0:02 [Music]
0:04 So Nepai, you're the CEO of Alphabet and
0:06 Google. Welcome back to the coder. Well,
0:08 good to be back. Feels like a nice
0:10 tradition post IO to be talking to you.
0:12 So good to be back. I think this is the
0:14 third year we've done this after IO. I'm
0:15 excited. Thank you for keeping the
0:18 tradition alive. Lots to talk about. You
0:20 announced a lot of things yesterday
0:22 during the keynote. Uh there's AI mode
0:24 rolling out for US users. Big updates to
0:27 Gemini. There's V3 and Imagine the
0:29 generators. You solved Pokemon with
0:31 robots, which is very exciting. You
0:33 know, my takeaway yesterday was that
0:34 Google feels very confident now. There's
0:37 there's a a real confidence about the
0:38 technology coming to life and the
0:41 products. A lot of things are shipping
0:44 imminently. What's the one piece that
0:46 gave you that confidence? Is it just the
0:47 volume of things that are shipping? Is
0:50 it one technology that clicked into
0:51 being ready for consumers? Where where
0:54 is it coming from? Look, I think it
0:55 comes from
0:59 the depth and breadth of uh of the AI
1:01 frontier. We are pushing like at a in a
1:03 in a more fundamental and foundational
1:05 way. We spoke a lot about this theme
1:07 called research becomes reality. But it
1:10 is you know always felt we are a deep
1:12 computer science company and you know
1:14 we've been AI first for a while. So
1:16 putting all that together and and
1:18 bringing it to our products at the at
1:20 the depth and breadth is what I think
1:23 it's really pleasing to see. for example
1:25 like you know people may not have
1:27 noticed it much it was so quick we spoke
1:29 about text diffusion models in the
1:31 middle of the whole thing but you know
1:33 but we're pushing the frontier on all
1:36 dimensions right and so de spoke about
1:38 world models so I think uh to me that's
1:41 the exciting part like how deep we are
1:42 pushing this frontier and then bringing
1:44 it to users and so maybe that's what
1:47 that's what makes it feel that way you
1:48 you mentioned research into reality
1:50 several times obviously a lot of these
1:51 projects have been cooking in labs for a
1:54 long Um, you've said many times you
1:55 think AI will be as profound as
1:57 electricity. Yes. Over the past many,
1:58 many years that you and I have talked
2:00 about it, but you said something
2:02 yesterday that I think adds to that,
2:04 which is that we're in a new phase of
2:06 the platform shift, right? And people
2:08 have talked about AI being a platform
2:09 shift for quite a while. That always has
2:11 meant to me that there's a user
2:13 interface platform shift coming, right?
2:14 We're going to interact with computers
2:15 in natural language and more natural
2:17 ways and they will interact with us back
2:18 in that same way and everything will
2:21 change. Is that the platform shift?
2:24 Yeah, there are you know you are right
2:26 uh each of these platform shifts uh you
2:29 know changes many things on the IO front
2:31 uh nothing to do with Google IO just IO
2:33 in the traditional uh computer science
2:35 sense you know you could feel it
2:37 yesterday when I watched the Android XR
2:38 I've used them and played around with
2:41 them but watching it to people talking
2:43 in different languages you can envision
2:45 the future one day where it'll actually
2:46 be seamless in a way you couldn't have
2:48 done it with phones you couldn't have
2:50 done it without AI because there's
2:51 nothing in your way you're looking at
2:54 the other person and talking right and
2:57 so that is an element of platform shift
2:59 but there are many more elements right
3:01 this is the only platform where I think
3:05 the actual platform is over time capable
3:08 of creating and self-improving and so on
3:10 in a way we could have never talked
3:12 about any other platform before so
3:13 that's why I think it's much more
3:16 profound than the other platform shifts
3:19 it'll allow people to create new things
3:21 in a way uh you know because at each
3:23 layer of the stack there's going to be
3:26 profound improvements and so I think
3:28 that virtuous cycle you get uh in terms
3:31 of how you can unleash this creative
3:33 power to all of society be it software
3:36 engineers be it creators I think that is
3:38 in going to happen in a a much more
3:40 multiplicative way and that that you
3:42 know so when I say it's a next phase I'm
3:44 talking about that part too let me just
3:45 put make that more concrete for people I
3:48 think the last platform shift we all
3:50 understand is the shift to
3:52 That's right. And that was right. We're
3:53 going to have multi-touch. We're going
3:55 to have faster cell connections. We're
3:57 going to increase processing power that
3:58 can go with you everywhere. And then
4:00 there was a layer of applications that
4:02 was enabled by all of those things. You
4:04 can push a button and a Toyota Camry
4:05 will show up wherever you are in the
4:06 world is like a very powerful thing that
4:09 required all of those ideas. Well, how
4:11 would you describe the phase we're in
4:14 now compared to that? The phase of this
4:16 because the first phase of AI was the
4:17 transformers work and the models work
4:19 and we can all see this capability.
4:21 Yeah. The second phase what what is it
4:23 to you?
