AI transcript
0:00:02 Is there a better way? I think there is.
0:00:05 Every single interface that I interact with,
0:00:08 every single problem space that I’m trying to solve
0:00:12 are going to be made easier by virtue of this new technology.
0:00:14 If you were starting from scratch today,
0:00:17 you probably wouldn’t build this app-centric world.
0:00:21 You can imagine a post-phone world.
0:00:24 The past 20 years of consumer technology
0:00:27 have been a story of apps, of touchscreens, and of smartphones.
0:00:31 These form factors seemingly appeared out of nowhere
0:00:34 and may be replaced just as quickly as they were ushered in,
0:00:37 perhaps by a new AI-enabled stack,
0:00:40 a new computing experience that is more agentic,
0:00:42 more adaptive, and more immersive.
0:00:46 Now, in today’s episode, A16C’s growth general partner,
0:00:48 David George, discusses this feature
0:00:52 with arguably one of the most influential builders of this era.
0:00:56 That is Meta CTO, Andrew Boz Bosworth,
0:00:58 who spent nearly two decades at the company,
0:01:01 shaping consumer interaction from the Facebook newsfeed
0:01:05 all the way through to their work on smart glasses and AR headsets.
0:01:09 Here, Boz explores the art of translating emerging technologies
0:01:12 into real products that people use and love,
0:01:14 plus how breakthroughs in AI and hardware
0:01:17 could turn the existing app model on its head.
0:01:22 In this world, what new interfaces and marketplaces need to be developed?
0:01:24 What competitive dynamics hold strong?
0:01:26 And which fall by the wayside?
0:01:28 For example, will brands still be a moat?
0:01:31 And if we get it right, Boz says,
0:01:34 the next wave of consumer tech won’t run on taps and swipes,
0:01:36 it’ll run on intent.
0:01:39 So, is the post-mobile phone era upon us?
0:01:40 Listen in to find out.
0:01:43 Oh, and if you do like this episode,
0:01:45 it comes straight from our AI Revolution series.
0:01:48 And if you missed previous episodes of this series
0:01:50 with guests like AMD CEO Lisa Su,
0:01:52 Anthropic co-founder Dario Amadei,
0:01:55 and the founders behind companies like Databricks,
0:01:56 Waymo, Figma, and more,
0:02:00 head on over to a16z.com slash AI Revolution.
0:02:06 As a reminder, the content here is for informational purposes only,
0:02:09 should not be taken as legal, business, tax, or investment advice,
0:02:11 or be used to evaluate any investment or security,
0:02:14 and is not directed at any investors or potential investors
0:02:15 in any A16Z fund.
0:02:18 Please note that A16Z and its affiliates
0:02:21 may also maintain investments in the companies discussed in this podcast.
0:02:24 For more details, including a link to our investments,
0:02:27 please see a16z.com slash disclosures.
0:02:34 Boz, thanks for being here.
0:02:35 Thanks for having me.
0:02:36 Appreciate it.
0:02:38 Okay, I want to jump right in.
0:02:42 How are we all going to be consuming content
0:02:44 five years from now and ten years from now?
0:02:45 Ten years, I feel pretty confident
0:02:49 that we will have a lot more ways to bring content into our view shed
0:02:50 than just taking out our fun.
0:02:54 I think augmented reality glasses, obviously, are a real possibility.
0:02:55 I’m also hoping that we can do better
0:02:58 for really engaging in immersive things.
0:03:00 Right now, you have to travel to, like, the sphere,
0:03:02 which is great, but there’s one of them.
0:03:03 It’s in Vegas since a trip.
0:03:05 Are there better ways that we can have access to
0:03:07 if we really want to be engaged in something,
0:03:09 not just immersively, but also socially?
0:03:10 So it’s like, oh, I want to watch the game.
0:03:11 I want to watch it with my dad.
0:03:12 I want to feel like we’re courtside.
0:03:15 Sure, we can go and pay a lot for tickets.
0:03:16 Is there a better way?
0:03:16 I think there is.
0:03:18 So ten years, I feel really good
0:03:20 about all these alternative content delivery vehicles.
0:03:21 Five years is trickier.
0:03:24 For example, I think the glasses, the smart glasses,
0:03:26 the AI glasses, the display glasses
0:03:28 that we’ll have in five years will be good.
0:03:31 Some of them will be super high-end
0:03:32 and pretty exceptional.
0:03:34 Some of them will be, like, actually little
0:03:37 and, like, not even tremendously high-resolution displays,
0:03:38 but they will be, like, always available
0:03:39 and on your face.
0:03:41 I wouldn’t be doing work there,
0:03:44 but, like, if I’m just trying to grab simple content
0:03:45 in moments between, it’s pretty good for that.
0:03:47 So I think what we are seeing is,
0:03:50 as you’d expect, we’re at the very beginning now
0:03:53 of a spectrum of super high-end
0:03:54 but probably very expensive experiences
0:03:57 that will not be evenly distributed across the population.
0:03:57 Yeah.
0:03:59 A much more broadly available set of experiences
0:04:01 that are, they’re not really rich enough
0:04:03 to replace, like, the devices that we have today.
0:04:05 And then hopefully a continually growing number
0:04:07 of people who are having experiences
0:04:10 that really could not be had any other way today.
0:04:12 You know, thinking about what you could do
0:04:13 with mixed reality and virtual reality.
0:04:13 Yeah.
0:04:15 We’re going to build up to a lot of that stuff.
0:04:17 So throughout your career,
0:04:20 I would say one of the observations I would have
0:04:22 is you’ve been uniquely good
0:04:26 at piecing together various big technology shifts
0:04:28 into new product experiences.
0:04:31 So in the case of Facebook, early days for you,
0:04:34 obviously you famously were part of the team
0:04:35 that created the news feed.
0:04:35 Yeah.
0:04:37 And that’s a combination of social media,
0:04:38 a mobile experience,
0:04:41 and applying your, like, old-school AI.
0:04:42 Yeah, that’s right.
0:04:42 To it.
0:04:43 The old-school AI.
0:04:43 Yeah, exactly.
0:04:44 But that’s pretty cool.
0:04:46 And, like, a lot of times these trends,
0:04:47 they come in bunches.
0:04:47 Yeah.
0:04:49 And that’s what creates the breakthrough products.
0:04:53 So maybe take that and apply it to where we are today
0:04:55 with the major trends that are in front of you.
0:04:56 Let me say two things about this.
0:04:58 The first one is I think if there was a thing that,
0:04:59 not me specifically,
0:05:02 but I think me and my cohorts at Meta were really good at,
0:05:04 was, like, we really immersed in, like, what the problem was.
0:05:05 Like, what were people trying to do?
0:05:06 What did they want to do?
0:05:08 And when you do that,
0:05:12 you are going to reach for whatever tool is available
0:05:13 to accomplish that goal.
0:05:15 That allows you to be really honest about
0:05:18 what tools are available and see trends.
0:05:21 I think the more oriented you are towards the technology side,
0:05:24 you get caught in a wave of technology,
0:05:25 and you don’t want to admit when that wave is over
0:05:27 and you don’t want to embrace the next wave.
0:05:29 And you’re building technology for technology’s sake.
0:05:29 Yeah, yeah.
0:05:31 So, like, solving a product problem.
0:05:31 But if you’re embracing, like,
0:05:34 what are the issues that people are really going through in their life
0:05:35 and they don’t have to be profound,
0:05:37 I bring that up just because I think we’re in this interesting moment
0:05:40 where I think all of us have been through a phase
0:05:43 where a lot of people wanted a new wave to be coming
0:05:45 because it would have been advantageous to them.
0:05:45 Yeah.
0:05:48 But those things weren’t solving problems that regular people had.
0:05:50 I think the reason we’re so enthusiastic about
0:05:53 the AI revolution that’s happening right now
0:05:56 is it really feels tangible.
0:05:58 These are real problems that are being solved.
0:05:59 And it’s not solving every problem.
0:06:00 It creates new problems.
0:06:01 It’s fine.
0:06:04 So it feels like a substantial real NUCA capability that we have.
0:06:09 And what’s unusual about it is how broad-based it can be applied.
0:06:12 And while it has these interesting downsides today on factuality
0:06:15 and certainly compute in cost and inference,
0:06:18 those types of trade-offs feel really solvable
0:06:20 and the domains that it applies to are really broad.
0:06:22 And that’s very unusual.
0:06:23 Certainly in my career,
0:06:25 you almost always, when these technological breakthroughs happen,
0:06:27 they’re almost always very domain-specific.
0:06:28 It’s like, cool, this is going to get faster
0:06:31 or that’s going to get cheaper or that’s now possible.
0:06:33 This kind of feels like, oh, everything’s going to get better.
0:06:36 Every single interface that I interact with,
0:06:40 every single problem space that I’m trying to solve
0:06:43 are going to be made easier by virtue of this new technology.
0:06:44 That’s pretty rare.
0:06:47 Mark and I always believed that this AI revolution was coming.
0:06:48 We just thought it was going to take longer.
0:06:48 Yeah.
0:06:50 We thought we were probably still 10 years away at this point.
0:06:51 Yeah.
0:06:52 But what we thought would happen sooner
0:06:54 was this revolution in computing interfaces.
0:06:59 And we really started to feel 10 years ago
0:07:02 like the mobile phone form factor, as amazing as it was,
0:07:05 this is 2015, was like already saturated.
0:07:06 That was what it was going to be.
0:07:09 And once you get past the mobile phone,
0:07:10 which is, again, the greatest computing device
0:07:12 that any of us have ever used to this point,
0:07:14 of course, it’s like, okay, well, it has to be more natural
0:07:18 in terms of how you’re getting information into your body,
0:07:20 which is obviously ideally usually through our eyes and ears,
0:07:24 and how we’re getting our intentions expressed back to the machine.
0:07:25 You no longer have a touchscreen.
0:07:26 You no longer have a keyboard.
0:07:29 So once you like realize those are the problems,
0:07:31 it’s like, cool, we need to be on the face
0:07:33 because you need to have access to eyes and ears
0:07:35 to bring information from the machine to the person.
0:07:37 And you need to have these neural interfaces
0:07:40 to try to allow the person to manipulate the machine
0:07:41 and express their intentions to it
0:07:43 when they don’t have a keyboard or mouse or a touchscreen.
0:07:47 And so that has been an incredibly clear-eyed vision
0:07:49 we’ve been on for the last 10 years.
0:07:53 But we really did grow up in an entire generation of engineers
0:07:56 for whom the system was fixed.
0:07:57 The application model was fixed.
0:07:59 The interaction design.
0:08:02 Sure, we went from a mouse to a touchscreen,
0:08:04 but it’s still a direct manipulation interface,
0:08:06 which is literally the same thing that was pioneered in the 1960s.
0:08:09 So we really haven’t changed these modalities.
0:08:11 And there’s a cost to changing those modalities
0:08:14 because we as a society have learned
0:08:18 how to manipulate these digital artifacts through these tools.
0:08:20 So the challenge for us was,
0:08:22 okay, you have to build this hardware,
0:08:24 which has to do all these amazing things
0:08:27 and also be attractive and also be light
0:08:28 and also be affordable.
0:08:31 And none of these existed before.
0:08:32 And what I tell my team all the time is like,
0:08:34 that’s only half the problem.
0:08:37 The other half of the problem is, great, how do I use it?
0:08:39 Like, how do I make it feel natural to me?
0:08:41 I’m so good with my phone now.
0:08:44 It’s an extension of my body, of my intention at this point.
0:08:47 How do we make it even easier?
0:08:49 And so we were having these challenges.
0:08:51 And then, what a wonderful blessing.
0:08:54 AI came in two years ago, much sooner than we expected.
0:08:56 And it’s a tremendous opportunity
0:08:58 to make this even easier for us.
0:08:59 Because the AIs that we have today
0:09:01 have a much greater ability to understand
0:09:02 what my intentions are.
0:09:04 I can give vague reference,
0:09:06 and it’s able to work through the corpus of information
0:09:09 it has available to make specific outcomes happen from it.
0:09:11 There’s still a lot of work to be done
0:09:13 to actually adapt it.
0:09:15 And it’s still not yet a control interface.
0:09:17 Like, I can’t reliably work my machine with it.
0:09:19 There’s a lot of things that we have to do.
0:09:21 We know what those things are.
0:09:24 And so, now you’re in a much more exciting place, actually.
0:09:25 Whereas before, we thought, okay,
0:09:28 we’ve got this big hill to climb on the hardware.
0:09:30 We’ve got this big hill to climb on the interaction design.
0:09:31 But we think we can do it.
0:09:33 And now we’ve got a wonderful tailwind,
0:09:35 where on the interaction design side, at least,
0:09:39 there’s the potential of having this much more intelligent agent
0:09:43 that now has not only the ability for you
0:09:46 to converse with it naturally and get results out of it,
0:09:50 but also to know by context what you’re seeing,
0:09:51 what you’re hearing, what’s going on around you.
0:09:51 Yeah.
0:09:54 And make intelligent inference based on that information.
0:09:56 Let’s talk about, like, reality labs
0:09:58 and this suite of products, what it is today.
0:10:01 So, you have Quest headsets, you have the smart glasses,
0:10:03 and then on the far end of the spectrum is Orion
0:10:05 and some of the stuff that I demoed.
0:10:07 So, just talk about the evolution of those efforts
0:10:10 and what you think the markets are for them
0:10:12 and how they converge versus not over time.
0:10:14 So, when we started the Ray-Ban Meta Project,
0:10:16 they were going to be smart glasses.
0:10:18 And, in fact, they were entirely built,
0:10:20 and we were six months away from production
0:10:22 when Llama 3 hit.
0:10:24 And the team was like, no, we got to do this.
0:10:25 And so, now they’re AI glasses, right?
0:10:27 Like, they didn’t start as AI glasses,
0:10:28 but the form factor was already right.
0:10:30 We could already do the compute.
0:10:31 We already had the ability.
0:10:32 So, yeah, now you have these glasses
0:10:33 that you can ask questions to.
0:10:36 And, in December, to the early access program,
0:10:37 we launched what we call Live AI.
