard. So... Once you realize,

7:17

those are the problems, it's like, cool, we

7:19

need to be on the face, because you

7:21

need to have access to eyes and ears

7:23

to bring information from the machine to the

7:25

person. And you need to have these neural

7:28

interfaces to try to allow the person to

7:30

manipulate the machine and express their intentions to

7:32

it when they don't have a keyboard or

7:34

mouse or a touch screen. And so that

7:36

has been an incredibly clear-eyed vision we've been

7:38

on for the last 10 years. But we

7:41

really did grow up in an entire generation

7:43

of engineers for whom the system was fixed.

7:45

The application model was fixed. The like interaction

7:47

design. Sure, we went. from a mouse to

7:49

a touch screen, but like it's still direct

7:52

manipulation interface, which is literally the same thing

7:54

that was pioneered in the 1960s. So like

7:56

we really haven't changed these modalities. And there's

7:58

a cost to changing those modalities because we

8:00

as a society have learned how to manipulate

8:03

these digital artifacts through these tools. So the

8:05

challenge was for us was, okay, you have

8:07

to build this hardware, which has to do

8:09

all these amazing things and also be attractive

8:11

and also be light. and also be affordable.

8:13

And none of these existed before. And what

8:16

I tell my team was like, that's only

8:18

half the problem. The other half the problem

8:20

is great, how do I use it? Like,

8:22

how do I make it feel natural to

8:24

me? I'm so good with my phone now.

8:27

It's an extension of my body, of my

8:29

intention at this point. How do we make

8:31

it even easier? Yeah. And so we were

8:33

having these challenges. And then what a wonderful

8:35

blessing. AI came in two years ago much

8:38

sooner than we expected and is a tremendous

8:40

opportunity to make this even easier for us

8:42

because the AIs that we have today are

8:44

a much greater ability to understand what my

8:46

intentions are. I can give vague reference and

8:49

it's able to like work through the corpus

8:51

of information as available to like make specific

8:53

outcomes happen from it. There's still a lot

8:55

of work to be done to actually adapt

8:57

it and it's still not yet a control

8:59

interface like I can't reliably work my machine

9:02

with it. There's a lot of things that

9:04

we have to do. But we know what

9:06

those things are. And so now you know

9:08

much more exciting place, actually. Whereas before we

9:10

thought, okay, we've got this big hill to

9:13

climb on the hardware, get this big hill

9:15

to climb on the interaction design, but we

9:17

think we can do it. And now we've

9:19

got a wonderful tailwind where on the interaction

9:21

design side, at least, there's the potential of

9:24

having this much more intelligent agent that now

9:26

has not only the ability for you to

9:28

converse with it naturally and get results out

9:30

of it, but also. to know by context

9:32

what you're seeing, what you're hearing, what's going

9:34

on around you. Yeah. And make intelligent inference

9:37

based on that information. Let's talk about like

9:39

reality labs and the suite of products, what

9:41

it is today. So you have Quest headsets.

9:43

You had the smart glasses, and then on

9:45

the far end of the spectrum is Orion

9:48

and some of the stuff that I demoed.