4:27 Just just imagine um you know like the
4:29 when when the internet came blogging
4:32 became a thing. You know pre- internet
4:35 very few people had a means by which
4:37 they could put out their thoughts out to
4:38 the world right you know with the
4:41 internet came a new medium. It allowed
4:42 people to create and express themselves
4:45 in a new way. With mobile cameras and
4:47 you could shoot and like you know you
4:49 could create videos. um you know look at
4:52 what what's happened with YouTube. For
4:54 me the similar part of this is you know
4:56 we're all talking about things like VIP
5:00 coding yesterday you saw V3 right so we
5:02 are now in that phase I think people are
5:04 going to be able to create AI
5:07 applications you can call it but you
5:09 know VIP coding there are many names to
5:12 it but that that power is yet to be
5:14 unleashed we're barely scratching the
5:18 surface right and these models are now
5:20 you know they aren't quite there you can
5:22 you can kind of do oneshot coding but
5:24 you really need to know be a programmer
5:26 to kind of go iterate and create
5:28 something with polish right but that
5:31 frontier is evolving pretty rapidly so I
5:32 think we you're going to see a new wave
5:34 of things just like mobile did just the
5:37 internet did uh you know I came to go
5:40 Google at the time when it was Ajax was
5:42 the revolution uh the fact that the web
5:44 became dynamic you know you had things
5:47 like Google maps flicker Gmail all that
5:49 suddenly came into existence right and
5:51 but I think AI is going to turbo charge
5:53 uh in a way we haven't seen before. It
5:55 feels like to me like what you're
5:57 describing is we're in the phase where
5:58 the products are developed, right? The
6:00 capabilities were the first phase and
6:01 now we're going to make some actual
6:03 products and and more people can build
6:04 products than ever before. That's the
6:06 multiplicative part I'm talking about.
6:08 Not just this platform helps you create
6:11 more products. The the process of
6:12 creating, developing, etc. is going to
6:15 be accessible to a much wider swath of
6:17 humanity than ever before. I'm wondering
6:19 when you look at the landscape of
6:20 products that exist now most people
6:24 experience AI in Gemini or chat as a
6:26 chatbot the general purpose interface to
6:28 a bunch of knowledge that will talk to
6:30 you what products do you see that will
6:32 have the same kind of impact as the web
6:33 two products you were talking about
6:36 besides the general purpose chatbot you
6:39 know uh well obviously you've seen away
6:42 with uh coding ides right like you know
6:44 that entire landscape is I can't even
6:47 keep track of uh you know how many new
6:49 companies have come into and people are
6:51 using a lot of it right and yesterday we
6:53 showed a bunch of partners with whom we
6:55 are working so that's an area where
6:58 because coding is where maybe AI is
7:02 making the uh most progress you're
7:03 seeing the application layer at least in
7:06 terms of code uh code editors uh really
7:09 come into vogue we've had success with
7:11 notebook LM right we're launching
7:13 products like flow yesterday right flow
7:15 is a new product which allows you to
7:17 create and imagine So those are all
7:18 applications we are doing. Uh I think
7:20 others are beginning to do you know
7:22 people are working on like legal
7:24 assistance and there are all kinds of
7:27 startups uh you know I I was recently in
7:30 a doctor's office and they do have AI
7:31 kind of look at I mean transcribe the
7:33 whole thing put it all in reports and so
7:36 on that's at an enterprise application
7:37 layer kind of completely works different
7:40 than when I went to that visit 2 years
7:43 ago. So all that change is happening
7:45 across the board but I think we just in
7:46 the early stages. So you know you will
7:48 see it play out over the next three to
7:49 five years in a big way. Did you ask
7:51 your doctor what model their
7:53 transcription software is running? No I
7:55 didn't. No I didn't. Yeah. One of the
7:56 reasons I'm asking this and I'm pushing
8:00 on this is the amount of huge investment
8:02 in the capability from Google and others
8:04 has to pay off in some products that
8:06 return on that investment. Notebook LM
8:08 is great. I don't think it's going to
8:10 fully return on Google's data center
8:12 investment, let alone the investment in
8:15 pure AI research. Do you see a product
8:16 that can return on that investment at scale?
8:18 scale?
8:20 Look, do you think in 2004 if you had
8:23 looked at Gmail, which was a 20% project
8:25 which people were internally
8:29 using internally as an email service.
8:31 you know, how would we be able to think
8:34 about, you know, Gmail is what led us to
8:37 do workspace, get into the enterprise. I
8:39 made a big bet on Google Cloud, you
8:42 know, uh, which is, you know, tens of
8:43 billions of dollars in revenue today,
8:47 right? And, and so my point is, uh, you
8:50 know, things build out over time, right?
8:52 Um, think about the journey we've been
8:55 with Whimo, right? And so I think one of
8:57 the mistakes people make often in a
9:00 period of rapid innovation is think
9:02 about what is that next big business
9:03 versus looking at the underlying
9:05 innovation and say can you build
9:06 something and put out something which
9:10 people love and use right and and and
9:12 out of which you you do the next thing
9:15 and you know create value out of it. So
9:18 when I look at it AI is the such a
9:20 horizontal piece of technology across
9:23 our entire business. It's why you know I
9:26 mean it impacts not just Google search,
9:30 YouTube cloud, all of Android you saw XR
9:33 etc. Google Play things like
9:36 Whimo isomorphic which is based on Alpha
9:38 Fold. So I've never seen one piece of
9:40 technology which can impact and help us
9:42 create so many businesses right and you
9:44 know AI is going to be so useful as an
9:46 assistant. I think that people will be
9:48 willing to pay for it too. We're
9:50 introducing subscription plans. So
9:52 there's a lot of you know headroom ahead
9:55 I think and and obviously that's why we
9:56 are investing because we see that
9:59 opportunity some of it will take uh take
10:01 time and it may not be always
10:03 immediately obvious uh you know I gave
10:05 the way example in fact 3 years ago
10:08 people were the sentiment on was quite
10:10 negative 3 years ago you know but I
10:11 actually you know as a company we
10:12 increased our investment into way more
10:14 at that time right because you're
10:16 betting on the underlying technology and
10:18 you're seeing the progress of where it's
10:20 going but these are good questions right
10:24 in in some ways if you don't realize the
10:25 opportunities that may constrain the
10:28 pace of uh you know investment in this
10:30 area but I but I you know but it I'm
10:32 optimistic we'll be able to unlock new
10:34 opportunities one reason I wanted to
10:35 start here as the foundation of the
10:37 conversation is you showed off Android
10:38 XR yesterday you showed off some
10:40 prototype glasses you have some partners
10:43 making glasses a lot of people think
10:46 augmented reality glasses powered by AI
10:48 will be the realization of the full
10:50 platform shift Right. You will have an
10:51 always on assistant that can look at the
10:53 world around you. You showed some of
10:55 those demos yesterday. The form factor
10:56 will change. The interface will change.