0:10:39 So, you can start a Live AI session
0:10:41 with your Ray-Ban Meta glasses,
0:10:43 and for 30 minutes until the battery runs out,
0:10:44 it’s seeing what you’re seeing.
0:10:45 Yeah.
0:10:46 And it’s funny because, on paper,
0:10:49 the Ray-Ban Meta looks like an incremental improvement
0:10:50 to Ray-Ban Stories.
0:10:52 And this is kind of the story I’m trying to tell,
0:10:54 which is, the hardware isn’t that different
0:10:55 between the two,
0:10:58 but the interactions that we enable
0:11:00 with the person using it
0:11:02 are so much richer now.
0:11:03 When you use Orion,
0:11:05 when you use the full AI glasses,
0:11:08 you can imagine a post-phone world.
0:11:09 You’re like, oh, wow.
0:11:11 Like, if this was attractive enough
0:11:12 and light enough
0:11:13 and had battery life enough
0:11:14 to wear all day,
0:11:15 this would have all the stuff I need.
0:11:17 Like, it would all be right here.
0:11:18 And when you start to combine that
0:11:20 with the images that we have
0:11:21 of what AI is capable of.
0:11:21 So, you did the demo
0:11:23 where we showed you the breakfast.
0:11:24 Yeah, it did.
0:11:25 And it’s, yeah,
0:11:25 and for what it’s worth,
0:11:26 I mean, I’ll explain it
0:11:27 because it’s very cool.
0:11:28 Got to walk over
0:11:30 and there’s a bunch of breakfast ingredients laid out.
0:11:32 And I look at it
0:11:33 and I say,
0:11:33 hey, Meta,
0:11:35 what are some recipes?
0:11:35 That’s right.
0:11:36 And these ingredients.
0:11:38 So, that is, for me at least,
0:11:39 when we think about Orion,
0:11:41 initially,
0:11:43 it didn’t have that AI component
0:11:44 when we first thought about it.
0:11:45 It had this component
0:11:47 that was very direct manipulation.
0:11:48 So, it was very much modeled
0:11:49 on the app model
0:11:49 that we’re all familiar with.
0:11:50 Yeah, of course.
0:11:51 And I think there’s a version of that.
0:11:52 Yeah, of course,
0:11:53 you’re going to want to do calls
0:11:53 and you’re going to want to be able
0:11:54 to do your email
0:11:56 and be able to do your texting
0:11:57 and you want to be able to play games.
0:11:58 We have to play our Stargazer game
0:12:00 and you want to do your Instagram reels.
0:12:01 What we’re now excited about
0:12:01 is, okay,
0:12:02 take all those pieces
0:12:04 and layer on the ability
0:12:07 to have an interactive assistant
0:12:08 that really understands
0:12:10 not just what’s happening
0:12:11 on your device
0:12:13 and what email’s coming in,
0:12:14 but also what’s happening
0:12:16 in the physical world around you
0:12:17 and is able to connect
0:12:19 what you need in the moment
0:12:20 with what’s happening.
0:12:21 And so, these are concepts
0:12:21 where you’re like,
0:12:23 wow, what if the entire app model
0:12:23 is upside down?
0:12:24 What if it isn’t like,
0:12:25 hey, I want to go fetch
0:12:26 Instagram right now.
0:12:26 It’s like, hey,
0:12:28 the device realizes
0:12:28 that you have a moment
0:12:29 between meetings,
0:12:30 you’re a little bit bored.
0:12:30 Hey, do you want to catch up
0:12:31 on the latest highlights
0:12:33 from your favorite basketball team?
0:12:34 Those things become possible.
0:12:35 Having said that,
0:12:36 the hardware problems are hard
0:12:36 and they’re real
0:12:37 and the cost problems
0:12:38 are hard and they’re real.
0:12:39 And come for the king,
0:12:40 you best not miss.
0:12:41 The phone is an incredible
0:12:43 centerpiece of our lives today.
0:12:45 It’s how I operate my home.
0:12:46 I use it in my car.
0:12:46 I use it for work.
0:12:47 It’s everywhere, right?
0:12:50 And the world has adapted
0:12:51 itself to the phone.
0:12:53 So, it’s weird that my ice maker
0:12:53 has a phone app,
0:12:54 but it does.
0:12:54 Like, I don’t know.
0:12:55 I’m not sure.
0:12:56 It seems excessive,
0:12:57 but like,
0:12:58 so somebody today
0:12:58 who’s like,
0:12:59 I got to make an ice maker,
0:13:00 number one job,
0:13:01 got to have an app.
0:13:03 It’s like the smart refrigerator.
0:13:04 You’re like,
0:13:04 I don’t need this.
0:13:05 Take it out of me.
0:13:06 I do think it’s going
0:13:07 to be a long,
0:13:07 that’s why I said
0:13:09 the 10-year view for me
0:13:10 is, I think, much clearer.
0:13:11 I think these things
0:13:13 are going to be available,
0:13:14 widely accepted,
0:13:16 increasingly adopted.
0:13:17 The five-year view is harder
0:13:18 because, man,
0:13:19 like, even if it’s amazing.
0:13:20 Knocking out the dominance
0:13:21 of the phone in five years,
0:13:22 it just seems so hard.
0:13:22 It seems unthinkable.
0:13:24 It’s unthinkable for us, right?
0:13:24 That’s why I said,
0:13:25 like, Orion was the first
0:13:27 time I thought me.
0:13:27 Orion, like,
0:13:28 putting that in my head,
0:13:28 I was like,
0:13:31 okay, it could happen.
0:13:32 Like, there does exist
0:13:33 a life for us as a species
0:13:34 past the phone.
0:13:34 Yeah.
0:13:35 Yeah, it still has
0:13:36 the whole dynamic of,
0:13:37 well, how do I envision
0:13:38 my life without the operating
0:13:39 system that I’m so accustomed to?
0:13:39 Totally.
0:13:40 So I see the physical stuff
0:13:41 that you do,
0:13:42 but just the familiarity
0:13:43 and all the stuff
0:13:44 that’s working in there.
0:13:45 So what do you think
0:13:47 of the interim period?
0:13:49 So maybe you get to the point
0:13:50 where the hardware is capable,
0:13:52 it is market accessible,
0:13:54 but do you tether
0:13:54 to the phone?
0:13:56 Do you take a strong view
0:13:57 that you will never do that
0:13:58 and let the product stand?
0:13:59 Like, how do you think
0:14:00 about that piece?
0:14:02 The phones have this huge
0:14:03 advantage and disadvantage.
0:14:04 Huge advantage,
0:14:05 which is like,
0:14:06 the phone is already central
0:14:07 to our lives.
0:14:08 It’s already got this huge
0:14:09 developer ecosystem.
0:14:10 It’s this anchor device,
0:14:11 and it’s a wonderful anchor
0:14:12 device for that.
0:14:13 The disadvantages,
0:14:14 I actually think what we found
0:14:17 is the apps want to be different
0:14:19 when they’re not controlled
0:14:20 via touchscreen.
0:14:22 And that’s not super novel.
0:14:23 A lot of people failed
0:14:24 early in mobile,
0:14:25 including us,
0:14:26 by just taking our web stuff
0:14:27 and putting it on
0:14:27 the mobile phone
0:14:28 and being like,
0:14:29 oh, the mobile phone,
0:14:30 we’ll just put the web there.
0:14:32 But because it wasn’t native
0:14:33 to what the phone was,
0:14:34 and I mean everything
0:14:36 from interaction design
0:14:38 to the actual design
0:14:39 to the layout
0:14:40 to how it felt,
0:14:41 because we weren’t doing
0:14:43 phone native things,
0:14:44 we were failing
0:14:45 with one of the most popular
0:14:46 products in the history
0:14:46 of the web.
0:14:48 It’s just like the major
0:14:49 design field,
0:14:50 like the skeuomorphic idea
0:14:51 versus the native idea.
0:14:52 Yeah, and I think
0:14:53 having the developers
0:14:54 is a true value,
0:14:54 and I think having all
0:14:55 this application functionality
0:14:56 is a true value.
0:14:58 but then once you actually
0:15:00 reproject it into space
0:15:01 and you’re manipulating it
0:15:04 with your fingers like this
0:15:05 as opposed to a touchscreen,
0:15:06 you have much less precision.
0:15:08 It doesn’t respond
0:15:08 to voice commands
0:15:10 because there’s no tools
0:15:11 for that.
0:15:11 There’s no design
0:15:12 integration for that.
0:15:14 So having a phone platform
0:15:16 today feels like,
0:15:17 wow, I’ve got this huge base
0:15:17 to work from
0:15:18 on the hardware side,
0:15:19 but I’ve also actually got
0:15:20 this kind of huge anchor
0:15:22 to drag on the software side.
0:15:24 And so we’re not opposed
0:15:24 to these partnerships,
0:15:25 and I think it’ll be interesting
0:15:26 to see once the hardware
0:15:27 is a little bit more developed
0:15:28 how partners feel about it.
0:15:30 And I hope they continue
0:15:32 to support people
0:15:33 who buy these phones
0:15:34 for $1,200, $1,300,
0:15:35 being able to bring
0:15:36 whatever hardware
0:15:37 they want to bring
0:15:38 and take the full functionality
0:15:39 of that with them.
0:15:41 The biggest question I have
0:15:42 is whether the entire app model,
0:15:43 because we were imagining
0:15:45 a very phone-like app model
0:15:46 for these devices,
0:15:47 admittedly a very different
0:15:48 interaction design,
0:15:50 input, and control schemes
0:15:50 are very different
0:15:51 and that demands
0:15:52 like a little extra
0:15:52 developer attention.
0:15:54 I am wondering if like
0:15:55 the progression of AI
0:15:56 over the next several years
0:15:57 doesn’t turn the app model
0:15:58 in its head.
0:15:59 Like right now,
0:16:00 it’s kind of an unusual thing
0:16:01 where I’m like,
0:16:03 I want to play music.
0:16:04 So in my head,
0:16:05 I translate that to
0:16:06 I have to go open Spotify
0:16:07 or open Tidal,
0:16:08 and the first thing I think of
0:16:09 is who is my provider
0:16:10 going to be?
0:16:11 Yeah, of course.
0:16:12 As opposed to like,
0:16:12 that’s not what I want.
0:16:13 It’s extremely limiting.
0:16:14 What I want is to play music.
0:16:14 Yes.
0:16:15 And I just want to be like,
0:16:16 go to the AI,
0:16:16 I’m like, cool,
0:16:18 play this music for me.
0:16:18 Yeah.
0:16:20 And it should know,
0:16:21 oh, like you’re already
0:16:23 using this service.
0:16:24 We’ll use that one.
0:16:25 Or these two services
0:16:26 are both available to you.
0:16:27 This one has a better quality song.
0:16:29 Or this one has lower latency,
0:16:29 whatever the thing is.
0:16:30 Or it’s like, hey,
0:16:31 the song you want
0:16:31 isn’t available
0:16:32 on any of these services.
0:16:33 Do you want to sign up
0:16:34 for this other service
0:16:34 that does have the song
0:16:35 that you want?
0:16:36 I don’t want to have
0:16:36 to be responsible
0:16:37 for orchestrating like
0:16:38 what app I’m opening
0:16:39 to do a thing.
0:16:40 We’ve had to do that
0:16:41 because that’s how
0:16:41 things were done
0:16:43 in the entire history
0:16:43 of digital computing.
0:16:45 You had an application-based model
0:16:46 that was the system.
0:16:47 So I do wonder
0:16:49 how much AI inverts things.
0:16:50 That’s a pretty hot take.
0:16:51 Yeah, that’s a hot take.
0:16:51 Inverts things.
0:16:53 And that’s not about wearables.
0:16:54 That’s not about anything.
0:16:54 That’s just like,
0:16:55 even at the phone level,
0:16:57 if you were building
0:16:57 a phone today,
0:16:59 would you build an app store
0:16:59 the way you historically
0:17:00 built an app store?
0:17:01 Or would you say like,
0:17:03 hey, you as a consumer,
0:17:04 express your intention.
0:17:05 Express what you’re
0:17:06 trying to accomplish.
0:17:07 And let’s like,
0:17:07 see what we have.
0:17:08 Let the system see
0:17:08 what it can produce.
0:17:09 Yeah, for you.
0:17:10 But I do think
0:17:11 if you were starting
0:17:12 from scratch today,
0:17:14 you probably wouldn’t build
0:17:15 this like app-centric world
0:17:16 where I, as a consumer,
0:17:17 I’m trying to solve a problem
0:17:18 and first have to decide
0:17:20 which of the providers
0:17:20 I’m going to use
0:17:21 to solve that problem.
0:17:21 Yeah, of course.
0:17:23 That’s fascinating.
0:17:24 And again,
0:17:25 I think it’s a function
0:17:26 of where the capabilities
0:17:27 are today
0:17:27 and I think where we have
0:17:28 line of sight
0:17:29 into orchestration capabilities.
0:17:30 Because I’d say,
0:17:31 knowledge-wise,
0:17:33 that is probably capable today.
0:17:35 I think orchestration-wise,
0:17:35 it’s probably
0:17:37 we’re a little bit away.
0:17:38 And then, of course,
0:17:39 you’ve got to build
0:17:40 the developer ecosystem
0:17:41 to develop on the platform.
0:17:42 Which is incredibly hard.
0:17:43 That’s the thing
0:17:43 I want to see
0:17:44 That’s the hardest piece, right?
0:17:45 That’s the hardest piece.
0:17:45 Yeah.
0:17:46 The stronger we get
0:17:48 at agentic reasoning
0:17:49 and capabilities,
0:17:50 the more I can rely
0:17:52 on my AI
0:17:53 to do things in my absence.
0:17:54 And at first,
0:17:55 it will be knowledge work,
0:17:55 of course.
0:17:56 That’s fine.
0:17:57 But once you have a flow
0:17:59 of consumers
0:18:00 coming through here,
0:18:00 what you’re going to find
0:18:01 is that they’re going to have
0:18:02 a bunch of dead ends.
