Apple’s Anthropic Flirtation, Can Meta Build Superintelligence?, AI Browser Wars — With M.G. Siegler

0:00Apple is thinking about handing Siri's brain to Anthropic. Does Meta have what it takes to build superintelligence? And will you soon be using an AI browser? That's coming up right after this. Welcome to Big Technology Podcast, where we're kicking off a new monthly segment, or really monthly episode, with the great MG Siegler of Spyglass, who's a regular guest with us here. But I thought, you know what, let's amp it up. We'll spend the first Monday of every month speaking with MG about the latest in big tech and AI and spend a lot of time speaking about the great reporting and writing that he does within Spyglass.

0:35So MG, really happy to have you here. Very glad to kick this off. Yeah, thanks for asking me to do it, Alex. And I'm excited to chat with you on a regular basis. Yes, and I'm definitely stoked about this. So to kick it off, let's pick up our Apple discussion. And I think just having this regular cadence is great because we can sort of check in on these companies and say, hey, how have they been doing since our last conversation? Of course, the last time we spoke was the day after WWDC. Basically, no AI. We said, okay, they're going to have to make some dramatic moves, maybe grasp for a perplexity, do some acquisition, or potentially even rewrite Siri itself.

1:14And it looks like Apple is starting to consider this seriously with some news that came out over the past week or so, where they are considering taking either OpenAI or Anthropics Claudebot and putting it in basically lobotomizing Siri and changing the brain around. And Anthropics seems to be the one that is the leader, at least in the clubhouse. So what do you think about this potential move from Apple and what's the significance of it? I think it's long overdue. I've been writing about this for a while now. At one point, I think it was around, you remember when several months ago, maybe around the first part of the year, there was a whole meme going around about how, oh, it was around the time of the Super Bowl, because it was a whole meme around how, you know, Siri couldn't correctly know who had won previous Super Bowls, something as rudimentary as that from like a knowledge graph perspective.

2:07And they would get all sorts of stuff wrong, thought the Chiefs had won like 50 of them or something like that. And so it seemed like right after that point, I had written sort of what I thought was a little bit more provocative of a piece with, you know, the the old Betteridge's headline law of the question mark in it of being like, should Apple replace Siri with basically an LLM with, with, with chat GBT. And then after that, that Superbowl thing came out, it's like, I was pushed over the edge. It's like, they've got to replace Siri with something else, whether it be chat GBT or Claude or Google Gemini or one of the other products, because this is just really table stakes stuff that they should, you know they've been working on siri for 15 plus years um you know to their credit as as we've long talked about they were first mover uh in that and i think that hurt them in many ways but the fact that we're in 2025 or you know last year 2024 and they couldn't get something as simple as that right while all these other competitive products have no problems doing those things um and again they're trying to do it in-house they're trying to do their own systems and stuff but all of this stuff is out there and the fact that they already had the relationship with open ai per last year's wwdc in place to do some of the offloading of um you know the the wider uh knowledge siri uh stuff you know reaching out to the web like why couldn't they just have slotted that in obviously it's easier said than done but still to this to this reporting now they've been thinking about doing it and obviously they've been testing it um and doing this like bake-off internally to see which of the products might be better.

3:46I did think that it was super interesting that they wound up at least per this report with Claude rather than ChatGPT. Yeah, that was surprising to me. And I definitely want to get into the different decisions that they're making and why Anthropic is the lead here and what it means for Anthropic. These are all really interesting parts of the story. And I should also note the Betteridge Law of Headlines is if you put a question mark in the headline, typically the answer is no. But when you asked Jathay Lopata Myciri, your answer was I think so. Yeah. Yeah. And, you know, again, we'll see like the reporting on it, if I have it right from from Mark Gurman, who else, you know, basically indicates that they wouldn't do this until next year.

4:28You know, a 2026 thing. Does that mean that they wait until sort of WWDC 2026 to announce it as part of iOS? I'm going to bungle it now. iOS 27. Yeah. Because it's the future year. Right. So it's not right. Right. So would they announce it as iOS 27 or do they try to get it out ahead of time as a dot release on iOS 26, something, you know, ahead of that? But again, presumably they could just slowly, you know, behind the scenes, just start giving more of the workload over to one of these LLMs that they already have plugged in. As noted, you know, they have ChatGPT in there right now. Do they slot in Clawed?

5:08Do they slot in Gemini and then start to offload that work without making a big deal about it? Or do they make a big deal about it? And I could see an argument both ways of wanting to do that. Now, I think you and I are both in favor of this move and we can make the case for it. We have on the show numerous times. Maybe we will come back to that in a moment. But I think let's just make the case against because I saw that you sort of grappled with whether this is actually the right move in your post. And the argument that Apple shouldn't do something like this is, look, large language models are going to commoditize at some point.