10:57 This will be a market as big as
10:59 smartphones were. How close do you think
11:03 we are to that as a mainstream product?
11:05 Um, you know, it was a nice reflective
11:07 moment all the way back from Google
11:09 Glass. Uh, you know, wearing the
11:11 product. I think there's a difference
11:14 between goggles and glasses. uh you know
11:15 everyone at Verge understands this as
11:18 well but you know obviously we are also
11:20 shipping goggles you know we have
11:23 announced uh uh products with Samsung to
11:26 come later this year on the XR side I
11:28 think I'm excited about our partnership
11:31 with Gentle Monster and Barbie Parker
11:33 will have products in the hands of
11:36 developers this year and I think but I
11:38 think those products will be pretty
11:40 close to what people eventually see as
11:43 final products so I'm excited I The pace
11:47 is actually pretty uh we palpable. So
11:50 I'd be shocked if you you and I were
11:52 sitting next year. Uh you know I wasn't
11:54 wearing one of that uh when I'm doing
11:56 the Do you think but that will be like a
11:58 mainstream iPhone level replacement
12:00 product cuz there's a lot of hardware
12:01 that needs to get developed along the
12:02 way to pull that off. Yeah, there are
12:04 there are look you're wearing something
12:06 on your face. people like I have
12:10 prescription uh right and uh you know
12:12 the bar is higher I think in terms of
12:13 making the experience seamless enough
12:15 that you're willing to wear it in your
12:19 face and enjoy it for all so so I don't
12:20 think nely next year it's as mainstream
12:23 as what you're talking about but would
12:25 millions of people be trying it I think
12:27 so yeah so both are both are true I
12:29 think so I have to ask you just before
12:31 we sat down openai announced that Johnny
12:33 IV was selling a company he had started
12:37 called IO to the company and IV and his
12:38 design consultancy love from would take
12:40 over all design. They didn't announce a
12:42 product but they said it's the future of
12:44 computing and it's coming next year. Is
12:46 do you anticipate more of that
12:48 competition that your competitors who
12:49 don't have a smartphone operating system
12:52 will go even harder in this direction?
12:54 I'm looking forward to a open IO
12:57 announcement ahead of Google IO the
12:59 night before. No, first of all, look,
13:02 stepping back, I mean, Johnny IV is, uh,
13:05 you know, uh, one of a kind. You know,
13:06 you look at this track record over the
13:08 years. I've definitely, uh, I've met him
13:10 only once or twice, but, you know, I've
13:12 admired his work obviously like so many
13:14 of us. Uh, so I think it's it's exciting,
13:16 exciting,
13:18 shows, you know, this is why I feel like
13:21 it's such there's so much innovation
13:23 ahead. Uh, and I think people tend to
13:26 underestimate this moment in some ways.
13:29 people tend to I always like to point
13:31 out when the internet happened Google
13:34 didn't even exist right I think what
13:36 people so when you when you think about
13:37 I think AI is going to be bigger than
13:39 the internet there are going to be
13:42 companies products categories created
13:44 which we aren't aware of today so I
13:46 think the future looks exciting I think
13:48 there's a lot of opportunity to innovate
13:51 around uh uh innovate around hardware
13:53 form factors at this moment with this
13:55 platform shift so I'm looking forward to
13:57 seeing what they uh you know we are
13:59 going to be doing a lot as well and I
14:02 think you know it's it's a exciting time
14:04 to be a consumer it's an exciting time
14:05 to be a developer so I think looking
14:07 forward to it I've in that video
14:09 described the phone and the laptop is
14:11 legacy platforms which is very
14:14 interesting considering his own history
14:15 are you all the way there that the phone
14:18 and laptop are legacy platforms look I
14:22 think these things if anything uh I
14:24 found through this AI moment using the
14:26 web a lot
14:29 right? Because I'm like it's easier to
14:31 create a V3 video on my browser in a big
14:38 screen, right? And and and so um I think
14:40 the way I've internalized is computing
14:42 will be available in like you don't have
14:44 to make these hard choices. You know,
14:46 you will computing is will become so
14:49 essential to you. You're going to have
14:52 it in multiple ways around you when you
14:55 need it, right? like I use a phone, a
14:57 tablet, a laptop, and I have my
15:00 workstation, right? And so I have the
15:03 breadth of it. Um, and but but you know,
15:06 I I you know, over time it makes sense
15:09 to me at some point in the future
15:11 consuming content by pulling out this
15:13 black glass display rectangle in front
15:14 of you and looking at it is not the most
15:17 intuitive way to do it. But I, you know,
15:18 but I think it's going to take some
15:19 time. I feel like we could do a full
15:21 hour just on Android tablets and where
15:22 they could go. We're going to come back
15:24 for that. A big part of what you're
15:26 describing implicates search in in
15:28 really big ways, right? We're going to
15:30 be surrounded by information. Search or
15:32 Gemini or some future Google product
15:34 will organize that information, take
15:36 action for you across the web in some
15:38 way and you will have a a companion and
15:40 maybe you only pull out your tablet to
15:42 watch a video or something. A lot of
15:44 what's going on with search has
15:46 downstream effects on the web,
15:47 downstream effects on information
15:49 providers broadly. Are you starting?