0:18:02 Yeah.
0:18:03 Where they’re going to ask the AI,
0:18:05 hey, can you do this thing for me?
0:18:05 And it’s going to say,
0:18:06 no, I can’t.
0:18:09 That’s the goldmine
0:18:10 that you take to developers.
0:18:10 And you’re like,
0:18:13 hey, I’ve got 100,000 people a day
0:18:14 trying to solve this problem.
0:18:15 They’re trying to use your app.
0:18:15 Yeah.
0:18:16 They don’t know they are,
0:18:17 but they’re trying to use their app.
0:18:18 Look, here’s the query stream.
0:18:19 Here’s what’s coming through.
0:18:21 And we’re going to tell them no today.
0:18:23 If you build these hooks,
0:18:24 you’ve got 100,000 people
0:18:25 clamoring for something today.
0:18:27 Coming in for your service.
0:18:27 Yeah.
0:18:28 And it’s totally fine
0:18:30 for RAI to go back and say,
0:18:31 hey, you’ve got to pay for this.
0:18:32 There’s a guy who does this for you,
0:18:33 but you’ve got to pay for it.
0:18:34 Yeah.
0:18:34 And by the way,
0:18:35 I’m not just talking about apps.
0:18:38 There’s some kind of a marketplace here
0:18:41 that I think emerges over time.
0:18:43 So that’s how I see it playing out.
0:18:44 I don’t see it playing out
0:18:46 as like someone goes into a darkroom
0:18:47 and comes up with this app platform.
0:18:48 No.
0:18:49 What’s going to happen is
0:18:50 there’s going to become a query stream
0:18:53 of people using AI to do things.
0:18:55 And the AI will fail
0:18:57 repeatedly in certain areas
0:18:59 because that’s a type of functionality
0:19:00 that is currently behind
0:19:01 some kind of an app wall.
0:19:02 And there’s no…
0:19:04 Or it hasn’t been built native
0:19:06 to whatever consumption mechanism.
0:19:07 There’s no bridge that can be built.
0:19:07 Yeah, yeah, yeah.
0:19:08 And everyone wants to build the bridges.
0:19:08 They’re like, no, no.
0:19:10 It’s going to manipulate the pixels
0:19:12 and it’s going to manipulate…
0:19:13 It’s like, fine, it can do those things.
0:19:13 I’m not saying the AI
0:19:15 can’t cross those boundaries.
0:19:16 But I think over time,
0:19:18 that becomes the primary interface
0:19:21 for humans interacting with software
0:19:23 as opposed to the like
0:19:24 pick from the garden of applications.
0:19:25 Yeah, that makes a ton of sense.
0:19:29 That’s a very alluring end state
0:19:31 just as a consumer, right?
0:19:31 Yeah, it’s messy.
0:19:33 And I think it creates
0:19:34 these very exciting marketplaces
0:19:37 for functionality inside the AI.
0:19:39 It abstracts away
0:19:41 a lot of companies’ brand names,
0:19:42 which I think is going to be very hard
0:19:45 for an entire generation of brands.
0:19:45 Yeah.
0:19:47 Like the fact that I don’t care
0:19:48 if it’s being played
0:19:49 on one of these two music services,
0:19:51 that’s hard for those music services
0:19:54 who like really want me to care.
0:19:55 Yeah, yeah, yeah.
0:19:55 And like they want me
0:19:58 to have a stronger opinion about it.
0:19:58 And like they want me
0:19:59 to have an attachment.
0:19:59 Yeah.
0:20:00 I don’t want to have an attachment.
0:20:01 There are some things
0:20:02 where you may value the attachment
0:20:03 and you don’t whatever.
0:20:04 Yeah, in the world
0:20:04 where I’m like,
0:20:05 hey, there’s an app garden
0:20:06 and these two are competing
0:20:07 for my eyeballs,
0:20:09 the brand that they’ve built
0:20:11 is the hugely valuable asset.
0:20:13 In the world where
0:20:15 I just care if the song gets played
0:20:15 and sounds good,
0:20:17 a different set of priorities
0:20:18 are important.
0:20:20 I think that’s net positive
0:20:21 because what matters now
0:20:23 is performance on the job
0:20:23 being asked.
0:20:24 actual product experience
0:20:25 and value
0:20:26 and price per performance
0:20:27 like matters a lot.
0:20:28 Yeah.
0:20:29 I think a lot of companies
0:20:29 won’t love that.
0:20:31 Well, abstracting away,
0:20:33 that’s like effectively articulating,
0:20:35 abstracting away margin pools,
0:20:36 which puts a lot more pressure
0:20:39 on us trusting the AI
0:20:41 or the distributor of the AI.
0:20:42 And so far as I’m floating
0:20:43 between different companies
0:20:44 that are each providing AIs,
0:20:46 the degree which I trust them
0:20:47 to not be bought
0:20:48 and paid for in the back end,
0:20:49 they’re not giving me
0:20:50 the best experience
0:20:51 or the best price per money.
0:20:52 They’re giving the one
0:20:53 that gives them the most money.
0:20:54 Yeah, of course.
0:20:55 So, yeah, it’s the experience
0:20:56 of Google Search today, right?
0:20:57 It’s a very different world.
0:20:59 It’s a very different world.
0:21:00 It’s a very different world.
0:21:01 But you can actually see
0:21:03 inklings of it today, right?
0:21:04 So certain companies
0:21:05 are willing to work
0:21:06 with the new AI providers
0:21:08 in agentic task completion.
0:21:09 Yeah, yeah.
0:21:09 And then they’re like,
0:21:10 well, actually, wait a minute.
0:21:12 I don’t just want the bots
0:21:13 executing this stuff.
0:21:14 I want the humans coming to me.
0:21:15 I think I need that.
0:21:16 Yeah, right.
0:21:16 It’s existential
0:21:18 that I have this brand relationship
0:21:19 directly with the demand side.
0:21:20 Yeah.
0:21:22 So that’s potentially messy,
0:21:24 but a bright future,
0:21:24 especially if we don’t have
0:21:26 to pay that like brand tax.
0:21:27 Yeah, it’ll be very messy.
0:21:30 I don’t know it’s avoidable
0:21:31 because I think once consumers
0:21:33 start to get into these tight loops
0:21:35 where more and more
0:21:35 of their interactions
0:21:38 are being moderated by an AI,
0:21:39 you won’t have a choice.
0:21:40 That’s like where
0:21:41 your customers will be.
0:21:42 But it’s going to be
0:21:42 a pretty different world.
0:21:43 Yeah, it’ll be a different world
0:21:44 and there’ll probably be
0:21:45 some groups that try
0:21:46 to move fast to it.
0:21:47 as a way to compete
0:21:48 with things that are branded.
0:21:48 Yeah.
0:21:49 And just say,
0:21:49 I’m going to compete
0:21:50 on performance and price.
0:21:50 Yeah, that’s right.
0:21:51 Where do you think
0:21:52 that could potentially
0:21:53 happen first?
0:21:55 It probably will mirror
0:21:56 query volume.
0:21:57 I think of this a lot.
0:21:58 We do have a model of this,
0:22:00 which was in the web era
0:22:01 when Google became
0:22:03 the dominant search engine.
0:22:04 So before that,
0:22:05 the web era was like
0:22:07 very index based.
0:22:07 It was like Yahoo
0:22:09 and it was like links
0:22:11 and getting major sources
0:22:12 of traffic to link to you
0:22:12 was the game.
0:22:14 And then once Google
0:22:16 came to dominance,
0:22:17 which happened very quickly
0:22:18 over maybe a couple of years,
0:22:18 I feel like.
0:22:19 All that mattered
0:22:20 was like SEO.
0:22:21 All that mattered
0:22:21 was like where you were
0:22:22 in the query stream.
0:22:22 Yeah.
0:22:23 And the query stream
0:22:26 dictated what businesses
0:22:27 came over and succeeded.
0:22:28 Yeah.
0:22:29 Because like the queries
0:22:30 that were the most frequent,
0:22:31 those were the ones
0:22:32 that came first.
0:22:32 Yeah.
0:22:37 Travel came right away.
0:22:38 It was a huge disruption
0:22:39 and travel agents
0:22:40 went from a thing that existed
0:22:41 to a thing that didn’t exist
0:22:42 in a relatively short-
0:22:42 Immediately.
0:22:44 And they all competed
0:22:45 on the basis of like
0:22:46 execution of the best deal
0:22:47 in a seamless fashion
0:22:49 with the highest conversion.
0:22:50 I think SEO has gotten
0:22:51 to a point now
0:22:53 where it’s kind of a bummer.
0:22:54 It’s like made things worse.
0:22:55 It’s like everyone’s gotten so good.
0:22:56 It’s just like game.
0:22:57 Everyone’s gotten so good at it.
0:22:59 Especially with AI actually now.
0:22:59 That’s right.
0:23:00 So I actually think it’s like
0:23:01 we had this incredible
0:23:02 flattening curve
0:23:02 and now it’s like starting
0:23:03 to kind of rise up
0:23:03 in terms of-
0:23:05 Especially with paid placement too.
0:23:05 Yeah.
0:23:06 That’s so dominant.
0:23:07 So duh.
0:23:08 Yeah, that’s right.
0:23:09 And this is like probably
0:23:10 the cautionary tale
0:23:11 for how this plays out
0:23:12 in AIs as well.
0:23:13 I think there will be
0:23:16 a pretty good golden era here
0:23:17 where the query stream
0:23:18 will dictate
0:23:20 what businesses come first
0:23:21 because those are the queries
0:23:22 that are-
0:23:23 That’s the volume of people
0:23:24 unsatisfied with the existing
0:23:26 solutions that they have.
0:23:26 Yeah.
0:23:27 Otherwise they wouldn’t
0:23:28 be asking about it.
0:23:29 And product providers
0:23:30 and developers will follow that.
0:23:31 And build specifically-
0:23:32 solve those problems.
0:23:33 That’s right.
0:23:34 Once it tips
0:23:36 in each vertical
0:23:37 we get a lot of progress
0:23:37 very quickly
0:23:39 towards better solutions
0:23:39 for consumers.
0:23:41 And then once it hits
0:23:41 a steady state
0:23:42 it starts to be
0:23:43 gamesmanship.
0:23:43 Yeah.
0:23:44 And that’s the thing to fight.
0:23:46 And that’s decaying or-
0:23:47 That’ll be the true test of AI.
0:23:48 The true test.
0:23:48 Can it get through that?
0:23:50 Can it avoid falling into that trap?
0:23:51 Can it avoid that trap?
0:23:51 Yeah, yeah.
0:23:52 That’s right.
0:23:52 Exactly.
0:23:53 Well a lot of that
0:23:54 is business model driven
0:23:55 and we’ll see how that evolves
0:23:55 over time too.
0:23:56 That’s right.
0:23:57 You guys have also been
0:23:59 leading from the front
0:24:00 on this idea of open source.
0:24:01 Yeah.
0:24:02 And so talk about
0:24:03 some of your efforts
0:24:05 on that side of the business
0:24:05 and then
0:24:07 what is the ideal
0:24:08 market structure
0:24:10 of the AI model side
0:24:10 for you guys?
0:24:11 There’s two parts
0:24:12 that came together.
0:24:13 The first one is
0:24:14 Llama came out of FAIR
0:24:16 our fundamental AI research group
0:24:17 and that’s been
0:24:18 an open source
0:24:19 research group
0:24:20 since the beginning.
0:24:20 Yeah.
0:24:21 Since John LeCun came in
0:24:22 and they established that
0:24:24 it’s allowed us to attract
0:24:24 incredible researchers
0:24:26 who really believe
0:24:26 that we’re going to make
0:24:28 more progress as a society
0:24:29 working together
0:24:29 across boundaries
0:24:30 of individual labs
0:24:31 than not.
0:24:33 And to be fair
0:24:33 it’s not just us
0:24:34 obviously
0:24:36 the Transformer paper
0:24:36 was published at Google
0:24:37 and like you know
0:24:39 big self-supervised learning
0:24:39 was our contribution
0:24:40 like everyone’s contributing
0:24:41 to the knowledge base
0:24:43 but when we open source Llama
0:24:44 that’s how all models
0:24:45 were open source
0:24:45 at that point.
0:24:45 Yeah.
0:24:47 Of course.
0:24:48 Like everyone was open
0:24:49 the only thing that was unusual
0:24:50 was everything else
0:24:50 just went closed source
0:24:51 over time.
0:24:51 Yeah.
0:24:52 Effectively.
0:24:52 That’s right.
0:24:53 But before that
0:24:55 every time someone built a model
0:24:55 they open sourced it
0:24:56 so that other people
0:24:56 could use the model
0:24:57 and see how great
0:24:58 that model was.
0:24:58 Like that was like
0:24:59 mostly how it was done.
0:25:00 Sure.
0:25:01 If it was worth anything.
0:25:02 Certainly some specialized models
0:25:03 for translations and whatnot
0:25:04 were kept closed
0:25:05 but like if it was a general model
0:25:05 that was what was done.
0:25:06 Llama 2 was probably
0:25:08 the big decision point for us.
0:25:08 Llama 2
0:25:09 and this is where I think
0:25:10 the second thing
0:25:11 that came in
0:25:12 was a belief that I’ve had
0:25:12 that I was advancing
0:25:14 really strenuously internally
0:25:15 that Mark Ridley believes in too
0:25:16 and he’s written
0:25:16 his post about this
0:25:17 which is first of all
0:25:18 we’re going to make way more progress
0:25:19 if these models are open.
0:25:20 Yeah.
0:25:21 Because a lot of these contributions
0:25:22 aren’t going to come
0:25:24 from these big labs
0:25:24 like they’re going to come
0:25:25 from these little labs
0:25:26 and we’ve seen this already
0:25:26 with DeepSeq in China
0:25:28 which was put in a tough spot
0:25:29 and then innovated
0:25:31 incredibly in the memory architectures
0:25:31 and a couple other places
0:25:33 to really get amazing results.
0:25:34 And so we really believe
0:25:35 we’re going to get
0:25:36 the most progress collectively.