9:50

So just talk about the evolution of those

9:52

efforts and what you think the markets are

9:54

for them and how they converge versus not

9:56

over time. So when we started the Raybam

9:59

Meadow Project, they were going to be smart

10:01

glasses, and in fact, they were entirely built,

10:03

and we were six months away from production

10:05

when Lama 3 hit. And the team was

10:07

like, no, we got to do this. And

10:10

so now they're AI glasses, right? Like they

10:12

didn't start as AI glasses, but the form

10:14

factor was already right. We could already do

10:16

the compute. We already had the ability. So

10:18

yeah, now you have these glasses that you

10:20

can ask questions to. And in December to

10:23

the early access program, we launched, we called

10:25

Live AI. So you can start a live

10:27

AI session with your Raybam meta glasses, and

10:29

for 30 minutes until the battery runs out,

10:31

it's seeing what you're seeing what you're seeing

10:34

what you're seeing, what you're seeing, what you're

10:36

seeing, what you're seeing, The Ray-Ban meta looks

10:38

like an incremental improvement to Ray-Ban stories. And

10:40

this is kind of the story I'm trying

10:42

to tell, which is the hardware isn't that

10:45

different between the two, but the interactions that

10:47

we enable with the person using it are

10:49

so much richer now. When you use Orion,

10:51

you use the full AR glasses, you can

10:53

imagine a post-phone world. You're like, oh, wow,

10:55

like if this was attractive enough and light

10:58

enough. and added battery life enough to wear

11:00

all day, this would have all the stuff

11:02

I need. Like it would all be right

11:04

here. And when you start to combine that

11:06

with images that we have of what AI

11:09

is capable of, see you did the demo

11:11

where we showed you the breakfast. Yeah, I

11:13

did. And it's, yeah, and for what it's

11:15

worth, I'll explain it because it's very cool.

11:17

Got to walk over and there's a bunch

11:20

of breakfast ingredients laid out and I look

11:22

at it and I say, hey meta, what

11:24

are some recipes, what are some recipes, AI

11:26

component when we first thought about it. It

11:28

had this component that was very direct manipulation.

11:30

So it was very much the model on

11:33

the app model that we're all from that

11:35

within the course. And I think there's a

11:37

version of that. Yeah, of course, you're going

11:39

to want to do calls and you're going

11:41

to do your email and be able to

11:44

do your texting and you want to play

11:46

games. We have to play our Star Gazer

11:48

game and do your Instagram. What we're now

11:50

excited about is, okay, take all those pieces

11:52

and layer on the ability to have an

11:55

interactive assistant that really understands not just what's

11:57

happening on your device and what email is

11:59

coming in. But also what's happening in. Yeah,

12:01

of course. But also what's happening in the

12:03

physical world around you and is able to

12:06

connect what you need in the moment with

12:08

what's happening. And so these are concepts where

12:10

you're like, wow, what if the entire app

12:12

model is upside down? The device realizes that

12:14

you have a moment between meetings, you're a

12:16

little bit bored. Hey, do you want to

12:19

catch up on the latest highlights from your

12:21

favorite basketball team? Like those things become possible.

12:23

Having said that, the hardware problems are hard

12:25

and they're real and the cost problems are

12:27

hard and they're real. And you come for

12:30

the king, you best not miss. The phone

12:32

is an incredible centerpiece of our lives today.

12:34

It's how I operate my home. I use

12:36

my car, I use it for work. It's

12:38

everywhere, right. It's everywhere, right, right, right, right.

12:41

And... the world has adapted itself to the

12:43

phone. So it's weird that my ice maker

12:45

has a phone app, but it does. Like,

12:47

I don't know. I'm not sure. Seemed excessive,

12:49

but like, somebody today is like, I gotta

12:51

make an ice maker, number one job, gotta

12:54

have it out. It's like, the smart refrigerator,

12:56

you're like, you're like, I don't need this,

12:58

it's like, take it on me. I do

13:00

think it's going to be a 10 year

13:02

view is harder because, man, like, like. Even

13:05

if knocking out, it's dominant into the phone

13:07

in five years. It just seems so hard.

13:09

I'm thinking, it's like unthinkable for us, right?

13:11

That's what I said, like, Orion was the

13:13

first time I thought, me. Orion, like, put

13:16

it on my head, I was like, okay.

13:18

I was like, okay. I was like, okay,

13:20

like, it could happen, like, there does exist

13:22

a life for us, I've had into that.