15:51 Last year we spent a lot of time talking
15:53 about those effects. Are you seeing that
15:54 play out the way that you thought it would?
15:56 would?
15:57 Look, I
16:02 um you know it depends. Uh I think I do
16:04 think people are consuming a lot more
16:07 information. Um and you know web is one
16:09 specific format. So we should talk about
16:12 the web but zooming back out you know
16:13 there are new platforms like YouTube and
16:15 others too. So I think people are just
16:17 consuming a lot more information right.
16:20 So it feels like an expansionary moment.
16:22 Uh I think there are more creators,
16:25 people are putting out more content. Um
16:27 you know and so people are generally
16:29 doing a lot more maybe people have a
16:30 little bit extra time in their hands and
16:33 so it's a combination of all that on the
16:35 web. Look things which have been uh
16:37 interesting and you know we've had these
16:39 uh conversations for a while you know
16:41 obviously in 2015 there was this famous
16:44 the web is dead uh you know I I always
16:46 have it somewhere around you know which
16:48 I look at it once in a while uh
16:50 predictions it's existed for a while I
16:52 think the web is evolving pretty pretty
16:55 profoundly I think that is true when we
16:57 crawl when we look at the number of web
17:00 pages available to us that number has
17:02 gone up by 45% in the last two years alone
17:03 alone
17:04 Right. So that's a staggering thing to
17:06 think. Can you detect if the if that
17:08 number if that volume increase is
17:10 because more pages are generated by AI
17:12 or not? This is the thing I I worry
17:16 about the most. Right. I um it's a good
17:18 question. We generally have many
17:19 techniques in search to try and
17:21 understand the quality of a page
17:22 including whether it was machine generated
17:23 generated
17:27 etc. This trend that doesn't explain
17:29 this trend we are seeing right. So
17:31 generally there are more web pages
17:34 right? So uh you know so so you know at
17:36 an underlying level. So I think that's
17:38 an interesting uh phenomenon. I think
17:40 everybody as a creator like you do at
17:42 Verge I think today if you're doing
17:44 stuff you have to do it in a
17:48 crossplatform cross format way uh in a
17:50 uh you know I look at the quality of
17:52 video content you put out. It's very
17:54 sophisticated right? You know and uh
17:56 very different from how words used to be
17:58 maybe 5 to 10 years ago right? It's
18:00 profoundly changed. So I think things
18:03 are becoming more dynamic cross format.
18:05 I think the thing another thing people
18:08 are under underestimating with AI
18:11 is AI will make it zero friction to move
18:14 from one format to another right because
18:16 our models are natively multimodal. We
18:18 kind of tease people's imagination with
18:21 audio overviews in notebook right the
18:23 fact you can throw a bunch of documents
18:25 at it and you have a podcast and you can
18:28 join and learn from it. So I think this
18:31 notion the static moment of like you
18:34 produce content by format whereas I
18:37 think machines can help translate it
18:38 from it's almost like different
18:40 languages and they can go seamlessly
18:42 between I think it's one of the
18:44 incredible opportunities to be unlocked
18:48 right ahead and so uh but maybe I didn't
18:49 want to drift from the uh question we
18:52 were having but I look I I think I think
18:54 people are producing a lot of content
18:56 and I see consumers consuming consuming
19:00 a lot of content and and you know it's
19:03 we we see it in our products others are
19:05 seeing it too so that's that's how I
19:06 would probably answer at the highest
19:08 level the way I I see it currently is
19:10 that the web is at an all-time high as
19:13 an application platform right the fact
19:15 that Figma exists and is as successful
19:17 as it is and its primary interfaces as a
19:20 web app is I think remarkable a lot of
19:21 the products you were talking about are
19:23 expressed as web apps even some of the
19:24 most interesting search results you
19:27 showed yesterday are, you know, Google
19:28 would generate a custom web app for you
19:30 and display it in a search result to do
19:32 some data visualization. I think that's
19:34 all looking incredible. I think the web
19:35 as a transaction platform is reaching
19:37 new heights, especially with rulings
19:39 that mean smartphone makers have to let
19:41 people push transactions to the web.
19:42 There's something very interesting
19:44 happening there. As a media platform, it
19:46 feels like it's at an all-time low,
19:47 right? If I was starting the verge, the
19:49 web, that's a media platform. The web is
19:50 a media plat is an information platform.
19:52 If I was starting The Verge today uh
19:55 with 11 of my co-founder and friends, we
19:57 would start a Tik Tok channel. We we
19:59 might start a YouTube channel. We would
20:00 definitely not start a website with the
20:02 dependencies we have as a website today.
20:04 And that's the dynamic that it feels
20:07 like AI is pushing on even harder. I'm
20:09 not fully sure I agree. Right. I think
20:11 you know I think if you were to go and
20:13 restart Verge again, I bet you would
20:15 start a uh you would have a
20:17 extraordinary web presence at best. No,
20:18 I I've thought about this a lot. I think
20:20 at at best our web presence would look
20:22 like a Substack or a ghost or something,
20:24 right? Uh maybe. Look, I u you know I'm
20:27 not you know I'm not fully sold on that
20:29 like you know but you know you know the
20:30 space. I acknowledge you know that space
20:33 better than I do. Uh so I don't mean to
20:35 be uh mean to be you know I don't have
20:38 that intuition which you do here. Uh but
20:40 look I I
20:42 see in fact you say the web application
20:45 platform is an all-time high but I've
20:47 looked you know I was wipe coding with
20:51 replet a few weeks ago uh you know
20:54 create I mean the power of what you're
20:56 going to be able to create on the web we
20:59 haven't given that power to developers
21:01 in 25 years. Mhm. So that is going to
21:03 come ahead.