5:42You don't want to rely so much on a third party, a proprietary third party's technology. At the very least, you can develop this either in-house or use open source technology. And that importantly gives you way more control. And what Apple wants in every single one of its product experiences is the control. So is that the argument not to do this? And do you think that, I mean, I sort of have this feeling that Apple will eventually kind of fall back on that. Like, yeah, we'll just try to control the experience and do it ourselves. So what would be the counterpoint to that? Yeah. And a lot of that was sort of even thinking back to pre-last year's WWDC leading up to what they would eventually announce was the partnership with OpenAI.

6:25There had been the rumors that they were potentially going to do something with them. And what might that look like with regard to Siri? And definitely one of the things that jumps to mind when you first hear that is user privacy, right? And Apple's big proponent of user privacy and not wanting to send their own user data over to third parties unnecessarily. And so they obviously have sort of an opt-in mechanism to do that right now. But if they were to actually outsource this in a major way, presumably they would still have some sort of opt-in to begin with. But would they do it on the ongoing basis, like saying, hopefully not on every query that you would do that, but do they sort of prompt users like once a month or more often or less often than that in order to ensure that they know what they're sending?

7:13And, you know, I think that that's the one argument, the major argument on Apple side on top of the other stuff that you're talking about, where Apple just wants to control that user experience, regardless of sort of the data and user privacy aspect. They want to make sure also that the thing's not going to hallucinate. Right. I think this was the big fear leading up to, again, last WWDC, the LLM products were not as, you know, trustworthy as they maybe are now. I mean, in many cases, they're still not for certain types of queries. Right. But obviously, Apple is going to worry about that if if, you know, Gemini tells you to eat rocks or put rocks in the pizza or whatever.

7:50Apple's not going to want that as a part of their their experience. And so do they do something that's more like image playground, which has obviously been very structured and locked down? right like they only let you use up until recently up until maybe the new changes that they're rolling out in uh the new uh os26 um layers they basically only allowed you to do very specific types of um scenes and and using you know very guard railed um versions and so do they do something like that but again that's probably not going to make for a good product experience uh you know for a user perspective if they really try to lock it down that much yeah exactly as we outline sort of the Apple strategy here and why they may not want to do something like this or why they've dragged their feet to do this.

8:35It just seems to me like just all the arguments you can make on the stay closed and don't do a partnership or try to, you know, ride it out. They just don't hold water because this stuff is moving so quickly. I'll just give you one example. I'm on my phone all the time talking with ChatGPT. It's like this past weekend, I took it from like deep within the apps that I have on like the, you know, maybe third or fourth screen, and I put it into the doc screen. Because whenever I have a question, you know, this thing connects to the web, I'm like always talking with ChatGPT. It's kind of crazy how often I am speaking with these things.

9:11And this format of voice is just a thing I believe that's driving a lot of the growth within AI today. And if you wait for, let's say a Samsung or some other Android to deeply integrate with a competitor and provide this experience while you're just standing still and you don't have this integrated into your operating system. I think you could really end up losing phone sales because of it. Yeah, I totally agree. And I'm one step further than you. Not only is it in my dock, I use the action button to trigger ChatGPT. Really? So I have it mapped to do that. Yeah, from my iPhone. That's cool. You can do it via shortcuts and it works very well.

9:47Like they have one that's like, they have different granularity options. You could do it just to open up the app to chat with it. You can even do it so as an overlay screen so you can chat without having to leave another app. It's pretty well done. And you can trigger right into voice. You can trigger right into voice transcription. So, yeah, I use it all the time as well. Certainly I'm not going to use Siri all the time. But if it was baked in, you know, like I think then it's an interesting question of, you know, what the differences would be. Because I don't know about you, but I still sort of do these own, you know, internal bake-offs of my own, right?

10:21When I'm when I'm doing certain types of queries. Right. I'll do I'll use ChatGPT and I'll Gemini open and, you know, sometimes clawed for certain things, coding, of course, specifically and things like that. And so I do think that Apple runs a real risk without doing something here, because to your point, they're just not moving fast enough to sort of catch up with the current state of the art. And you and I have talked about this before, right? Beyond just even the product integrations, like one of the reasons that I would worry about Apple going forward is they don't have the right internal mentality about doing this stuff.

10:56And you can see it in the sort of products that they've shipped to date, specifically, obviously, with Siri, where they're clearly just moving too slowly and don't have the right product cadence in order to do that. And so offloading this will help them in that they would be using a third party to do that. I think in the best state, and again, this is something that I argued for in the original piece before this was rumored to be happening. But it's basically like, look, just outsource it for an indefinite amount of time while you work on your own stuff behind the scenes. Take the pressure off the team that's working on it internally.

11:30Again, we don't know the timetable of what that would mean. Does it take a year? Does it take longer than that? But again, if you fully outsourced it to something like ChatGPT or Claude, then you have the breathing room to be able to do that. And then behind the scenes, it's almost the inverse of what we were talking about before. I would just start to slowly roll out the actual Siri version of Siri, the Apple made version of Siri again, with certain types of queries that they feel confident that they can answer now and do a good job with. Obviously, they keep the timers and they keep the, you know, the sort of low hanging fruit stuff that's that's sort of simple right now baked into the system.