13:24

I was like, like, okay, like, like, like,

13:27

like, it could, it could, it could, it

13:29

could, yeah, yeah, yeah, like, like, it could,

13:31

it could, it could, it could, yeah, it

13:33

could, it could, yeah, it could, yeah, yeah,

13:35

it could, it could, yeah, it could, it

13:37

could, yeah, it could, it could, yeah, it

13:40

could, yeah, it could, it could, it could,

13:42

it could, it could, it could, it could,

13:44

it could, So maybe you get to the

13:46

point where the hardware is capable, it is

13:48

market accessible, but do you tether to the

13:51

phone? Do you take a strong view that

13:53

you will never do that and let the

13:55

product stand? Like how do you think about

13:57

that piece? The phones have this huge advantage

13:59

and disadvantage. Huge advantage, which is like the

14:02

phone is already central to our lives. It's

14:04

already got this huge developer ecosystem. It's its

14:06

anchor device. And it's a wonderful anchor device

14:08

for that. The disadvantages, I actually think what

14:10

we found is the apps want to be

14:12

different when they're not controlled via touch screen.

14:15

And that's not super novel. A lot of

14:17

people failed early in mobile, including us, by

14:19

just taking our web stuff and putting on

14:21

the mobile phone, and be like, oh, the

14:23

mobile phone, we'll just put the web there.

14:26

But because it wasn't native to what the

14:28

phone was, and I mean, everything from interaction

14:30

design to the actual design, to the layout,

14:32

to how it felt, like, because we weren't

14:34

doing phone native things. we were failing with

14:37

one of the most popular products in the

14:39

history of the web. This is like the

14:41

major design field, like the scumorphic idea versus

14:43

the native idea. Yeah, and I think having

14:45

the developers is a true value and I

14:48

think having all this application functionality is a

14:50

true value, but then once you actually project

14:52

it into space and you're manipulating it with

14:54

your fingers like this as opposed to a

14:56

touch screen, you have much less precision, it

14:58

doesn't respond to voice commands because there's no...

15:01

tools for that. It was no design integration

15:03

with that. So having a phone platform today

15:05

feels like, wow, I've got this huge base

15:07

to work from on the hardware side, but

15:09

I've also actually got this kind of huge

15:12

anchor to drag on the software side. And

15:14

so we're not opposed to these partnerships. And

15:16

I think it'll be interesting to see once

15:18

the hardware is a little bit more developed

15:20

how partners feel about it. And I hope

15:23

they continue to support people who buy these

15:25

phones for $1,200 dollars, $1, $1,200,000. The biggest

15:27

question I have is whether the entire app

15:29

model, we were imagining a very phone-like app

15:31

model for these devices, admittedly a very different

15:33

interaction design, input, and control schemes are very

15:36

different, and that demands like a little extra

15:38

developer tension. I am wondering if like the

15:40

progression of AI over the next several years

15:42

doesn't turn the app model in its head.

15:44

Right now it's a common unusual... thing where

15:47

I'm like I want to play music. So

15:49

in my head I translate that to I

15:51

have to go open Spotify or open title

15:53

and the first thing I think of is

15:55

who is my provider going to be? Yeah

15:58

of course. As opposed like that's not what

16:00

I want. What's extremely limiting? What I want

16:02

is to play music. Yes. And I just

16:04

want to be like go to the AI

16:06

and like cool play this music for me.

16:09

Yeah. And it should know. Oh like you're

16:11

already using this service. We'll use this service.

16:13

Song or this one has lower latency whatever

16:15

thing is or it's like hey the song

16:17

you want isn't available on any of these

16:19

services Do you want to sign up for

16:22

this other service that does have the song

16:24

that you want? I don't want to have

16:26

to be responsible for orchestrating like what app

16:28

I'm opening to do a thing We've had

16:30

to do that because that's how things were

16:33

done in the entire history of digital computing

16:35

you have an application based model that was

16:37

the system So I do wonder how much

16:39

AI inverts things. That's a pretty hot take.

16:41

Yeah, it's a hot take. Inverts things. And

16:44

that's not about wearables. That's not about anything.