21:07 So, you know, it's not exactly clear to
21:10 me. Uh, you know, maybe today you're
21:11 looking at and say, I won't put all the
21:12 investment in because it looks like a
21:14 lot of investment to do that. But that
21:19 may not be true 2 years out, right?
21:21 Like, you know, if you feel like you
21:23 would create a Tik Tok channel, then,
21:26 you know, maybe with like 2% extra
21:27 effort, if you could have a robust web
21:29 presence, why wouldn't you? Right? And
21:32 so so I'm not fully I you know I'm not
21:33 fully sold on it but I think it's a good
21:36 question to ask but you know you have to
21:38 somehow reconcile that with the fact
21:41 that overall that traffic seems to be
21:44 growing we see more web pages in so
21:45 somewhere we have to explain all of that
21:47 too. You know the publishers as they
21:49 often do responded to Google IO
21:51 announcements. So the news media
21:54 alliance uh after AI mode was announced
21:56 yesterday I would say they're very
21:58 upset. Uh here's here's a statement from
21:59 the president of the news media
22:01 alliance. Links were the last redeeming
22:02 quality of search that gave publishers
22:05 traffic and revenue. Now Google takes
22:07 content by force and uses it with no
22:09 return, no economic return. That's the
22:10 definition of theft. And they go on to
22:12 say the DOJ lawsuits must address it.
22:14 That's pretty furious. That's not a
22:15 negotiation, right? That's a we just
22:17 want this to stop. How do you respond to
22:20 that very loud set of people who say,
22:22 "Yeah, okay, maybe it's growing
22:24 somewhere, but for us it's crushing our
22:28 businesses." Look I um you know first of
22:32 all through all the products I mean AI
22:34 mode is going to have sources and you
22:37 know one of we're very committed as a
22:40 direction as a product direction uh to
22:42 make I think part of why people come to
22:46 Google is to experience that breadth uh
22:48 of the web and and go in the direction
22:50 they want to right so I view us as
22:52 giving more context yes there are
22:54 certain questions which may get answers
22:56 but overall and that's the pattern we
22:58 see today right and and if anything over
23:00 the last year it's clear to us the
23:02 breadth of where we are sending people
23:05 to is increasing uh and and so I I
23:07 expect that to be true with AI mode as
23:08 well but if it was increasing wouldn't
23:12 they be less angry with you look I more
23:15 than you're you're always going to have
23:17 areas where people are robustly debating
23:19 value exchanges etc right like app
23:22 developers and platforms that's not on
23:25 the web etc right it's inherently the uh
23:26 you know you know there's always going
23:29 to be when you're running a platform
23:31 these debates I would challenge I think
23:35 more than any other company we we think
23:39 about we prioritize sending traffic to
23:41 the web no one sends traffic to the web
23:44 in the way we do I look at other
23:46 companies newer emerging companies they
23:48 openly talk about it as something they
23:50 are not going to do right we are the
23:52 only ones which make it a high priority
23:55 agonize nice and so on. Look, we'll
23:56 engage and you know, we've always
23:58 engaged. There are going to be uh
24:00 debates through it all, but we are
24:02 committed to, you know, I've said this
24:06 before, everything we do across all you
24:09 will see us 5 years from now, sending a
24:10 lot of traffic out to the web. I think
24:11 that's the product direction we are
24:13 committed to. I think it's what users
24:15 look for when they come to Google,
24:18 right? and uh and the nature of it will
24:21 evolve. But you know I I I you know I'm
24:22 confident that that's the direction
24:24 we'll continue taking. Is there public
24:27 data that shows that AI overviews and AI
24:28 mode actually send more traffic out than
24:30 the previous search engine results page?
24:32 Look, it's you know the way we look at
24:34 it is I mean obviously we take a lot of
24:37 uh we definitely uh sending traffic to a
24:39 wider range of uh uh uh sources
24:42 publishers and and because just like
24:44 we've done over 25 years we went through
24:47 the same with featured snippets the the
24:50 quality of you know it's higher quality
24:52 referral traffic too so right and we can
24:54 observe it because the time when people
24:56 spend as one metric and there are other
24:58 ways by which we measure quality of our
25:00 outbound traffic is also increasing
25:02 So and overall through this transition I
25:04 think generally AI overviews are also
25:06 growing and you know the growth
25:08 compounds over time. So whenever we have
25:11 worked through these transitions it ends
25:14 up post that's how Google has worked for
25:16 uh 25 years and you know and we end up
25:18 sending more traffic over time. So
25:19 that's that's how I would expect all
25:21 this to play out. So why do you think
25:23 that there is so much general economic
25:25 turmoil on that side of the the house
25:28 right? If you're sending more traffic
25:30 and the goal over time is to make sure
25:32 that that sending traffic to we're a
25:34 year into it, right? And it it doesn't
25:35 seem to have gotten better over there.
25:37 No, look, we are sending traffic to a
25:40 broader source of people. People may be
25:42 you know uh surfacing more content,
25:43 looking at more content. So somebody
25:46 individually may see less. I mean there
25:47 are all kinds of at the end of the day
25:50 we are reflecting what users want. Mhm.