0:25:37 The second thing is
0:25:38 inside this piece is
0:25:39 you know this is a classic
0:25:40 I believe these are going to be commodities
0:25:42 and you want to commoditize
0:25:42 your complements.
0:25:43 Yes.
0:25:44 And we’re in a unique position
0:25:45 strategically where
0:25:46 our products are made better
0:25:47 through AI
0:25:48 which is why we’ve been
0:25:49 investing in it for so long.
0:25:50 Whether it’s recommendation systems
0:25:51 in what you’re seeing
0:25:52 in feed or reels
0:25:54 whether it’s simple things
0:25:55 like what friend
0:25:56 do I put at the top
0:25:57 when you type
0:25:57 you want to make a new message
0:25:58 who do I think
0:25:58 you’re going to message right now?
0:26:00 Little things like that
0:26:01 to really big expansive things
0:26:03 like hey here’s an entire answer
0:26:04 here’s an entire search interface
0:26:05 that we couldn’t do before
0:26:06 in WhatsApp
0:26:07 that like now
0:26:09 is a super popular surface.
0:26:10 So there’s all these things
0:26:11 that are possible for us
0:26:12 that are made better
0:26:13 by this AI
0:26:14 but nobody else
0:26:15 having this AI
0:26:16 can then build our product.
0:26:18 The asymmetry works in our favor.
0:26:18 Yeah of course.
0:26:19 And so for us
0:26:20 like commoditizing your complements
0:26:21 is just good business sense
0:26:22 and making sure that there is
0:26:24 a lot of competitively priced
0:26:26 if not almost free
0:26:27 models out there
0:26:29 helps the entire industry
0:26:31 helps a bunch of small startups
0:26:32 and academic labs
0:26:33 it also helps us.
0:26:34 Yeah you as the application provider
0:26:35 are huge beneficiaries.
0:26:36 So we’re all super aligned.
0:26:37 Yeah you’re aligned.
0:26:38 Business model alignment
0:26:38 and industry alignment.
0:26:40 It’s a strong alignment there.
0:26:40 Yes.
0:26:42 So it comes from both
0:26:43 this fundamental belief
0:26:44 in how this kind of research
0:26:45 should be done
0:26:46 and then aligns
0:26:47 with the other business model
0:26:48 and so there’s no conflict.
0:26:49 Yeah societal progress
0:26:50 plus business model alignment.
0:26:51 It’s all together.
0:26:52 It’s all great.
0:26:53 It’s all going the same direction.
0:26:53 That’s awesome.
0:26:54 It’s great.
0:26:55 I want to shift gears
0:26:56 to talking about
0:26:58 the impediments to progress
0:26:59 and like what you think
0:27:00 you know are kind of
0:27:01 linear versus not.
0:27:03 So the risks to the vision
0:27:04 to the overall vision
0:27:05 that you articulated
0:27:06 obviously hardware
0:27:07 Yep.
0:27:08 AI capabilities.
0:27:08 Yep.
0:27:10 Vision capabilities
0:27:11 and screens
0:27:12 and all that
0:27:12 resolutions.
0:27:14 We talked about the ecosystem
0:27:15 and developers
0:27:17 and native products.
0:27:18 So maybe just talk about
0:27:18 what you see
0:27:19 are kind of
0:27:20 the linear path things
0:27:22 and the things
0:27:22 that may be harder
0:27:23 or riskier.
0:27:26 we have real invention risk.
0:27:27 There exists risk
0:27:28 that the things
0:27:28 that we want to build
0:27:30 we don’t have the capacity
0:27:31 to build as a society
0:27:32 as a species yet.
0:27:33 Yeah.
0:27:34 And that’s not a guarantee.
0:27:36 I think we have windows to it.
0:27:37 You’ve seen Orion
0:27:38 so like it can be done.
0:27:38 Yeah there’s probably.
0:27:39 Yeah it feels like
0:27:40 it’s a cost reduction exercise
0:27:42 it’s a materials improvement exercise
0:27:43 but it can be done.
0:27:44 There is still some invention risk.
0:27:46 Far bigger than the invention risk
0:27:47 I think is the adoption risk.
0:27:48 Is it considered socially acceptable?
0:27:50 Are people willing
0:27:51 to learn a new modality?
0:27:52 Like we all learned to type
0:27:53 when we were kids at this point.
0:27:54 We were born with phones
0:27:55 in our hands at this point.
0:27:55 Yeah.
0:27:56 Are people willing to learn
0:27:57 a new modality?
0:27:57 Is it worth it to them?
0:27:58 Ecosystem risk
0:27:59 even bigger than that.
0:28:00 Like great you build this thing
0:28:02 but if it just does like
0:28:03 your email and reels
0:28:04 that’s probably not enough.
0:28:05 Do people bring
0:28:06 the suite of software
0:28:07 that we require
0:28:08 to interact with
0:28:09 modern human society
0:28:11 to bear on the device?
0:28:13 Those are all huge risks.
0:28:14 I will say
0:28:15 we feel pretty good
0:28:16 about where we’re getting
0:28:17 on the hardware
0:28:18 on acceptability.
0:28:19 We think we can do
0:28:20 those things.
0:28:20 That was not a guarantee
0:28:21 before I think
0:28:23 with the Ray-Van Metaglasses
0:28:23 we’re feeling like
0:28:24 okay we can get through
0:28:25 You feel like the acceptability
0:28:26 Humans will accept
0:28:28 that I’m using technology.
0:28:29 Within that
0:28:31 super interesting
0:28:32 regulatory challenges
0:28:33 here I have
0:28:34 an always on machine
0:28:35 that gives me
0:28:36 super human sensing.
0:28:37 My vision is better.
0:28:38 My hearing is better.
0:28:39 My memory is better.
0:28:40 That means
0:28:42 when I see you
0:28:43 a couple years from now
0:28:44 and I haven’t seen you
0:28:44 on the internet
0:28:45 I’m like
0:28:45 oh god I don’t remember
0:28:47 we did a podcast together
0:28:47 what’s the guy’s name?
0:28:49 Can I ask that question?
0:28:49 Am I allowed
0:28:50 to ask that question?
0:28:50 Yes.
0:28:51 What is your right
0:28:53 it’s your face
0:28:54 you showed me your face
0:28:55 and if I was somebody
0:28:55 with a better memory
0:28:57 I could remember the face
0:28:59 so like that happened
0:29:00 but I don’t have
0:29:00 a great memory
0:29:01 so am I allowed
0:29:02 to use a tool
0:29:03 to assist me or not?
0:29:04 So there’s really
0:29:05 subtle regulatory
0:29:06 privacy
0:29:07 social
0:29:07 acceptability
0:29:08 questions
0:29:08 that are like
0:29:09 embedded here
0:29:10 that are super deep
0:29:10 individually
0:29:12 and can derail
0:29:13 the whole thing
0:29:14 like you can easily
0:29:14 derail
0:29:15 easily derail
0:29:16 the whole thing
0:29:17 and slow progress
0:29:17 that’s the thing
0:29:18 I think we sometimes
0:29:20 think in our industry
0:29:21 it’s like feel the dreams
0:29:21 if you build it
0:29:22 they will come
0:29:22 and it’s like
0:29:23 no a lot of things
0:29:24 have to happen right
0:29:25 well you can also
0:29:26 overstep too
0:29:27 that’s the risk
0:29:28 you’re sure you get
0:29:28 your hands locked
0:29:29 great technology
0:29:31 can get derailed
0:29:32 for long periods
0:29:32 of time
0:29:33 nuclear power
0:29:34 got derailed
0:29:35 yeah for absolutely
0:29:36 stupid reasons
0:29:37 for 70 years
0:29:38 for bad reasons
0:29:39 we know we’re bad now
0:29:40 and it was like
0:29:41 they just played it wrong
0:29:42 yeah of course
0:29:42 and they were like
0:29:43 ah ignore this
0:29:43 it’s like no
0:29:45 people actually feel this way
0:29:46 so I think yeah
0:29:47 I feel pretty good
0:29:47 about the invention risk
0:29:48 acceptability risk
0:29:49 is looking better
0:29:50 than it has been
0:29:51 but like I think
0:29:51 there’s still a lot
0:29:53 of big hedges
0:29:53 to cross there
0:29:55 I actually think
0:29:56 the ecosystem risk
0:29:56 was one
0:29:57 I would have said
0:29:57 previously
0:29:58 was the biggest one
0:30:00 but AI is now
0:30:01 my potential
0:30:02 silver bullet there
0:30:03 if AI becomes
0:30:04 the major interface
0:30:05 then it comes for free
0:30:06 yeah
0:30:07 and I will also say
0:30:08 that we’ve had
0:30:09 such a positive response
0:30:10 from even just
0:30:11 set aside Orion
0:30:12 even the Ray-Ban Metas
0:30:14 companies that want
0:30:14 to work for us
0:30:15 and build on that platform
0:30:16 it’s not a platform yet
0:30:17 yeah it’s not
0:30:17 there’s so little
0:30:18 there’s so little compute
0:30:19 there’s so little compute
0:30:20 we just connect an app
0:30:21 we literally don’t
0:30:22 have any space yet
0:30:22 yeah
0:30:23 but we did do a partnership
0:30:24 with Be My Eyes
0:30:25 which like helps blind
0:30:26 and hard of vision
0:30:26 people navigate
0:30:28 and it’s really spectacular
0:30:29 and so there’s a little window
0:30:29 there where we can start
0:30:30 building
0:30:30 so yeah
0:30:31 I would say the response
0:30:32 has been more positive
0:30:33 than I had expected
0:30:35 so everything right now
0:30:36 tailwinds abound
0:30:36 right now
0:30:37 and to be honest
0:30:38 after eight years
0:30:40 of nine years
0:30:42 of headwinds
0:30:43 having a year of tailwinds
0:30:43 is nice
0:30:43 yeah
0:30:44 I’ll take it
0:30:44 I’ll take it
0:30:45 I’m not gonna look
0:30:45 in the face
0:30:46 no victory laps
0:30:47 yeah but that’s good
0:30:47 okay
0:30:48 but it’s all hard
0:30:50 at every point
0:30:50 it could all fail
0:30:51 yeah I like that you
0:30:51 just started with
0:30:52 it’s invention risk
0:30:53 it’s I don’t know
0:30:54 there’s many ways
0:30:55 this just won’t work
0:30:55 yeah that’s right
0:30:56 even if it does work
0:30:57 it might not take
0:30:59 well I’ll say two things
0:31:00 about this
0:31:01 and this is where
0:31:02 Mark just deserves
0:31:02 so much credit
0:31:04 is we’re true believers
0:31:06 like we have actual conviction
0:31:07 yeah
0:31:08 Mark believes
0:31:10 this is the next thing
0:31:12 it needs to happen
0:31:13 and it doesn’t happen
0:31:13 for free
0:31:15 like we can be the ones
0:31:15 to do it
0:31:16 our chief scientist
0:31:17 Michael Arash
0:31:18 who’s one of my favorite people
0:31:18 I’ve ever gotten a chance
0:31:19 to work with
0:31:20 he talks a lot about
0:31:22 the myth of technological eventualism
0:31:23 it doesn’t eventually happen
0:31:24 there’s a lot of people in tech
0:31:25 who are like
0:31:26 yeah AR will eventually happen
0:31:27 that’s not how it fucking works
0:31:28 that would not
0:31:28 that would actually
0:31:30 AR is a specific one
0:31:30 that would just absolutely not
0:31:32 you have to stop
0:31:33 and put the money
0:31:34 and the time
0:31:34 and do it
0:31:36 somebody has to stop
0:31:36 and do it
0:31:37 and that is the difference
0:31:38 the number one thing I’d say
0:31:39 is like the difference
0:31:40 between us and anybody else
0:31:41 is we believe in this stuff
0:31:42 in our cores
0:31:44 this is the most important work
0:31:45 I’ll ever get a chance to do
0:31:46 this is Xerox PARC level
0:31:48 new stuff
0:31:49 where we’re rethinking
0:31:50 how humans are going to interact
0:31:50 with computers
0:31:52 it’s like JCR Licklider
0:31:53 and the human in the loop computing
0:31:54 we’re seeing that with AI
0:31:55 it’s a rare moment
0:31:56 it’s a rare moment
0:31:57 it doesn’t even happen
0:31:58 once a generation I think
0:31:59 it may happen every other generation
0:31:59 every third generation
0:32:00 like you don’t get a chance
0:32:01 to do this all the time
0:32:03 so we’re not missing it
0:32:03 we’re just like
0:32:04 we’re going to do it
0:32:05 and we may fail
0:32:06 like it’s possible
0:32:07 but we will not fail
0:32:09 for lack of effort or belief
0:32:09 great
0:32:10 thanks a ton boss
0:32:11 cheers
0:32:11 yeah cheers
0:32:20 and we’ll see you in the next video
0:32:21 we’ll see you in the next video
0:32:22 we’ll see you in the next video
0:32:23 we’ll see you in the next video
0:32:23 we’ll see you in the next video
0:32:23 we’ll see you in the next video
0:32:24 we’ll see you in the next video
0:32:24 we’ll see you in the next video
0:32:24 we’ll see you in the next video
0:32:25 we’ll see you in the next video
0:32:25 we’ll see you in the next video
0:32:26 we’ll see you in the next video
0:32:26 we’ll see you in the next video
0:32:27 we’ll see you in the next video
0:32:27 we’ll see you in the next video
0:32:28 we’ll see you in the next video
0:32:28 we’ll see you in the next video
0:32:29 we’ll see you in the next video
0:32:30 we’ll see you in the next video
0:00:05 Every single interface that I interact with,
0:00:08 every single problem space that I’m trying to solve
0:00:12 are going to be made easier by virtue of this new technology.
0:00:14 If you were starting from scratch today,
0:00:17 you probably wouldn’t build this app-centric world.
0:00:21 You can imagine a post-phone world.
0:00:24 The past 20 years of consumer technology
0:00:27 have been a story of apps, of touchscreens, and of smartphones.