16:46

That's just like, even at the phone level,

16:48

if you were building a phone today, would

16:50

you build an app store the way you

16:52

historically built an app store? Or would you

16:54

say like, hey, you as a consumer, express

16:57

your intention, express what you're trying to accomplish?

16:59

And let's like see what we have. you

17:01

probably wouldn't build this like, absentric world where

17:03

I, as a consumer, I'm trying to solve

17:05

a problem and first have to decide which

17:08

of the providers I'm going to use to

17:10

solve that problem. Yeah, of course. That's fascinating.

17:12

And again, I think it's a function of

17:14

where the capabilities are today and I think

17:16

where we have line of sight into orchestration

17:19

capabilities. And then, of course, you get a...

17:21

build the developer ecosystem to develop on the

17:23

platform. Which is incredibly hard. That's the thing

17:25

I want to see. That's the hardest piece.

17:27

Yeah. The stronger we get at agentic reasoning

17:30

and capabilities, the more I can rely on

17:32

my AI to do things in my absence.

17:34

And at first it will be knowledge work,

17:36

of course. That's fine. But once you have

17:38

a flow of consumers coming. through here, what

17:40

you're going to find is that they're going

17:43

to have a bunch of dead ends. Yeah.

17:45

Where they're going to ask the AI, hey,

17:47

can you do this thing for me? And

17:49

it's going to say, no, I can't. That's

17:51

the gold mine that you take to developers.

17:54

And you're like, hey, I've got 100,000 people

17:56

a day, use your app, use your app,

17:58

yeah, trying to use your app, look, here's

18:00

what's what's coming through, here, and you're, and

18:02

you're like, I've got, I've got a, I've

18:05

got a, I've got a, I've got a,

18:07

I've got a, a, a, a, a, a,

18:09

a, a, a, a, a, a, a, a,

18:11

a, a, a, a, a, a, a, a,

18:13

a, a, a, a, Right. I'd go back

18:15

and say, hey, you gotta pay for this.

18:18

There's a guy who does this for you,

18:20

but you gotta pay for it. Yeah. And

18:22

I, by the way, I'm not just talking

18:24

about apps. I'm like, it's a plumber. There's

18:26

something of a marketplace here that I think

18:29

emerges over time. So that's how I see

18:31

it playing out. I don't see it playing

18:33

out. I don't see it playing out as

18:35

like someone goes. repeatedly in certain areas because

18:37

that's a type of functionality that is currently

18:40

behind some kind of an app wall and

18:42

there's no There's no word hasn't been built

18:44

native to whatever consumption mechanism. There's no bridge.

18:46

Yeah. Yeah. And everyone wants to build the

18:48

bridges like no it's gonna manipulate the pixels

18:51

and it's gonna manipulate it's like fine it

18:53

can do those things. I'm not saying the

18:55

AI can't cross those boundaries. But I think

18:57

over time that becomes the primary interface for

18:59

humans interacting with software as opposed to the

19:01

like pick from the garden of applications. Yeah,

19:04

that makes a ton of sense. That's a

19:06

very alluring in state, just as a consumer,

19:08

right? Yeah, it's messy. And I think it

19:10

creates these very exciting marketplaces for functionality inside

19:12

the AI. It abstracts away a lot of

19:15

companies' brand names, which I think is going

19:17

to be very hard for an entire generation

19:19

of brands. Yeah. Like the fact that I

19:21

don't care if it's being played on one

19:23

of these two music services. That's hard for

19:26

those music services who like really want me

19:28

to care. Yeah. And like they want me

19:30

to have a stronger opinion about it. And

19:32

like they want me to have an attachment.

19:34

Yeah. I don't want to have an attachment.