25:53 Right. You know if you do the wrong
25:55 thing users won't use our product go
25:57 somewhere else. Right. And and and so
26:00 you have uh uh you know you have all
26:02 these dynamics underway and I think we
26:04 have genuinely you know we took a long
26:07 time designing a AI overviews and we are
26:09 constantly iterating
26:11 uh in a way that we
26:13 prioritizes you know sources and sending
26:15 traffic to the web. I mean, my my
26:16 criticism of this industry, just be
26:17 clear, is that everyone's addicted to
26:19 Google, and it would be better if they
26:21 weren't, but they're addicted to Google,
26:23 right? And they're they're feeling it.
26:25 And then on top of that, you see, you've
26:26 mentioned several times, like overall
26:28 queries are increasing on Google
26:30 services, but they're they're changing,
26:31 right? They're getting longer. They're
26:33 getting more complicated. AI mode might
26:36 walk you through several steps. Maybe
26:38 some people are searching on TikTok now.
26:40 Eddie EQ uh on the stand in the trial
26:43 the other day said uh search and Safari
26:45 for the last month dropped for the first
26:47 time in 22 years. That's a big stat.
26:49 Obviously your stock price uh was
26:51 affected by it. There was a statement.
26:53 Is that trend bearing out that the
26:54 standard Google search is is dropping
26:56 from devices and different kinds of
26:57 searches are
26:59 increasing? No, look uh we've been very
27:01 clear. We are seeing overall query
27:05 growth in search. Um you know it looks
27:06 but have you actually seen the drop in Safari?
27:08 Safari?
27:10 Look, we have a comprehensive view of
27:13 how we look at data across the board. Uh
27:14 there's a lot of there can be a lot of
27:17 noise in search data. Uh but everything
27:19 we see tells us we are seeing query
27:21 growth including across Apple's devices
27:24 and platforms. Uh and specifically, you
27:26 know, I think we quantified the query
27:29 growth from AI overviews and what's
27:31 what's healthy is that the query growth
27:33 is continuing to grow over time. This is
27:35 what I I've said before. It feels very
27:38 far from a zero sum game to me. I said
27:40 this last year. It's interesting. We
27:41 spoke about Tik Tok, right? Think about
27:45 like how profound a new product Tik Tok
27:47 was. How has YouTube done since Tik Tok
27:50 has come? Right? You could ask all these
27:52 questions there. Like why is it that Tik
27:55 Tok comes and YouTube has grown? I think
27:57 what we always underestimate in these
27:59 moments is people are engaging more,
28:01 doing more with it. we are improving our
28:04 products and and so you know so that's
28:05 how I would I would think about these
28:08 moments let me just broaden that out to
28:10 agents right uh I watched Dennis Asabis
28:12 yesterday he was on stage with Sergey
28:15 Brin and Alex Hantritz asked
28:17 him what does a web look like in 10
28:19 years and Dennis said I don't know that
28:21 an Asian first web looks anything like
28:23 the web that we have today I don't know
28:24 that we have to render web pages for
28:26 agents the way that we have to see them
28:28 that kind of implies that the web will
28:30 turn into a series of databases for
28:32 various agents to go and ask questions
28:35 to and return those answers. And I I've
28:36 been thinking about this in the context
28:38 of services like Uber and Door Dash and
28:40 Airbnb. Why would they want to
28:42 participate in that and be abstracted
28:45 away for agents to use the services
28:46 they've spent a lot of time and money
28:48 building? Two things though. First,
28:49 there's a lot to unpack. It's a
28:52 fascinating topic. Um the web is a
28:54 series of databases, etc. We build a UI
28:56 on top of it for for all of us to
28:57 consume. This is exactly what I wanted
28:59 is the web is a series of databases. It
29:01 is like you know but I think I think I I
29:03 listened to the uh te surrogate
29:05 conversation yesterday I enjoyed it. I
29:08 think he's saying for a agent first web
29:09 like so you know for a web which is
29:12 interacting with agents you would you
29:14 would think about how to make that
29:15 process more
29:17 efficient. Today you're running a
29:19 restaurant people are coming dining and
29:22 eating and people are ordering takeout
29:25 and delivery. Obviously for for you to
29:26 service the
29:28 takeout you would think about it
29:30 different than all the tables and the
29:32 clothing and the furniture and the like
29:33 you know so but both are important to
29:35 you. You could be a restaurant which
29:37 decides I'm not going to participate in
29:40 the takeout business. I'm only going to
29:42 focus on uh on on on the dining
29:44 experience. You're going to have some
29:46 people vice versa. I'm going to say I'm
29:48 going to go all in on this and run a
29:52 different uh experience. So to your
29:53 question on agents right I think think
29:57 of agents as uh you know a new powerful
30:00 format. I do think it'll happen in
30:02 enterprises faster than consumer. Mhm.
30:04 Because in the context of an enterprise
30:06 you have a CIO who's able to go and say
30:08 I really don't know why these two things
30:10 don't talk to each other. Yeah. Right.
30:12 I'm not going to buy more of this unless
30:15 you interoperate with this. uh so I I
30:16 think partly why you see on the
30:19 enterprise side a lot more uh uh agentic
30:21 experiences on the consumer side I think
30:23 what you're saying is a good point right
30:25 people have to think about and say what
30:28 is the value for me to participate in
30:31 this world and it could come in many
30:33 ways it could be because I participated
30:35 in it and overall my business grows
30:38 right some people may fa feel that it's
30:40 disintermediating right and like it
30:42 doesn't make sense I think all of that
30:44 are you
30:46 uh all of that can
30:49 happen but users may vote with their
30:52 feet right like you know you may find
30:55 like some people are supporting the
30:56 agentic experience and your life is
30:59 better because of it and so you're going
31:01 to have all these dynamics and like you
31:02 know and I think they're going to try
31:04 and find an equilibrium somewhere is how
31:06 you know everything evolves yeah I mean
31:07 I think the idea that the web is a
31:08 series of databases and we change the
31:09 inter first of all this is like the most
31:11 decoder conversation that we've ever had
31:13 I'm very happy with it but you know I
31:15 had DAR from Uber on the show. I asked
31:17 him this question from his perspective.