0:00:31 These form factors seemingly appeared out of nowhere
0:00:34 and may be replaced just as quickly as they were ushered in,
0:00:37 perhaps by a new AI-enabled stack,
0:00:40 a new computing experience that is more agentic,
0:00:42 more adaptive, and more immersive.
0:00:46 Now, in today’s episode, A16C’s growth general partner,
0:00:48 David George, discusses this feature
0:00:52 with arguably one of the most influential builders of this era.
0:00:56 That is Meta CTO, Andrew Boz Bosworth,
0:00:58 who spent nearly two decades at the company,
0:01:01 shaping consumer interaction from the Facebook newsfeed
0:01:05 all the way through to their work on smart glasses and AR headsets.
0:01:09 Here, Boz explores the art of translating emerging technologies
0:01:12 into real products that people use and love,
0:01:14 plus how breakthroughs in AI and hardware
0:01:17 could turn the existing app model on its head.
0:01:22 In this world, what new interfaces and marketplaces need to be developed?
0:01:24 What competitive dynamics hold strong?
0:01:26 And which fall by the wayside?
0:01:28 For example, will brands still be a moat?
0:01:31 And if we get it right, Boz says,
0:01:34 the next wave of consumer tech won’t run on taps and swipes,
0:01:36 it’ll run on intent.
0:01:39 So, is the post-mobile phone era upon us?
0:01:40 Listen in to find out.
0:01:43 Oh, and if you do like this episode,
0:01:45 it comes straight from our AI Revolution series.
0:01:48 And if you missed previous episodes of this series
0:01:50 with guests like AMD CEO Lisa Su,
0:01:52 Anthropic co-founder Dario Amadei,
0:01:55 and the founders behind companies like Databricks,
0:01:56 Waymo, Figma, and more,
0:02:00 head on over to a16z.com slash AI Revolution.
0:02:06 As a reminder, the content here is for informational purposes only,
0:02:09 should not be taken as legal, business, tax, or investment advice,
0:02:11 or be used to evaluate any investment or security,
0:02:14 and is not directed at any investors or potential investors
0:02:15 in any A16Z fund.
0:02:18 Please note that A16Z and its affiliates
0:02:21 may also maintain investments in the companies discussed in this podcast.
0:02:24 For more details, including a link to our investments,
0:02:27 please see a16z.com slash disclosures.
0:02:34 Boz, thanks for being here.
0:02:35 Thanks for having me.
0:02:36 Appreciate it.
0:02:38 Okay, I want to jump right in.
0:02:42 How are we all going to be consuming content
0:02:44 five years from now and ten years from now?
0:02:45 Ten years, I feel pretty confident
0:02:49 that we will have a lot more ways to bring content into our view shed
0:02:50 than just taking out our fun.
0:02:54 I think augmented reality glasses, obviously, are a real possibility.
0:02:55 I’m also hoping that we can do better
0:02:58 for really engaging in immersive things.
0:03:00 Right now, you have to travel to, like, the sphere,
0:03:02 which is great, but there’s one of them.
0:03:03 It’s in Vegas since a trip.
0:03:05 Are there better ways that we can have access to
0:03:07 if we really want to be engaged in something,
0:03:09 not just immersively, but also socially?
0:03:10 So it’s like, oh, I want to watch the game.
0:03:11 I want to watch it with my dad.
0:03:12 I want to feel like we’re courtside.
0:03:15 Sure, we can go and pay a lot for tickets.
0:03:16 Is there a better way?
0:03:16 I think there is.
0:03:18 So ten years, I feel really good
0:03:20 about all these alternative content delivery vehicles.
0:03:21 Five years is trickier.
0:03:24 For example, I think the glasses, the smart glasses,
0:03:26 the AI glasses, the display glasses
0:03:28 that we’ll have in five years will be good.
0:03:31 Some of them will be super high-end
0:03:32 and pretty exceptional.
0:03:34 Some of them will be, like, actually little
0:03:37 and, like, not even tremendously high-resolution displays,
0:03:38 but they will be, like, always available
0:03:39 and on your face.
0:03:41 I wouldn’t be doing work there,
0:03:44 but, like, if I’m just trying to grab simple content
0:03:45 in moments between, it’s pretty good for that.
0:03:47 So I think what we are seeing is,
0:03:50 as you’d expect, we’re at the very beginning now
0:03:53 of a spectrum of super high-end
0:03:54 but probably very expensive experiences
0:03:57 that will not be evenly distributed across the population.
0:03:57 Yeah.
0:03:59 A much more broadly available set of experiences
0:04:01 that are, they’re not really rich enough
0:04:03 to replace, like, the devices that we have today.
0:04:05 And then hopefully a continually growing number
0:04:07 of people who are having experiences
0:04:10 that really could not be had any other way today.
0:04:12 You know, thinking about what you could do
0:04:13 with mixed reality and virtual reality.
0:04:13 Yeah.
0:04:15 We’re going to build up to a lot of that stuff.
0:04:17 So throughout your career,
0:04:20 I would say one of the observations I would have
0:04:22 is you’ve been uniquely good
0:04:26 at piecing together various big technology shifts
0:04:28 into new product experiences.
0:04:31 So in the case of Facebook, early days for you,
0:04:34 obviously you famously were part of the team
0:04:35 that created the news feed.
0:04:35 Yeah.
0:04:37 And that’s a combination of social media,
0:04:38 a mobile experience,
0:04:41 and applying your, like, old-school AI.
0:04:42 Yeah, that’s right.
0:04:42 To it.
0:04:43 The old-school AI.
0:04:43 Yeah, exactly.
0:04:44 But that’s pretty cool.
0:04:46 And, like, a lot of times these trends,
0:04:47 they come in bunches.
0:04:47 Yeah.
0:04:49 And that’s what creates the breakthrough products.
0:04:53 So maybe take that and apply it to where we are today
0:04:55 with the major trends that are in front of you.
0:04:56 Let me say two things about this.
0:04:58 The first one is I think if there was a thing that,
0:04:59 not me specifically,
0:05:02 but I think me and my cohorts at Meta were really good at,
0:05:04 was, like, we really immersed in, like, what the problem was.
0:05:05 Like, what were people trying to do?
0:05:06 What did they want to do?
0:05:08 And when you do that,
0:05:12 you are going to reach for whatever tool is available
0:05:13 to accomplish that goal.
0:05:15 That allows you to be really honest about
0:05:18 what tools are available and see trends.
0:05:21 I think the more oriented you are towards the technology side,
0:05:24 you get caught in a wave of technology,
0:05:25 and you don’t want to admit when that wave is over
0:05:27 and you don’t want to embrace the next wave.
0:05:29 And you’re building technology for technology’s sake.
0:05:29 Yeah, yeah.
0:05:31 So, like, solving a product problem.
0:05:31 But if you’re embracing, like,
0:05:34 what are the issues that people are really going through in their life
0:05:35 and they don’t have to be profound,
0:05:37 I bring that up just because I think we’re in this interesting moment
0:05:40 where I think all of us have been through a phase
0:05:43 where a lot of people wanted a new wave to be coming
0:05:45 because it would have been advantageous to them.
0:05:45 Yeah.
0:05:48 But those things weren’t solving problems that regular people had.
0:05:50 I think the reason we’re so enthusiastic about
0:05:53 the AI revolution that’s happening right now
0:05:56 is it really feels tangible.
0:05:58 These are real problems that are being solved.
0:05:59 And it’s not solving every problem.
0:06:00 It creates new problems.
0:06:01 It’s fine.
0:06:04 So it feels like a substantial real NUCA capability that we have.
0:06:09 And what’s unusual about it is how broad-based it can be applied.
0:06:12 And while it has these interesting downsides today on factuality
0:06:15 and certainly compute in cost and inference,
0:06:18 those types of trade-offs feel really solvable
0:06:20 and the domains that it applies to are really broad.
0:06:22 And that’s very unusual.
0:06:23 Certainly in my career,
0:06:25 you almost always, when these technological breakthroughs happen,
0:06:27 they’re almost always very domain-specific.
0:06:28 It’s like, cool, this is going to get faster
0:06:31 or that’s going to get cheaper or that’s now possible.
0:06:33 This kind of feels like, oh, everything’s going to get better.
0:06:36 Every single interface that I interact with,
0:06:40 every single problem space that I’m trying to solve
0:06:43 are going to be made easier by virtue of this new technology.
0:06:44 That’s pretty rare.
0:06:47 Mark and I always believed that this AI revolution was coming.
0:06:48 We just thought it was going to take longer.
0:06:48 Yeah.
0:06:50 We thought we were probably still 10 years away at this point.
0:06:51 Yeah.
0:06:52 But what we thought would happen sooner
0:06:54 was this revolution in computing interfaces.
0:06:59 And we really started to feel 10 years ago
0:07:02 like the mobile phone form factor, as amazing as it was,
0:07:05 this is 2015, was like already saturated.
0:07:06 That was what it was going to be.
0:07:09 And once you get past the mobile phone,
0:07:10 which is, again, the greatest computing device
0:07:12 that any of us have ever used to this point,
0:07:14 of course, it’s like, okay, well, it has to be more natural
0:07:18 in terms of how you’re getting information into your body,
0:07:20 which is obviously ideally usually through our eyes and ears,
0:07:24 and how we’re getting our intentions expressed back to the machine.
0:07:25 You no longer have a touchscreen.
0:07:26 You no longer have a keyboard.
0:07:29 So once you like realize those are the problems,
0:07:31 it’s like, cool, we need to be on the face
0:07:33 because you need to have access to eyes and ears
0:07:35 to bring information from the machine to the person.
0:07:37 And you need to have these neural interfaces
0:07:40 to try to allow the person to manipulate the machine
0:07:41 and express their intentions to it
0:07:43 when they don’t have a keyboard or mouse or a touchscreen.
0:07:47 And so that has been an incredibly clear-eyed vision
0:07:49 we’ve been on for the last 10 years.
0:07:53 But we really did grow up in an entire generation of engineers
0:07:56 for whom the system was fixed.
0:07:57 The application model was fixed.
0:07:59 The interaction design.
0:08:02 Sure, we went from a mouse to a touchscreen,
0:08:04 but it’s still a direct manipulation interface,
0:08:06 which is literally the same thing that was pioneered in the 1960s.
0:08:09 So we really haven’t changed these modalities.
0:08:11 And there’s a cost to changing those modalities
0:08:14 because we as a society have learned
0:08:18 how to manipulate these digital artifacts through these tools.
0:08:20 So the challenge for us was,
0:08:22 okay, you have to build this hardware,
0:08:24 which has to do all these amazing things
0:08:27 and also be attractive and also be light
0:08:28 and also be affordable.
0:08:31 And none of these existed before.
0:08:32 And what I tell my team all the time is like,
0:08:34 that’s only half the problem.
0:08:37 The other half of the problem is, great, how do I use it?
0:08:39 Like, how do I make it feel natural to me?
0:08:41 I’m so good with my phone now.
0:08:44 It’s an extension of my body, of my intention at this point.
0:08:47 How do we make it even easier?
0:08:49 And so we were having these challenges.
0:08:51 And then, what a wonderful blessing.
0:08:54 AI came in two years ago, much sooner than we expected.
0:08:56 And it’s a tremendous opportunity
0:08:58 to make this even easier for us.
0:08:59 Because the AIs that we have today
0:09:01 have a much greater ability to understand
0:09:02 what my intentions are.
0:09:04 I can give vague reference,
0:09:06 and it’s able to work through the corpus of information
0:09:09 it has available to make specific outcomes happen from it.
0:09:11 There’s still a lot of work to be done
0:09:13 to actually adapt it.
0:09:15 And it’s still not yet a control interface.
0:09:17 Like, I can’t reliably work my machine with it.
0:09:19 There’s a lot of things that we have to do.
0:09:21 We know what those things are.
0:09:24 And so, now you’re in a much more exciting place, actually.
0:09:25 Whereas before, we thought, okay,
0:09:28 we’ve got this big hill to climb on the hardware.
0:09:30 We’ve got this big hill to climb on the interaction design.
0:09:31 But we think we can do it.
0:09:33 And now we’ve got a wonderful tailwind,
0:09:35 where on the interaction design side, at least,
0:09:39 there’s the potential of having this much more intelligent agent
0:09:43 that now has not only the ability for you
0:09:46 to converse with it naturally and get results out of it,
0:09:50 but also to know by context what you’re seeing,
0:09:51 what you’re hearing, what’s going on around you.
0:09:51 Yeah.
0:09:54 And make intelligent inference based on that information.
0:09:56 Let’s talk about, like, reality labs
0:09:58 and this suite of products, what it is today.
0:10:01 So, you have Quest headsets, you have the smart glasses,
0:10:03 and then on the far end of the spectrum is Orion
0:10:05 and some of the stuff that I demoed.
0:10:07 So, just talk about the evolution of those efforts
0:10:10 and what you think the markets are for them
0:10:12 and how they converge versus not over time.
0:10:14 So, when we started the Ray-Ban Meta Project,
0:10:16 they were going to be smart glasses.
0:10:18 And, in fact, they were entirely built,
0:10:20 and we were six months away from production
0:10:22 when Llama 3 hit.
0:10:24 And the team was like, no, we got to do this.
0:10:25 And so, now they’re AI glasses, right?
0:10:27 Like, they didn’t start as AI glasses,
0:10:28 but the form factor was already right.
0:10:30 We could already do the compute.
0:10:31 We already had the ability.
0:10:32 So, yeah, now you have these glasses
0:10:33 that you can ask questions to.
0:10:36 And, in December, to the early access program,
0:10:37 we launched what we call Live AI.
0:10:39 So, you can start a Live AI session
0:10:41 with your Ray-Ban Meta glasses,
0:10:43 and for 30 minutes until the battery runs out,
0:10:44 it’s seeing what you’re seeing.
0:10:45 Yeah.
0:10:46 And it’s funny because, on paper,
0:10:49 the Ray-Ban Meta looks like an incremental improvement
0:10:50 to Ray-Ban Stories.
0:10:52 And this is kind of the story I’m trying to tell,
0:10:54 which is, the hardware isn’t that different
0:10:55 between the two,
0:10:58 but the interactions that we enable
0:11:00 with the person using it
0:11:02 are so much richer now.