19:36

There are some things where you may value

19:39

the attachment. In the world where I'm like,

19:41

here's an app garden, and these two are

19:43

competing for my eyeballs, the brand that they've

19:45

built is the hugely valuable asset. In the

19:47

world where I just care if the song

19:50

gets played and the song gets played and

19:52

sounds good, a different set of priorities are

19:54

important. I think that's net positive because what

19:56

matters now is performance on the job being

19:58

asked. Well abstracting away, that's like effectively articulating

20:01

abstracting away margin pools, which puts a lot

20:03

more pressure on us trusting the AI or

20:05

the distributor of the AI. And so far

20:07

as I'm floating between different companies that are

20:09

each providing AIs, the degree which I trust

20:12

them to not be bought and paid for

20:14

in the back end. they're not giving me

20:16

the best experience or the best price for

20:18

money. They're giving the one that gives them

20:20

the most money. Yeah, of course. So yeah,

20:22

it's the experience of people's search today, right?

20:25

It's a very different world. It's a very

20:27

different world. But you can actually see inklings

20:29

of it today, right? So certain companies are

20:31

willing to work with the new AI providers

20:33

in agientic task completion. Yeah. And then they're

20:36

like, well, actually, wait a minute, wait a

20:38

minute. that I have this brand relationship directly

20:40

with the demand side. So that's potentially messy,

20:42

but a bright future, especially if we don't

20:44

have to pay that like brand tax. Yeah,

20:47

it'll be very messy. I don't know it's

20:49

avoidable, because I think once consumers start to

20:51

get into these tight loops where more and

20:53

more of their interactions are being moderated by

20:55

an AI. you won't have a choice. That's

20:57

like where your customers will be. Yeah. But

21:00

it's going to be a pretty different world.

21:02

Yeah, it'll be a different world and there

21:04

will probably be some groups that try to

21:06

move fast to it as a way to

21:08

compete with things that are branded. Yeah. And

21:11

just say I'm going to compete on performance

21:13

and price. Yeah, that's right. Where do you

21:15

think that could potentially happen first? It probably

21:17

will mirror query volume. web era when Google

21:19

became the dominant search engine. So before that,

21:22

the web era was like very index-based. It

21:24

was like Yahoo and it was like links

21:26

and getting major sources of traffic to link

21:28

to you was the game. And then once

21:30

Google came to dominance, which happened very quickly

21:33

over maybe a couple of years, I feel

21:35

like. All that mattered was like SEO. All

21:37

that mattered was like where you were in

21:39

the query stream. And the query stream dictated.

21:41

what businesses came over and succeeded? Yeah. Because

21:43

like the queries that were the most frequent,

21:46

those were the ones that came first. Yeah.

21:48

And so like I travel travel travel travel

21:50

travel travel is the one that travel came

21:52

right away. Right. Like it is a huge

21:54

disruption and travel agents went from a thing

21:57

that existed. Do I think that didn't exist

21:59

in a relatively short? And they all created

22:01

on the basis of like execution of the

22:03

best deal. It was literally like seamless fashion

22:05

with the highest conversion. I think SEO has

22:08

gotten to a point now. Where it's kind

22:10

of a bummer, it's like made things worse.

22:12

Like everyone's got so good. This is like

22:14

game. Everyone's gotten so good at it. Especially

22:16

with AI, actually. That's right. So I actually

22:18

think it's like we had this incredible flattening

22:21

curve. Now it's like starting to kind of

22:23

rise up in terms of. Especially with paid

22:25

placement too. Yeah, that's so dominant. Yeah, that's

22:27

right. And this is like probably the cautionary

22:29

tale for how this plays out in AIs

22:32

as well. because those are the queries that

22:34

are, that's the volume of people unsatisfied with

22:36

the existing solutions that they have. Yeah. Otherwise

22:38

they wouldn't be asking about it. And product

22:40

providers and developers will follow that and build

22:43

specifically to solve those problems. That's right. Once

22:45

it tips in each vertical, we get a

22:47

lot of progress very quickly. Yeah. Towards better

22:49

solutions for consumers. And then once it's a

22:51

steady state. it starts to be gamesmanship. Yeah,

22:54

and that's something to fight. And that's decaying

22:56

or... That'll be the true test of it.