31:18 And his answer attracts yours broadly.
31:20 He said, "First, we'll do it because
31:22 it's cool and we'll see if there's value
31:23 there and if there is, you know, he's
31:25 going to charge a a big fee for the
31:27 agent to come and use Uber because
31:30 losing the customer for him, losing the
31:32 ability to upsell or sell a
31:34 subscription." None of that is great,
31:36 right? That it's the same is true for
31:37 Airbnb. I keep calling it the Door Dash
31:40 problem. Like Door Dash should not be a
31:41 dumb pipe for sandwiches, right? that
31:42 they they're trying to actually run a
31:44 business and they want the customer
31:46 relationship. And so if the agents are
31:48 going across the web and they're looking
31:49 in all these databases and saying,
31:51 "Okay, this is where I get food from and
31:52 this is where I get cars from and this
31:55 is where I book I think the demo was
31:56 booking a a vacation home in Spanish,
31:58 right? And I'm going to connect you to
32:02 that agent uh that travel agent." Is it
32:03 just going to be tolls that everyone
32:05 pays to let the agents work? Because the
32:08 the price I I still don't the CIO gets
32:10 to just spend money to solve the
32:11 problem. Yeah, he says, "I want this
32:12 capability for you and I'm just going to
32:15 pay you to do it." The the market, the
32:16 consumer market doesn't have that
32:18 capability. Right?
32:21 All kinds of models may emerge, right?
32:25 Uh I can literally see envision 20
32:26 different ways this could work, right?
32:28 Consumers could pay a subscription for
32:31 agents and the agents could rev share
32:34 back, right? So, you know, so that that
32:36 is a CIO like use case you're talking
32:37 about. That's possible. We can't rule
32:39 that out.
32:40 uh rule that out. I don't think we
32:42 should underestimate people may actually
32:44 see more value participating in it. You
32:47 know, I think this is uh you know it's
32:51 tough to predict. Uh but I do think uh
32:53 over time like you know like if you're
32:56 removing friction and improving user
32:58 experience, you know, it's tough to bet
33:00 against those in the long run, right?
33:02 And so I think it, you know, I think if
33:05 in general you're lowering friction for
33:07 it, uh, you know, and people are
33:10 enjoying using it, right? Somebody's
33:12 going to want to participate in it and
33:14 grow their business. Yeah. Right. And
33:16 like would brands want to be in
33:18 retailers, why don't they sell directly
33:22 today? Yeah. Like why won't they do
33:24 that? I I don't know. Because retailers
33:26 provide value in the middle, right? And
33:29 like you know uh why do merchants take
33:33 credit cards right like why pay I'm just
33:35 saying so like there are many parts like
33:37 and and you find equilibrium because
33:39 merchants take credit cards because they
33:40 see more business as part of taking
33:43 credit cards than not right and which
33:45 justifies the increased cost of taking
33:47 credit cards um it may not be the
33:49 perfect analogy but I think there are
33:51 all these kinds of effects going around
33:53 but but what you're saying is true that
33:55 you know some of this will slow uh
33:57 progress and agents just because we all
34:00 are excited about A2A and MCP and we
34:03 think no like some of it will slow uh uh
34:06 slow uh slow progress. Um but I think
34:08 it'll be it'll be very dynamic. Yeah.
34:10 Yeah. There's other pressures on Google.
34:11 There's antitrust pressures. The
34:12 government would like you to sell
34:14 Chrome. Can you do all the things you
34:15 want to do if you're made to sell Chrome?
34:17 Chrome?
34:19 Look, I mean I you know I don't want to
34:22 comment on Look, we are uh you know we
34:25 are in a legal process. Um you know I
34:28 look at having directly been involved in
34:31 building Chrome uh right I look at the I
34:33 think there are very few companies which
34:35 would have appro you know we we not only
34:38 improved our product we improved like
34:40 the state of the web by building Chrome
34:43 we open sourced it we provided as
34:45 Chromium everyone else has access to the
34:48 browser so I think the amount of R&D the
34:50 amount of innovation we put into it uh
34:53 the investments in security etc we do so
34:55 I think we to sell it. Can you do all
34:57 the things that you want to do? Um,
35:00 look, I I don't think that's the uh
35:02 scenario we're looking at. stepping back
35:05 you know as a company look I I think as
35:08 a company I think of ourselves as a
35:11 deeply foundational technology company
35:13 uh which then translates it into
35:15 products that touch billions of people
35:17 and you know so we do it across many
35:19 many things and so of course I think
35:20 look as a company we are going to
35:22 continue investing and doing our best to
35:24 innovate and build a successful business
35:26 in all scenarios uh so this how I would
35:29 answer it the Trump administration is uh
35:31 extremely transactional I would say, you
35:32 know, the tech industry has a new
35:34 relationship with Trump in a second
35:36 term. You were at the inauguration. Have
35:38 you had conversations about what a
35:39 settlement might look like and what the
35:41 Trump administration might demand to
35:43 make these problems go away? No. No.
35:46 Like, look, we we've always we we've
35:48 engaged with the DOJ or like we do o
35:51 over the years on in the context of all
35:53 the all the cases we have. So, that's
35:55 you know, that's that's how we normally
35:57 do these conversations. Trump has very
35:58 publicly said he doesn't like his search
36:00 ranking and he wants it changed in some
36:02 way. Would you ever adjust the search
36:04 ranking for Donald Trump?