0:11:03 When you use Orion,
0:11:05 when you use the full AI glasses,
0:11:08 you can imagine a post-phone world.
0:11:09 You’re like, oh, wow.
0:11:11 Like, if this was attractive enough
0:11:12 and light enough
0:11:13 and had battery life enough
0:11:14 to wear all day,
0:11:15 this would have all the stuff I need.
0:11:17 Like, it would all be right here.
0:11:18 And when you start to combine that
0:11:20 with the images that we have
0:11:21 of what AI is capable of.
0:11:21 So, you did the demo
0:11:23 where we showed you the breakfast.
0:11:24 Yeah, it did.
0:11:25 And it’s, yeah,
0:11:25 and for what it’s worth,
0:11:26 I mean, I’ll explain it
0:11:27 because it’s very cool.
0:11:28 Got to walk over
0:11:30 and there’s a bunch of breakfast ingredients laid out.
0:11:32 And I look at it
0:11:33 and I say,
0:11:33 hey, Meta,
0:11:35 what are some recipes?
0:11:35 That’s right.
0:11:36 And these ingredients.
0:11:38 So, that is, for me at least,
0:11:39 when we think about Orion,
0:11:41 initially,
0:11:43 it didn’t have that AI component
0:11:44 when we first thought about it.
0:11:45 It had this component
0:11:47 that was very direct manipulation.
0:11:48 So, it was very much modeled
0:11:49 on the app model
0:11:49 that we’re all familiar with.
0:11:50 Yeah, of course.
0:11:51 And I think there’s a version of that.
0:11:52 Yeah, of course,
0:11:53 you’re going to want to do calls
0:11:53 and you’re going to want to be able
0:11:54 to do your email
0:11:56 and be able to do your texting
0:11:57 and you want to be able to play games.
0:11:58 We have to play our Stargazer game
0:12:00 and you want to do your Instagram reels.
0:12:01 What we’re now excited about
0:12:01 is, okay,
0:12:02 take all those pieces
0:12:04 and layer on the ability
0:12:07 to have an interactive assistant
0:12:08 that really understands
0:12:10 not just what’s happening
0:12:11 on your device
0:12:13 and what email’s coming in,
0:12:14 but also what’s happening
0:12:16 in the physical world around you
0:12:17 and is able to connect
0:12:19 what you need in the moment
0:12:20 with what’s happening.
0:12:21 And so, these are concepts
0:12:21 where you’re like,
0:12:23 wow, what if the entire app model
0:12:23 is upside down?
0:12:24 What if it isn’t like,
0:12:25 hey, I want to go fetch
0:12:26 Instagram right now.
0:12:26 It’s like, hey,
0:12:28 the device realizes
0:12:28 that you have a moment
0:12:29 between meetings,
0:12:30 you’re a little bit bored.
0:12:30 Hey, do you want to catch up
0:12:31 on the latest highlights
0:12:33 from your favorite basketball team?
0:12:34 Those things become possible.
0:12:35 Having said that,
0:12:36 the hardware problems are hard
0:12:36 and they’re real
0:12:37 and the cost problems
0:12:38 are hard and they’re real.
0:12:39 And come for the king,
0:12:40 you best not miss.
0:12:41 The phone is an incredible
0:12:43 centerpiece of our lives today.
0:12:45 It’s how I operate my home.
0:12:46 I use it in my car.
0:12:46 I use it for work.
0:12:47 It’s everywhere, right?
0:12:50 And the world has adapted
0:12:51 itself to the phone.
0:12:53 So, it’s weird that my ice maker
0:12:53 has a phone app,
0:12:54 but it does.
0:12:54 Like, I don’t know.
0:12:55 I’m not sure.
0:12:56 It seems excessive,
0:12:57 but like,
0:12:58 so somebody today
0:12:58 who’s like,
0:12:59 I got to make an ice maker,
0:13:00 number one job,
0:13:01 got to have an app.
0:13:03 It’s like the smart refrigerator.
0:13:04 You’re like,
0:13:04 I don’t need this.
0:13:05 Take it out of me.
0:13:06 I do think it’s going
0:13:07 to be a long,
0:13:07 that’s why I said
0:13:09 the 10-year view for me
0:13:10 is, I think, much clearer.
0:13:11 I think these things
0:13:13 are going to be available,
0:13:14 widely accepted,
0:13:16 increasingly adopted.
0:13:17 The five-year view is harder
0:13:18 because, man,
0:13:19 like, even if it’s amazing.
0:13:20 Knocking out the dominance
0:13:21 of the phone in five years,
0:13:22 it just seems so hard.
0:13:22 It seems unthinkable.
0:13:24 It’s unthinkable for us, right?
0:13:24 That’s why I said,
0:13:25 like, Orion was the first
0:13:27 time I thought me.
0:13:27 Orion, like,
0:13:28 putting that in my head,
0:13:28 I was like,
0:13:31 okay, it could happen.
0:13:32 Like, there does exist
0:13:33 a life for us as a species
0:13:34 past the phone.
0:13:34 Yeah.
0:13:35 Yeah, it still has
0:13:36 the whole dynamic of,
0:13:37 well, how do I envision
0:13:38 my life without the operating
0:13:39 system that I’m so accustomed to?
0:13:39 Totally.
0:13:40 So I see the physical stuff
0:13:41 that you do,
0:13:42 but just the familiarity
0:13:43 and all the stuff
0:13:44 that’s working in there.
0:13:45 So what do you think
0:13:47 of the interim period?
0:13:49 So maybe you get to the point
0:13:50 where the hardware is capable,
0:13:52 it is market accessible,
0:13:54 but do you tether
0:13:54 to the phone?
0:13:56 Do you take a strong view
0:13:57 that you will never do that
0:13:58 and let the product stand?
0:13:59 Like, how do you think
0:14:00 about that piece?
0:14:02 The phones have this huge
0:14:03 advantage and disadvantage.
0:14:04 Huge advantage,
0:14:05 which is like,
0:14:06 the phone is already central
0:14:07 to our lives.
0:14:08 It’s already got this huge
0:14:09 developer ecosystem.
0:14:10 It’s this anchor device,
0:14:11 and it’s a wonderful anchor
0:14:12 device for that.
0:14:13 The disadvantages,
0:14:14 I actually think what we found
0:14:17 is the apps want to be different
0:14:19 when they’re not controlled
0:14:20 via touchscreen.
0:14:22 And that’s not super novel.
0:14:23 A lot of people failed
0:14:24 early in mobile,
0:14:25 including us,
0:14:26 by just taking our web stuff
0:14:27 and putting it on
0:14:27 the mobile phone
0:14:28 and being like,
0:14:29 oh, the mobile phone,
0:14:30 we’ll just put the web there.
0:14:32 But because it wasn’t native
0:14:33 to what the phone was,
0:14:34 and I mean everything
0:14:36 from interaction design
0:14:38 to the actual design
0:14:39 to the layout
0:14:40 to how it felt,
0:14:41 because we weren’t doing
0:14:43 phone native things,
0:14:44 we were failing
0:14:45 with one of the most popular
0:14:46 products in the history
0:14:46 of the web.
0:14:48 It’s just like the major
0:14:49 design field,
0:14:50 like the skeuomorphic idea
0:14:51 versus the native idea.
0:14:52 Yeah, and I think
0:14:53 having the developers
0:14:54 is a true value,
0:14:54 and I think having all
0:14:55 this application functionality
0:14:56 is a true value.
0:14:58 but then once you actually
0:15:00 reproject it into space
0:15:01 and you’re manipulating it
0:15:04 with your fingers like this
0:15:05 as opposed to a touchscreen,
0:15:06 you have much less precision.
0:15:08 It doesn’t respond
0:15:08 to voice commands
0:15:10 because there’s no tools
0:15:11 for that.
0:15:11 There’s no design
0:15:12 integration for that.
0:15:14 So having a phone platform
0:15:16 today feels like,
0:15:17 wow, I’ve got this huge base
0:15:17 to work from
0:15:18 on the hardware side,
0:15:19 but I’ve also actually got
0:15:20 this kind of huge anchor
0:15:22 to drag on the software side.
0:15:24 And so we’re not opposed
0:15:24 to these partnerships,
0:15:25 and I think it’ll be interesting
0:15:26 to see once the hardware
0:15:27 is a little bit more developed
0:15:28 how partners feel about it.
0:15:30 And I hope they continue
0:15:32 to support people
0:15:33 who buy these phones
0:15:34 for $1,200, $1,300,
0:15:35 being able to bring
0:15:36 whatever hardware
0:15:37 they want to bring
0:15:38 and take the full functionality
0:15:39 of that with them.
0:15:41 The biggest question I have
0:15:42 is whether the entire app model,
0:15:43 because we were imagining
0:15:45 a very phone-like app model
0:15:46 for these devices,
0:15:47 admittedly a very different
0:15:48 interaction design,
0:15:50 input, and control schemes
0:15:50 are very different
0:15:51 and that demands
0:15:52 like a little extra
0:15:52 developer attention.
0:15:54 I am wondering if like
0:15:55 the progression of AI
0:15:56 over the next several years
0:15:57 doesn’t turn the app model
0:15:58 in its head.
0:15:59 Like right now,
0:16:00 it’s kind of an unusual thing
0:16:01 where I’m like,
0:16:03 I want to play music.
0:16:04 So in my head,
0:16:05 I translate that to
0:16:06 I have to go open Spotify
0:16:07 or open Tidal,
0:16:08 and the first thing I think of
0:16:09 is who is my provider
0:16:10 going to be?
0:16:11 Yeah, of course.
0:16:12 As opposed to like,
0:16:12 that’s not what I want.
0:16:13 It’s extremely limiting.
0:16:14 What I want is to play music.
0:16:14 Yes.
0:16:15 And I just want to be like,
0:16:16 go to the AI,
0:16:16 I’m like, cool,
0:16:18 play this music for me.
0:16:18 Yeah.
0:16:20 And it should know,
0:16:21 oh, like you’re already
0:16:23 using this service.
0:16:24 We’ll use that one.
0:16:25 Or these two services
0:16:26 are both available to you.
0:16:27 This one has a better quality song.
0:16:29 Or this one has lower latency,
0:16:29 whatever the thing is.
0:16:30 Or it’s like, hey,
0:16:31 the song you want
0:16:31 isn’t available
0:16:32 on any of these services.
0:16:33 Do you want to sign up
0:16:34 for this other service
0:16:34 that does have the song
0:16:35 that you want?
0:16:36 I don’t want to have
0:16:36 to be responsible
0:16:37 for orchestrating like
0:16:38 what app I’m opening
0:16:39 to do a thing.
0:16:40 We’ve had to do that
0:16:41 because that’s how
0:16:41 things were done
0:16:43 in the entire history
0:16:43 of digital computing.
0:16:45 You had an application-based model
0:16:46 that was the system.
0:16:47 So I do wonder
0:16:49 how much AI inverts things.
0:16:50 That’s a pretty hot take.
0:16:51 Yeah, that’s a hot take.
0:16:51 Inverts things.
0:16:53 And that’s not about wearables.
0:16:54 That’s not about anything.
0:16:54 That’s just like,
0:16:55 even at the phone level,
0:16:57 if you were building
0:16:57 a phone today,
0:16:59 would you build an app store
0:16:59 the way you historically
0:17:00 built an app store?
0:17:01 Or would you say like,
0:17:03 hey, you as a consumer,
0:17:04 express your intention.
0:17:05 Express what you’re
0:17:06 trying to accomplish.
0:17:07 And let’s like,
0:17:07 see what we have.
0:17:08 Let the system see
0:17:08 what it can produce.
0:17:09 Yeah, for you.
0:17:10 But I do think
0:17:11 if you were starting
0:17:12 from scratch today,
0:17:14 you probably wouldn’t build
0:17:15 this like app-centric world
0:17:16 where I, as a consumer,
0:17:17 I’m trying to solve a problem
0:17:18 and first have to decide
0:17:20 which of the providers
0:17:20 I’m going to use
0:17:21 to solve that problem.
0:17:21 Yeah, of course.
0:17:23 That’s fascinating.
0:17:24 And again,
0:17:25 I think it’s a function
0:17:26 of where the capabilities
0:17:27 are today
0:17:27 and I think where we have
0:17:28 line of sight
0:17:29 into orchestration capabilities.
0:17:30 Because I’d say,
0:17:31 knowledge-wise,
0:17:33 that is probably capable today.
0:17:35 I think orchestration-wise,
0:17:35 it’s probably
0:17:37 we’re a little bit away.
0:17:38 And then, of course,
0:17:39 you’ve got to build
0:17:40 the developer ecosystem
0:17:41 to develop on the platform.
0:17:42 Which is incredibly hard.
0:17:43 That’s the thing
0:17:43 I want to see
0:17:44 That’s the hardest piece, right?
0:17:45 That’s the hardest piece.
0:17:45 Yeah.
0:17:46 The stronger we get
0:17:48 at agentic reasoning
0:17:49 and capabilities,
0:17:50 the more I can rely
0:17:52 on my AI
0:17:53 to do things in my absence.
0:17:54 And at first,
0:17:55 it will be knowledge work,
0:17:55 of course.
0:17:56 That’s fine.
0:17:57 But once you have a flow
0:17:59 of consumers
0:18:00 coming through here,
0:18:00 what you’re going to find
0:18:01 is that they’re going to have
0:18:02 a bunch of dead ends.
0:18:02 Yeah.
0:18:03 Where they’re going to ask the AI,
0:18:05 hey, can you do this thing for me?
0:18:05 And it’s going to say,
0:18:06 no, I can’t.
0:18:09 That’s the goldmine
0:18:10 that you take to developers.
0:18:10 And you’re like,
0:18:13 hey, I’ve got 100,000 people a day
0:18:14 trying to solve this problem.
0:18:15 They’re trying to use your app.
0:18:15 Yeah.
0:18:16 They don’t know they are,
0:18:17 but they’re trying to use their app.