22:58

The true test. Can it get through that?

23:00

Can it avoid falling into that. Can it

23:02

avoid that travel? Yeah, yeah, that's right. Well,

23:04

a lot of that is business model driven

23:07

and we'll see how that evolves over time

23:09

too. That's right. You guys have also been

23:11

leading from the front on this... idea of

23:13

open source. Yeah. And so talk about some

23:15

of your efforts on that side of the

23:18

business and then what is the ideal market

23:20

structure of the AI model side for you

23:22

guys? There's two parts that came together. The

23:24

first one is Lama came out of fair,

23:26

our fundamental AI research group, and that's been

23:29

an open source research group since the beginning.

23:31

You know, since Jean Lacoon came in and

23:33

they established that. It allowed us to attract

23:35

incredible researchers who really believe that we're going

23:37

to make more progress as a society working

23:39

together across boundaries of individual labs than not.

23:42

And to be fair, it's not just us,

23:44

obviously. the transformer paper was published at Google.

23:46

And like, you know, big, we self-supervised learning

23:48

was our contribution. Like, everyone's contributing to the

23:50

knowledge base. But when we open source Lama,

23:53

that's how all models were open source of

23:55

that. Like, of course, like, everyone's open. The

23:57

only thing that was unusual, everything else just

23:59

went close source over time, effectively. That's right.

24:01

But before that. Every time someone built a

24:04

model, they open-source it so that other people

24:06

could use the model and see how great

24:08

that model was. I was like, mostly how

24:10

it was done. If it was worth anything.

24:12

There's certainly some specialized models for translations and

24:15

whatnot where kept closed. But like, if it

24:17

was a general model, that was what was

24:19

done. Lama 2 was probably the big decision

24:21

point for us. Lama 2, and this is

24:23

where I think the second thing that came

24:25

into some belief that I've had that I

24:28

was advancing really continuously internally, that Mark really

24:30

believes in, believes in, believes in, and he

24:32

believes in, and he's, and he's, and he's,

24:34

and he's, and he's, and he's, and he's,

24:36

and he's, he's, he's, he's, he's, he's, he's,

24:39

he's, he's, he's, he's, he's, he's, he's, he's,

24:41

he's, he's, he's, he's, he's, he's, Which is

24:43

first of all, we're going to make way

24:45

more progress if these models are open. Yeah.

24:47

Because a lot of these contributions aren't going

24:50

to come from these big labs. Like they're

24:52

going to come from these little labs. We've

24:54

seen this already with Deep Seek in China,

24:56

which was put in a tough spot and

24:58

then innovated incredibly in the memory architecture and

25:00

a couple other places to really get amazing

25:03

results. And so we really believe we're going

25:05

to get the most progress collectively. The second

25:07

thing is, inside this piece, is, you know,

25:09

you know, you know, this is a classic,

25:11

I believe this is you're going to be

25:14

going to make going to make commodities, you're

25:16

going to make commodities, you're going to make

25:18

commodities, And you want to commoditize your compliments.