36:06 No. Like you know we have a I can't
36:08 today the way Google search works is I cannot
36:10 cannot
36:14 uh no person Google can influence the
36:15 ranking algorithm in AI mode is
36:17 different right we've seen system
36:19 prompts adjusted in very chaotic ways
36:21 from some of your competitors. Is that
36:22 something that you would be open to in a
36:24 world where you're serving the full
36:26 answer? So, so would you adjust the AI
36:28 mode responses in response to political
36:31 pressure? No. Because we we've seen
36:33 certainly in Grock and others that
36:35 system prompts change the answers in
36:37 dramatic ways. No, like the way we do
36:39 ranking like you know the way we do
36:41 ranking is sacrosan to us. Uh you know
36:45 we've done it over 25 years. uh you know
36:47 we make a lot of when there's a lot of
36:50 ranking signals we take into account and
36:52 stuff and if there's broad feedback from
36:54 people that something isn't working
36:56 right we will look at it systematically
36:58 and try and make changes but you know we
37:00 don't take look at individual cases and
37:02 ever change ranking when you think about
37:04 those sources of information one of the
37:05 things that I have been thinking about a
37:07 lot is I don't know this the CDC web
37:09 pages have changed a lot recently right
37:11 diversity equity inclusion language has
37:13 been removed from pages across the
37:15 government those used to be very high
37:17 ranking sources in Google search, right?
37:19 We would we just implicitly trusted the
37:22 CDC's web pages in some ways. Are you
37:23 re-evaluating that that there might be
37:25 misinformation on some of these pages
37:27 that then gets synthesized into AI
37:29 results? No, it's a misunderstanding of
37:30 how search works, right? We don't
37:31 individually evaluate the
37:34 authoritiveness of like a page, right?
37:36 And like it's what our signals does, you
37:38 know, page ranket. It's, you know,
37:41 obviously our signals
37:43 are multiple orders of magnitude more
37:45 complicated than page rank today, but to
37:47 use page rank as an example, we weren't
37:49 the ones determining how authoritative a
37:51 page is. It's how other pages were
37:53 linking to it like an academic citation,
37:56 etc. So, so we don't, you know, so we
37:58 are not making those decisions today.
37:59 And so I don't I don't see that
38:01 changing. As you synthesize more of the
38:02 answers, do you think you're going to
38:03 have to take more responsibility for the results?
38:05 results?
38:07 Look, we are giving context around it
38:09 and but we still anchoring it in, you
38:12 know, the sources we find. Uh, you know,
38:14 but we've always felt a high bar, you
38:18 know, in in in Google. I mean, last year
38:20 when we launched AI overviews, I think,
38:23 you know, people were adversely querying
38:25 to find errors and the error rate was 1
38:29 in 7 million for adversal queries and so
38:30 but that was a big but that's the bar.
38:32 We've always operated as a company and
38:35 so I think to me nothing is nothing has
38:37 changed like you know Google operates
38:39 under a very high bar that's the bar we
38:41 strive to meet and and you know our
38:43 search page results are there for
38:45 everyone to see with that comes that
38:48 natural uh accountability and you know
38:49 we have to constantly earn people's
38:51 trust so that's how I would approach it
38:52 what do you think the marker is for the
38:54 next phase of the platform shift after
38:56 this one we we open by talking about
38:57 we're in a second phase what's the
38:59 marker for the final phase or the third
39:02 phase of the platform shift you mean of
39:03 the AI platform. What what are you
39:05 looking for as the next marker? Oh, you
39:07 know, look, I think, you know, the the
39:10 the real thing about AI, which I think
39:12 why I've always called it more profound
39:14 is, you know, self-improving technology,
39:17 right? And you know uh having watched alpha
39:18 alpha
39:21 go you know start from scratch not
39:24 knowing how to play go and you know
39:27 within couple hours or 4 hours be better
39:30 than uh a top level human players and in
39:32 8 hours you know no human can ever
39:35 aspire to play against it right so you
39:36 know and that's the essence of the
39:39 technology obviously in in in a in a in
39:42 a deep way so I think look I think
39:44 there's so much ahead
39:47 uh you know on the opportunity side you
39:51 know I I'm blown away by the ability to
39:53 discover new drugs you know completely
39:56 uh change how we treat diseases like
39:58 cancer over time etc you know the
40:02 opportunities there um you know I uh the
40:04 creative power which I talked about
40:06 which we're putting in everyone's hands
40:08 like the next generation of kids
40:10 everyone can program and will to if you
40:13 think of something you can create it I
40:14 think I I I don't think we have
40:16 comprehended what that means but that's
40:18 going to be true. The part which the
40:19 next phase of the shift which is going
40:21 to be really meaningful is when this
40:23 translates into the physical world
40:26 through robotics. Mhm. Right. So that
40:28 aha moment of robotics I think when it
40:31 happens that's going to be the next big
40:33 thing we will all uh grapple with.
40:34 Right. Today they're all online and
40:37 you're doing things with it. But you
40:40 know one hand you know today I think of
40:42 way more as a robot right? So we are
40:44 running around uh driving around a robot
40:47 but I'm talking more uh general purpose
40:50 robot and you know and when AI creates
40:52 that magical moment with robotics I
40:54 think that'll be a big platform shift as
40:55 well. Yeah I'm looking forward to it
40:56 next year we're going to do this with
40:58 glasses and robots. It's going to be
40:59 great. We'll we'll give it a shot. Thank
41:01 you so much. All right. Thanks. I appreciate
41:02 appreciate
41:05 it. Are these setups getting more and
41:08 more stuff? I just feel like I have to
41:09 say something more important just for