0:18:18 Look, here’s the query stream.
0:18:19 Here’s what’s coming through.
0:18:21 And we’re going to tell them no today.
0:18:23 If you build these hooks,
0:18:24 you’ve got 100,000 people
0:18:25 clamoring for something today.
0:18:27 Coming in for your service.
0:18:27 Yeah.
0:18:28 And it’s totally fine
0:18:30 for RAI to go back and say,
0:18:31 hey, you’ve got to pay for this.
0:18:32 There’s a guy who does this for you,
0:18:33 but you’ve got to pay for it.
0:18:34 Yeah.
0:18:34 And by the way,
0:18:35 I’m not just talking about apps.
0:18:38 There’s some kind of a marketplace here
0:18:41 that I think emerges over time.
0:18:43 So that’s how I see it playing out.
0:18:44 I don’t see it playing out
0:18:46 as like someone goes into a darkroom
0:18:47 and comes up with this app platform.
0:18:48 No.
0:18:49 What’s going to happen is
0:18:50 there’s going to become a query stream
0:18:53 of people using AI to do things.
0:18:55 And the AI will fail
0:18:57 repeatedly in certain areas
0:18:59 because that’s a type of functionality
0:19:00 that is currently behind
0:19:01 some kind of an app wall.
0:19:02 And there’s no…
0:19:04 Or it hasn’t been built native
0:19:06 to whatever consumption mechanism.
0:19:07 There’s no bridge that can be built.
0:19:07 Yeah, yeah, yeah.
0:19:08 And everyone wants to build the bridges.
0:19:08 They’re like, no, no.
0:19:10 It’s going to manipulate the pixels
0:19:12 and it’s going to manipulate…
0:19:13 It’s like, fine, it can do those things.
0:19:13 I’m not saying the AI
0:19:15 can’t cross those boundaries.
0:19:16 But I think over time,
0:19:18 that becomes the primary interface
0:19:21 for humans interacting with software
0:19:23 as opposed to the like
0:19:24 pick from the garden of applications.
0:19:25 Yeah, that makes a ton of sense.
0:19:29 That’s a very alluring end state
0:19:31 just as a consumer, right?
0:19:31 Yeah, it’s messy.
0:19:33 And I think it creates
0:19:34 these very exciting marketplaces
0:19:37 for functionality inside the AI.
0:19:39 It abstracts away
0:19:41 a lot of companies’ brand names,
0:19:42 which I think is going to be very hard
0:19:45 for an entire generation of brands.
0:19:45 Yeah.
0:19:47 Like the fact that I don’t care
0:19:48 if it’s being played
0:19:49 on one of these two music services,
0:19:51 that’s hard for those music services
0:19:54 who like really want me to care.
0:19:55 Yeah, yeah, yeah.
0:19:55 And like they want me
0:19:58 to have a stronger opinion about it.
0:19:58 And like they want me
0:19:59 to have an attachment.
0:19:59 Yeah.
0:20:00 I don’t want to have an attachment.
0:20:01 There are some things
0:20:02 where you may value the attachment
0:20:03 and you don’t whatever.
0:20:04 Yeah, in the world
0:20:04 where I’m like,
0:20:05 hey, there’s an app garden
0:20:06 and these two are competing
0:20:07 for my eyeballs,
0:20:09 the brand that they’ve built
0:20:11 is the hugely valuable asset.
0:20:13 In the world where
0:20:15 I just care if the song gets played
0:20:15 and sounds good,
0:20:17 a different set of priorities
0:20:18 are important.
0:20:20 I think that’s net positive
0:20:21 because what matters now
0:20:23 is performance on the job
0:20:23 being asked.
0:20:24 actual product experience
0:20:25 and value
0:20:26 and price per performance
0:20:27 like matters a lot.
0:20:28 Yeah.
0:20:29 I think a lot of companies
0:20:29 won’t love that.
0:20:31 Well, abstracting away,
0:20:33 that’s like effectively articulating,
0:20:35 abstracting away margin pools,
0:20:36 which puts a lot more pressure
0:20:39 on us trusting the AI
0:20:41 or the distributor of the AI.
0:20:42 And so far as I’m floating
0:20:43 between different companies
0:20:44 that are each providing AIs,
0:20:46 the degree which I trust them
0:20:47 to not be bought
0:20:48 and paid for in the back end,
0:20:49 they’re not giving me
0:20:50 the best experience
0:20:51 or the best price per money.
0:20:52 They’re giving the one
0:20:53 that gives them the most money.
0:20:54 Yeah, of course.
0:20:55 So, yeah, it’s the experience
0:20:56 of Google Search today, right?
0:20:57 It’s a very different world.
0:20:59 It’s a very different world.
0:21:00 It’s a very different world.
0:21:01 But you can actually see
0:21:03 inklings of it today, right?
0:21:04 So certain companies
0:21:05 are willing to work
0:21:06 with the new AI providers
0:21:08 in agentic task completion.
0:21:09 Yeah, yeah.
0:21:09 And then they’re like,
0:21:10 well, actually, wait a minute.
0:21:12 I don’t just want the bots
0:21:13 executing this stuff.
0:21:14 I want the humans coming to me.
0:21:15 I think I need that.
0:21:16 Yeah, right.
0:21:16 It’s existential
0:21:18 that I have this brand relationship
0:21:19 directly with the demand side.
0:21:20 Yeah.
0:21:22 So that’s potentially messy,
0:21:24 but a bright future,
0:21:24 especially if we don’t have
0:21:26 to pay that like brand tax.
0:21:27 Yeah, it’ll be very messy.
0:21:30 I don’t know it’s avoidable
0:21:31 because I think once consumers
0:21:33 start to get into these tight loops
0:21:35 where more and more
0:21:35 of their interactions
0:21:38 are being moderated by an AI,
0:21:39 you won’t have a choice.
0:21:40 That’s like where
0:21:41 your customers will be.
0:21:42 But it’s going to be
0:21:42 a pretty different world.
0:21:43 Yeah, it’ll be a different world
0:21:44 and there’ll probably be
0:21:45 some groups that try
0:21:46 to move fast to it.
0:21:47 as a way to compete
0:21:48 with things that are branded.
0:21:48 Yeah.
0:21:49 And just say,
0:21:49 I’m going to compete
0:21:50 on performance and price.
0:21:50 Yeah, that’s right.
0:21:51 Where do you think
0:21:52 that could potentially
0:21:53 happen first?
0:21:55 It probably will mirror
0:21:56 query volume.
0:21:57 I think of this a lot.
0:21:58 We do have a model of this,
0:22:00 which was in the web era
0:22:01 when Google became
0:22:03 the dominant search engine.
0:22:04 So before that,
0:22:05 the web era was like
0:22:07 very index based.
0:22:07 It was like Yahoo
0:22:09 and it was like links
0:22:11 and getting major sources
0:22:12 of traffic to link to you
0:22:12 was the game.
0:22:14 And then once Google
0:22:16 came to dominance,
0:22:17 which happened very quickly
0:22:18 over maybe a couple of years,
0:22:18 I feel like.
0:22:19 All that mattered
0:22:20 was like SEO.
0:22:21 All that mattered
0:22:21 was like where you were
0:22:22 in the query stream.
0:22:22 Yeah.
0:22:23 And the query stream
0:22:26 dictated what businesses
0:22:27 came over and succeeded.
0:22:28 Yeah.
0:22:29 Because like the queries
0:22:30 that were the most frequent,
0:22:31 those were the ones
0:22:32 that came first.
0:22:32 Yeah.
0:22:37 Travel came right away.
0:22:38 It was a huge disruption
0:22:39 and travel agents
0:22:40 went from a thing that existed
0:22:41 to a thing that didn’t exist
0:22:42 in a relatively short-
0:22:42 Immediately.
0:22:44 And they all competed
0:22:45 on the basis of like
0:22:46 execution of the best deal
0:22:47 in a seamless fashion
0:22:49 with the highest conversion.
0:22:50 I think SEO has gotten
0:22:51 to a point now
0:22:53 where it’s kind of a bummer.
0:22:54 It’s like made things worse.
0:22:55 It’s like everyone’s gotten so good.
0:22:56 It’s just like game.
0:22:57 Everyone’s gotten so good at it.
0:22:59 Especially with AI actually now.
0:22:59 That’s right.
0:23:00 So I actually think it’s like
0:23:01 we had this incredible
0:23:02 flattening curve
0:23:02 and now it’s like starting
0:23:03 to kind of rise up
0:23:03 in terms of-
0:23:05 Especially with paid placement too.
0:23:05 Yeah.
0:23:06 That’s so dominant.
0:23:07 So duh.
0:23:08 Yeah, that’s right.
0:23:09 And this is like probably
0:23:10 the cautionary tale
0:23:11 for how this plays out
0:23:12 in AIs as well.
0:23:13 I think there will be
0:23:16 a pretty good golden era here
0:23:17 where the query stream
0:23:18 will dictate
0:23:20 what businesses come first
0:23:21 because those are the queries
0:23:22 that are-
0:23:23 That’s the volume of people
0:23:24 unsatisfied with the existing
0:23:26 solutions that they have.
0:23:26 Yeah.
0:23:27 Otherwise they wouldn’t
0:23:28 be asking about it.
0:23:29 And product providers
0:23:30 and developers will follow that.
0:23:31 And build specifically-
0:23:32 solve those problems.
0:23:33 That’s right.
0:23:34 Once it tips
0:23:36 in each vertical
0:23:37 we get a lot of progress
0:23:37 very quickly
0:23:39 towards better solutions
0:23:39 for consumers.
0:23:41 And then once it hits
0:23:41 a steady state
0:23:42 it starts to be
0:23:43 gamesmanship.
0:23:43 Yeah.
0:23:44 And that’s the thing to fight.
0:23:46 And that’s decaying or-
0:23:47 That’ll be the true test of AI.
0:23:48 The true test.
0:23:48 Can it get through that?
0:23:50 Can it avoid falling into that trap?
0:23:51 Can it avoid that trap?
0:23:51 Yeah, yeah.
0:23:52 That’s right.
0:23:52 Exactly.
0:23:53 Well a lot of that
0:23:54 is business model driven
0:23:55 and we’ll see how that evolves
0:23:55 over time too.
0:23:56 That’s right.
0:23:57 You guys have also been
0:23:59 leading from the front
0:24:00 on this idea of open source.
0:24:01 Yeah.
0:24:02 And so talk about
0:24:03 some of your efforts
0:24:05 on that side of the business
0:24:05 and then
0:24:07 what is the ideal
0:24:08 market structure
0:24:10 of the AI model side
0:24:10 for you guys?
0:24:11 There’s two parts
0:24:12 that came together.
0:24:13 The first one is
0:24:14 Llama came out of FAIR
0:24:16 our fundamental AI research group
0:24:17 and that’s been
0:24:18 an open source
0:24:19 research group
0:24:20 since the beginning.
0:24:20 Yeah.
0:24:21 Since John LeCun came in
0:24:22 and they established that
0:24:24 it’s allowed us to attract
0:24:24 incredible researchers
0:24:26 who really believe
0:24:26 that we’re going to make
0:24:28 more progress as a society
0:24:29 working together
0:24:29 across boundaries
0:24:30 of individual labs
0:24:31 than not.
0:24:33 And to be fair
0:24:33 it’s not just us
0:24:34 obviously
0:24:36 the Transformer paper
0:24:36 was published at Google
0:24:37 and like you know
0:24:39 big self-supervised learning
0:24:39 was our contribution
0:24:40 like everyone’s contributing
0:24:41 to the knowledge base
0:24:43 but when we open source Llama
0:24:44 that’s how all models
0:24:45 were open source
0:24:45 at that point.
0:24:45 Yeah.
0:24:47 Of course.
0:24:48 Like everyone was open
0:24:49 the only thing that was unusual
0:24:50 was everything else
0:24:50 just went closed source
0:24:51 over time.
0:24:51 Yeah.
0:24:52 Effectively.
0:24:52 That’s right.
0:24:53 But before that
0:24:55 every time someone built a model
0:24:55 they open sourced it
0:24:56 so that other people
0:24:56 could use the model
0:24:57 and see how great
0:24:58 that model was.
0:24:58 Like that was like
0:24:59 mostly how it was done.
0:25:00 Sure.
0:25:01 If it was worth anything.
0:25:02 Certainly some specialized models
0:25:03 for translations and whatnot
0:25:04 were kept closed
0:25:05 but like if it was a general model
0:25:05 that was what was done.
0:25:06 Llama 2 was probably
0:25:08 the big decision point for us.
0:25:08 Llama 2
0:25:09 and this is where I think
0:25:10 the second thing
0:25:11 that came in
0:25:12 was a belief that I’ve had
0:25:12 that I was advancing
0:25:14 really strenuously internally
0:25:15 that Mark Ridley believes in too
0:25:16 and he’s written
0:25:16 his post about this
0:25:17 which is first of all
0:25:18 we’re going to make way more progress
0:25:19 if these models are open.
0:25:20 Yeah.
0:25:21 Because a lot of these contributions
0:25:22 aren’t going to come
0:25:24 from these big labs
0:25:24 like they’re going to come
0:25:25 from these little labs
0:25:26 and we’ve seen this already
0:25:26 with DeepSeq in China
0:25:28 which was put in a tough spot
0:25:29 and then innovated
0:25:31 incredibly in the memory architectures
0:25:31 and a couple other places
0:25:33 to really get amazing results.
0:25:34 And so we really believe
0:25:35 we’re going to get
0:25:36 the most progress collectively.
0:25:37 The second thing is
0:25:38 inside this piece is
0:25:39 you know this is a classic
0:25:40 I believe these are going to be commodities
0:25:42 and you want to commoditize
0:25:42 your complements.
0:25:43 Yes.
0:25:44 And we’re in a unique position
0:25:45 strategically where
0:25:46 our products are made better
0:25:47 through AI
0:25:48 which is why we’ve been
0:25:49 investing in it for so long.