25:20

Yes. And we're in a unique position strategically

25:22

where our products are made better through AI,

25:25

which is why we go investing it for

25:27

so long. Whether it's recommendation systems in what

25:29

you're seeing in feed, reals, whether it's simple

25:31

things like what friend do I put at

25:33

the top when you type you want to

25:36

make a new message? Who do I think

25:38

you're going to message right now? Yeah. that

25:40

to really big expansive things like, hey, here's

25:42

an entire answer, here's an entire search interface

25:44

that we couldn't do it for in what's

25:46

out. Yeah, yeah, yeah. That like now is

25:49

a super popular surface. Yeah. So there's all

25:51

these things that are possible for us that

25:53

are made better by this AI, but nobody

25:55

else having this AI can then build our

25:57

product. The asymmetry works in our favorite. Yeah,

26:00

of course. And so for us, like commoditizing

26:02

your compliments is just good business sense and

26:04

making. helps a bunch of small startups and

26:06

academic labs. It also helps us. Yeah, so

26:08

we're the application provider. So we're all super

26:11

aligned on a business model alignment and industry

26:13

alignment there. So it comes from both this

26:15

fundamental belief in how this kind of research

26:17

should be done and then aligns with the

26:19

business model and so there's no conflict. Yeah,

26:21

societal progress plus business model alignment. It's all

26:24

together. It's all going the same direction. That's

26:26

great. I want to shift gears to talking

26:28

about the impediments to progress and like what

26:30

you think. or kind of linear versus not.

26:32

So the risks to the vision, to the

26:35

overall vision that you articulated, obviously hardware, AI

26:37

capabilities, vision capabilities, and screens and all that,

26:39

resolutions. We talked about the ecosystem and developers

26:41

and native products. So maybe just talk about

26:43

what you see or kind of the linear

26:46

path things and the things that may be

26:48

harder or riskier. We have real invention risk.

26:50

There exists risk that the things that we

26:52

want to build. We don't have the capacity

26:54

to build as a society as a species

26:57

yet. Yeah, and that's not a guarantee I

26:59

think we have Windows to us you've seen

27:01

Orion so like it can be done Yeah,

27:03

there's but yeah, it feels like it's a

27:05

cost reduction exercise It's materials improvement exercise, but

27:07

it can be done There is still some

27:10

invention risk far bigger than the invention risk.

27:12

I think is the adoption risk is it

27:14

considered socially acceptable? Are people willing to learn

27:16

a new modality? Like we all learned a

27:18

type when we were kids at this point.

27:21

We were born with phones in our hands

27:23

at this point. Are people willing to learn

27:25

a new modality? Is it worth it to

27:27

them? Ecosystemrous. Even bigger than that. Like, great.

27:29

build this thing, but if it just does

27:32

like your email and reals, that's probably not

27:34

enough, do people bring the suite of software

27:36

that we require to interact with modern human

27:38

society to bear on the device. Those are

27:40

all huge risks. I will say we feel

27:42

pretty good about where we're getting on the

27:45

hardware on acceptability. We think we can do

27:47

those things. That was not a guarantee before.

27:49

I think with the Rayvan meta glasses, we're

27:51

feeling like I'm using technology. Within that. super

27:53

interesting regulatory challenges. Here I have an always

27:56

on machine that gives me superhuman sensing. My

27:58

vision is better. My hearing is better. My

28:00

memory is better. That means when I see

28:02

you a couple years from now, and I

28:04

haven't seen you in the internet, I'm like,

28:07

oh God, I don't remember. We did a

28:09

podcast together. What's the guy's name? Can I

28:11

ask that question? Am I allowed to ask

28:13

that question? What is your right? Your face.

28:15

You showed me your face. You showed me.

28:18

You showed me your face. You showed me

28:20

your face. You showed me your face. You

28:22

showed me your face. You showed me your

28:24

face. You showed me your face. You showed

28:26

me your face. But I don't have a

28:28

great memory, so am I allowed to use

28:31

a tool to assist me or not? So

28:33

there's really subtle regulatory, privacy, social, acceptability questions

28:35

that are like embedded here that are super

28:37

deep individually and can derail the whole thing.

28:39

Like you can easily derail the whole thing

28:42

in slow progress. Yeah, that's the things I

28:44

think we sometimes think. in our industry, it's

28:46

like feel the dreams, if you build it,

28:48

they will come. And it's like, no, a

28:50

lot of things have to happen right. Well,

28:53

you can also, we're stuck, totally. That's the

28:55

whole, that's the risk is, you're sure you

28:57

can get your hands locked. Right technology can

28:59

get derailed for a long periods of time.