0:25:50 Whether it’s recommendation systems
0:25:51 in what you’re seeing
0:25:52 in feed or reels
0:25:54 whether it’s simple things
0:25:55 like what friend
0:25:56 do I put at the top
0:25:57 when you type
0:25:57 you want to make a new message
0:25:58 who do I think
0:25:58 you’re going to message right now?
0:26:00 Little things like that
0:26:01 to really big expansive things
0:26:03 like hey here’s an entire answer
0:26:04 here’s an entire search interface
0:26:05 that we couldn’t do before
0:26:06 in WhatsApp
0:26:07 that like now
0:26:09 is a super popular surface.
0:26:10 So there’s all these things
0:26:11 that are possible for us
0:26:12 that are made better
0:26:13 by this AI
0:26:14 but nobody else
0:26:15 having this AI
0:26:16 can then build our product.
0:26:18 The asymmetry works in our favor.
0:26:18 Yeah of course.
0:26:19 And so for us
0:26:20 like commoditizing your complements
0:26:21 is just good business sense
0:26:22 and making sure that there is
0:26:24 a lot of competitively priced
0:26:26 if not almost free
0:26:27 models out there
0:26:29 helps the entire industry
0:26:31 helps a bunch of small startups
0:26:32 and academic labs
0:26:33 it also helps us.
0:26:34 Yeah you as the application provider
0:26:35 are huge beneficiaries.
0:26:36 So we’re all super aligned.
0:26:37 Yeah you’re aligned.
0:26:38 Business model alignment
0:26:38 and industry alignment.
0:26:40 It’s a strong alignment there.
0:26:40 Yes.
0:26:42 So it comes from both
0:26:43 this fundamental belief
0:26:44 in how this kind of research
0:26:45 should be done
0:26:46 and then aligns
0:26:47 with the other business model
0:26:48 and so there’s no conflict.
0:26:49 Yeah societal progress
0:26:50 plus business model alignment.
0:26:51 It’s all together.
0:26:52 It’s all great.
0:26:53 It’s all going the same direction.
0:26:53 That’s awesome.
0:26:54 It’s great.
0:26:55 I want to shift gears
0:26:56 to talking about
0:26:58 the impediments to progress
0:26:59 and like what you think
0:27:00 you know are kind of
0:27:01 linear versus not.
0:27:03 So the risks to the vision
0:27:04 to the overall vision
0:27:05 that you articulated
0:27:06 obviously hardware
0:27:07 Yep.
0:27:08 AI capabilities.
0:27:08 Yep.
0:27:10 Vision capabilities
0:27:11 and screens
0:27:12 and all that
0:27:12 resolutions.
0:27:14 We talked about the ecosystem
0:27:15 and developers
0:27:17 and native products.
0:27:18 So maybe just talk about
0:27:18 what you see
0:27:19 are kind of
0:27:20 the linear path things
0:27:22 and the things
0:27:22 that may be harder
0:27:23 or riskier.
0:27:26 we have real invention risk.
0:27:27 There exists risk
0:27:28 that the things
0:27:28 that we want to build
0:27:30 we don’t have the capacity
0:27:31 to build as a society
0:27:32 as a species yet.
0:27:33 Yeah.
0:27:34 And that’s not a guarantee.
0:27:36 I think we have windows to it.
0:27:37 You’ve seen Orion
0:27:38 so like it can be done.
0:27:38 Yeah there’s probably.
0:27:39 Yeah it feels like
0:27:40 it’s a cost reduction exercise
0:27:42 it’s a materials improvement exercise
0:27:43 but it can be done.
0:27:44 There is still some invention risk.
0:27:46 Far bigger than the invention risk
0:27:47 I think is the adoption risk.
0:27:48 Is it considered socially acceptable?
0:27:50 Are people willing
0:27:51 to learn a new modality?
0:27:52 Like we all learned to type
0:27:53 when we were kids at this point.
0:27:54 We were born with phones
0:27:55 in our hands at this point.
0:27:55 Yeah.
0:27:56 Are people willing to learn
0:27:57 a new modality?
0:27:57 Is it worth it to them?
0:27:58 Ecosystem risk
0:27:59 even bigger than that.
0:28:00 Like great you build this thing
0:28:02 but if it just does like
0:28:03 your email and reels
0:28:04 that’s probably not enough.
0:28:05 Do people bring
0:28:06 the suite of software
0:28:07 that we require
0:28:08 to interact with
0:28:09 modern human society
0:28:11 to bear on the device?
0:28:13 Those are all huge risks.
0:28:14 I will say
0:28:15 we feel pretty good
0:28:16 about where we’re getting
0:28:17 on the hardware
0:28:18 on acceptability.
0:28:19 We think we can do
0:28:20 those things.
0:28:20 That was not a guarantee
0:28:21 before I think
0:28:23 with the Ray-Van Metaglasses
0:28:23 we’re feeling like
0:28:24 okay we can get through
0:28:25 You feel like the acceptability
0:28:26 Humans will accept
0:28:28 that I’m using technology.
0:28:29 Within that
0:28:31 super interesting
0:28:32 regulatory challenges
0:28:33 here I have
0:28:34 an always on machine
0:28:35 that gives me
0:28:36 super human sensing.
0:28:37 My vision is better.
0:28:38 My hearing is better.
0:28:39 My memory is better.
0:28:40 That means
0:28:42 when I see you
0:28:43 a couple years from now
0:28:44 and I haven’t seen you
0:28:44 on the internet
0:28:45 I’m like
0:28:45 oh god I don’t remember
0:28:47 we did a podcast together
0:28:47 what’s the guy’s name?
0:28:49 Can I ask that question?
0:28:49 Am I allowed
0:28:50 to ask that question?
0:28:50 Yes.
0:28:51 What is your right
0:28:53 it’s your face
0:28:54 you showed me your face
0:28:55 and if I was somebody
0:28:55 with a better memory
0:28:57 I could remember the face
0:28:59 so like that happened
0:29:00 but I don’t have
0:29:00 a great memory
0:29:01 so am I allowed
0:29:02 to use a tool
0:29:03 to assist me or not?
0:29:04 So there’s really
0:29:05 subtle regulatory
0:29:06 privacy
0:29:07 social
0:29:07 acceptability
0:29:08 questions
0:29:08 that are like
0:29:09 embedded here
0:29:10 that are super deep
0:29:10 individually
0:29:12 and can derail
0:29:13 the whole thing
0:29:14 like you can easily
0:29:14 derail
0:29:15 easily derail
0:29:16 the whole thing
0:29:17 and slow progress
0:29:17 that’s the thing
0:29:18 I think we sometimes
0:29:20 think in our industry
0:29:21 it’s like feel the dreams
0:29:21 if you build it
0:29:22 they will come
0:29:22 and it’s like
0:29:23 no a lot of things
0:29:24 have to happen right
0:29:25 well you can also
0:29:26 overstep too
0:29:27 that’s the risk
0:29:28 you’re sure you get
0:29:28 your hands locked
0:29:29 great technology
0:29:31 can get derailed
0:29:32 for long periods
0:29:32 of time
0:29:33 nuclear power
0:29:34 got derailed
0:29:35 yeah for absolutely
0:29:36 stupid reasons
0:29:37 for 70 years
0:29:38 for bad reasons
0:29:39 we know we’re bad now
0:29:40 and it was like
0:29:41 they just played it wrong
0:29:42 yeah of course
0:29:42 and they were like
0:29:43 ah ignore this
0:29:43 it’s like no
0:29:45 people actually feel this way
0:29:46 so I think yeah
0:29:47 I feel pretty good
0:29:47 about the invention risk
0:29:48 acceptability risk
0:29:49 is looking better
0:29:50 than it has been
0:29:51 but like I think
0:29:51 there’s still a lot
0:29:53 of big hedges
0:29:53 to cross there
0:29:55 I actually think
0:29:56 the ecosystem risk
0:29:56 was one
0:29:57 I would have said
0:29:57 previously
0:29:58 was the biggest one
0:30:00 but AI is now
0:30:01 my potential
0:30:02 silver bullet there
0:30:03 if AI becomes
0:30:04 the major interface
0:30:05 then it comes for free
0:30:06 yeah
0:30:07 and I will also say
0:30:08 that we’ve had
0:30:09 such a positive response
0:30:10 from even just
0:30:11 set aside Orion
0:30:12 even the Ray-Ban Metas
0:30:14 companies that want
0:30:14 to work for us
0:30:15 and build on that platform
0:30:16 it’s not a platform yet
0:30:17 yeah it’s not
0:30:17 there’s so little
0:30:18 there’s so little compute
0:30:19 there’s so little compute
0:30:20 we just connect an app
0:30:21 we literally don’t
0:30:22 have any space yet
0:30:22 yeah
0:30:23 but we did do a partnership
0:30:24 with Be My Eyes
0:30:25 which like helps blind
0:30:26 and hard of vision
0:30:26 people navigate
0:30:28 and it’s really spectacular
0:30:29 and so there’s a little window
0:30:29 there where we can start
0:30:30 building
0:30:30 so yeah
0:30:31 I would say the response
0:30:32 has been more positive
0:30:33 than I had expected
0:30:35 so everything right now
0:30:36 tailwinds abound
0:30:36 right now
0:30:37 and to be honest
0:30:38 after eight years
0:30:40 of nine years
0:30:42 of headwinds
0:30:43 having a year of tailwinds
0:30:43 is nice
0:30:43 yeah
0:30:44 I’ll take it
0:30:44 I’ll take it
0:30:45 I’m not gonna look
0:30:45 in the face
0:30:46 no victory laps
0:30:47 yeah but that’s good
0:30:47 okay
0:30:48 but it’s all hard
0:30:50 at every point
0:30:50 it could all fail
0:30:51 yeah I like that you
0:30:51 just started with
0:30:52 it’s invention risk
0:30:53 it’s I don’t know
0:30:54 there’s many ways
0:30:55 this just won’t work
0:30:55 yeah that’s right
0:30:56 even if it does work
0:30:57 it might not take
0:30:59 well I’ll say two things
0:31:00 about this
0:31:01 and this is where
0:31:02 Mark just deserves
0:31:02 so much credit
0:31:04 is we’re true believers
0:31:06 like we have actual conviction
0:31:07 yeah
0:31:08 Mark believes
0:31:10 this is the next thing
0:31:12 it needs to happen
0:31:13 and it doesn’t happen
0:31:13 for free
0:31:15 like we can be the ones
0:31:15 to do it
0:31:16 our chief scientist
0:31:17 Michael Arash
0:31:18 who’s one of my favorite people
0:31:18 I’ve ever gotten a chance
0:31:19 to work with
0:31:20 he talks a lot about
0:31:22 the myth of technological eventualism
0:31:23 it doesn’t eventually happen
0:31:24 there’s a lot of people in tech
0:31:25 who are like
0:31:26 yeah AR will eventually happen
0:31:27 that’s not how it fucking works
0:31:28 that would not
0:31:28 that would actually
0:31:30 AR is a specific one
0:31:30 that would just absolutely not
0:31:32 you have to stop
0:31:33 and put the money
0:31:34 and the time
0:31:34 and do it
0:31:36 somebody has to stop
0:31:36 and do it
0:31:37 and that is the difference
0:31:38 the number one thing I’d say
0:31:39 is like the difference
0:31:40 between us and anybody else
0:31:41 is we believe in this stuff
0:31:42 in our cores
0:31:44 this is the most important work
0:31:45 I’ll ever get a chance to do
0:31:46 this is Xerox PARC level
0:31:48 new stuff
0:31:49 where we’re rethinking
0:31:50 how humans are going to interact
0:31:50 with computers
0:31:52 it’s like JCR Licklider
0:31:53 and the human in the loop computing
0:31:54 we’re seeing that with AI
0:31:55 it’s a rare moment
0:31:56 it’s a rare moment
0:31:57 it doesn’t even happen
0:31:58 once a generation I think
0:31:59 it may happen every other generation
0:31:59 every third generation
0:32:00 like you don’t get a chance
0:32:01 to do this all the time
0:32:03 so we’re not missing it
0:32:03 we’re just like
0:32:04 we’re going to do it
0:32:05 and we may fail
0:32:06 like it’s possible
0:32:07 but we will not fail
0:32:09 for lack of effort or belief
0:32:09 great
0:32:10 thanks a ton boss
0:32:11 cheers
0:32:11 yeah cheers
0:32:20 and we’ll see you in the next video
0:32:21 we’ll see you in the next video
0:32:22 we’ll see you in the next video
0:32:23 we’ll see you in the next video
0:32:23 we’ll see you in the next video
0:32:23 we’ll see you in the next video
0:32:24 we’ll see you in the next video
0:32:24 we’ll see you in the next video
0:32:24 we’ll see you in the next video
0:32:25 we’ll see you in the next video
0:32:25 we’ll see you in the next video
0:32:26 we’ll see you in the next video
0:32:26 we’ll see you in the next video
0:32:27 we’ll see you in the next video
0:32:27 we’ll see you in the next video
0:32:28 we’ll see you in the next video
0:32:28 we’ll see you in the next video
0:32:29 we’ll see you in the next video
0:32:30 we’ll see you in the next video
Are we nearing the end of the smartphone era?
In this episode, a16z Growth General Partner David George talks with Meta CTO Andrew “Boz” Bosworth about what comes after apps and touchscreens. From smart glasses to AR headsets, Boz shares how AI is powering a new wave of computing—one that’s ambient, agentic, and driven by human intent.
They explore what it takes to build for this future, the risks of changing interaction models, and why the next big platform shift may already be in motion.
This episode is part of our AI Revolution series, where we explore how industry leaders are leveraging generative AI to steer innovation and navigate the next major platform shift. Discover more insights and content from the AI Revolution series at a16z.com/AIRevolution.
Resources:
Find Boz on X: https://x.com/boztank
Find David on X: https://x.com/davidgeorge8
Stay Updated:
Let us know what you think: https://ratethispodcast.com/a16z
Find a16z on Twitter: https://twitter.com/a16z
Find a16z on LinkedIn: https://www.linkedin.com/company/a16z
Subscribe on your favorite podcast app: https://a16z.simplecast.com/
Follow our host: https://twitter.com/stephsmithio
Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.