29:01

Nuclear power, nuclear power, nuclear power, got derail.

29:03

Yeah, nuclear power, it, it's, can get derailed,

29:06

for long periods of time. And they just,

29:08

like they just, like they just, like, like,

29:10

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

29:12

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

29:14

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

29:17

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

29:19

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

29:21

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

29:23

is looking better than it has been, but

29:25

I think there's still a lot of big

29:28

hedges across there. I actually think the ecosystemers

29:30

was one I would have said previously was

29:32

the biggest one, but AI. is now my

29:34

potential silver bullet there. If AI becomes the

29:36

major interface, then it comes for free. And

29:39

I will also say that we've had such

29:41

a positive response from even just set aside

29:43

Orion, even the Raybam Meadows, companies that want

29:45

to work for us. and building that platform.

29:47

It's not a platform yet. We're so little

29:49

competing. There's so little computing. We just connect

29:52

an app. We literally don't have any space

29:54

yet. But we did do a partnership with

29:56

Be My Eyes, which like helps blind and

29:58

harder vision people navigate. And it's really spectacular.

30:00

And so there's a little window there where

30:03

we can start building. So yeah, the response

30:05

has been more positive than I had expected.

30:07

So everything right now, tailwinds abound. And to

30:09

be honest after eight years after eight years

30:11

of eight years of headwinds. Having a year

30:14

of tailwinds is nice. Yeah, I'll take it.

30:16

I'll take it. I'm not going to look

30:18

at it in the face. No victory left.

30:20

Yeah, but that's good. Okay. But it's all

30:22

hard. Yeah, yeah. At every point it could

30:24

all fail. Yeah, I like these uses start

30:27

with it's invention risk. It's I don't know.

30:29

There's like, there's many ways this just won't

30:31

work. Yeah, that's like, there's many ways this

30:33

just won't work. This just won't work. This

30:35

just won't work. This just won't work. We're

30:38

true believers. We're true believers. We're true believers.

30:40

We're true believers. We're true believers. Like we're

30:42

true believers. Like we're true believers. Like we're

30:44

true believers. Like we're true believers. Like we're

30:46

true believers. Like we're true believers. Like we're

30:49

true believers. Like we're true believers. It needs

30:51

to happen and it doesn't happen for free.

30:53

We can be the ones to do it.

30:55

Our chief scientist, Michael A. Rash, who's one

30:57

of my favorite people I've ever gotten a

31:00

chance to work with, he talks a lot

31:02

about the myth of technological eventualism. It doesn't

31:04

eventually happen. There's a lot of people in

31:06

tech. Like, yeah, AR eventually happened. That's not

31:08

a what... fucking works now. That would actually,

31:10

you have to, you have, or it's a

31:13

specific one that would just absolutely not. You

31:15

have to stop and put the money and

31:17

do it. Somebody has to stop and like

31:19

do it. And that is the difference. The

31:21

number one thing I say is like the

31:24

difference between us and anybody else is we

31:26

believe in this stuff in our course. This

31:28

is the most important work I'll ever get

31:30

a chance to do. This is Xerox Park

31:32

level. new stuff where we're rethinking how humans

31:35

are going to interact with computers. It's like

31:37

JCR, Licklider, and the human computing. We're seeing

31:39

that with AI. It's a rare moment. It's

31:41

a rare moment. It doesn't even happen once

31:43

a generation, I think. It may happen every

31:45

other generation, every third generation. you

31:48

don't get a chance

31:50

to do this all

31:52

the time. this all the time.

31:54

not missing it. We're it.

31:56

We're to do it

31:59

and we may fail.

32:01

Like it's possible, like it's possible.

32:03

not fail for lack

32:05

of effort or belief.

32:07

effort or belief. Great. ton, boss.

32:10

Nice cheers. boss. cheers. cheers. Yeah,

32:12

cheers.