12:02But I mean, for the world queries and things, if they're building it behind the scenes and they can eventually sort of roll in, roll back to, you know, Siri as they, you know, build it behind the scenes with this integration. By the way, it's kind of crazy for me to hear how ingrained AI voice is in your life. I mean, I'll give you kind of an example of what's happened with me. I've kind of ping ponged from Siri back to open AI. So using ChatGPT, I'm like typing all these things and I'm asking like it's become my search. And then, you know, I'm expecting AI to be able to do this stuff for me. So when I like need to know the answer to something, like I'll just yell, hey, Siri, what's the answer?

12:42And then it tends to get that right. I don't know, 10, 20 % of the time. And then I was like, well, this is an experience I want to have. But obviously Siri's not holding it up to me. So that's where OpenAI got the promotion and the higher prominence for me. And I'm just like, wow, this is something that I'm just going to use all the time. I'm curious, what type of things are you speaking with ChatGPT about that you've given it such a prominent place in your phone? Honestly, it's almost, I mean, it's most of the Google queries that I used to be doing, right? Like, I mean, Google now I predominantly use when I'm on my desktop if I'm doing research for a post or if I'm trying to look up things to link to, then I will use sort of traditional Google search, but for pretty much everything else.

13:30And yeah, like you, I would use Siri for things like, you know, we're say we're traveling and we're walking around the city. I don't want to know the population like of a place, right? I would use Siri for that, like silly, silly things like that, but sort of fun, especially with kids It's like, you know, they like those little fun things. But OpenAI and ChatGPT have gotten so good at beyond just that, you know, very simple example of giving you, like, my older daughter is obsessed right now with, like, the London Underground system. And, like, knowing all the different tube lines and, like, when they came about.

14:03And so she's, like, constantly, like, thirsting for information about these. And obviously, I know, you know, the high-level names of them, but I can't go deep on it. So previously, right, we would have done like a Google search about that probably on your phone, like while you're waiting for a tube to come on a certain line. But now you can just you can just chat with ChatGPT about it. And it's really an incredible experience, like being able to go back and forth and and sort of answer asking follow up questions. And yeah, that's like one example of something that just has totally changed. Yeah, it's so good.

14:34And one of the things that's really nice about ChatGPT is when you're speaking with it in the voice interface, there's almost no latency. maybe a little latency if it has to go search the web for something but otherwise it's spitting these answers back so quickly and they have a bunch of different voices that you can pick from and i felt sometimes like when i'm switching the voice i feel like i'm murdering that old i feel kind of bad for it because it just goes to show you you develop this rapport your old friend is gone yeah that and that actually speaks to how uh my you know my older daughter and i started using it when they you know the cute like gimmicky thing when they rolled out the santa voice Right.

15:07It was just like it was such a fun thing to do during the holidays. Right. It's like, hey, Maisie, like, come come listen to this. And, you know, it's like, oh, oh, oh, hello there, Maisie. How are you? You tell you can tell, you know, your daughter's name, even though they know you're not talking to it directly. And yeah, it's all sorts of clever, clever user interaction. See, it's one of the more underrated releases from Shipmas. Was that? Yes, exactly. Twelve days of Shipmas. But let's talk about the decision between OpenAI and Anthropic, because this to me is almost as interesting as the news that they're actually thinking about this.

15:40So the report from Gurman, like we mentioned, is that they are thinking about Anthropic over OpenAI. Let's just talk about the bake off there. Why do you think that is? Because if you just put it product to product, OpenAI's voice mode is mature and much better than Anthropic's. Anthropic just released the Claude voice mode. So what do you think is going on behind the scenes there? I sort of wonder when you and I last spoke, we spoke about, you know, the list of companies that Apple could potentially acquire right in the space. And the one that I had sort of at the top, you know, we talked about perplexity.

16:16But the one above that even for me was Anthropic, even though I don't think it's possible for them to do that for all sorts of reasons, namely the price at this point. But I do just feel like overall from everything you hear externally, the mentality that Anthropic, that team has, and obviously they spun out right from OpenAI famously, you know, some of the team spun out and started Anthropic in that way. Just the mentality is more in line with what I think Apple would want from an AI perspective. And so part of me wonders if it's not that. That said, I don't think that they would, if it was much worse, I don't think that they would pick that over ChatGPT.

16:56So obviously they're using it and they do think it's better. I mean, there's been the reports as well that they've been using it for Xcode, right, internally. And that was going to be the thing that they outsourced, that Anthropic was going to be the tool that they used on the AI side to outsource Vibe coding to. and you know there's there's some work it seems like that's underway now within xcode that they'll be rolling out along those lines but at wwc they also on the flip side of that they talked about using open ai more on that side rather than anthropics so there's all sorts of weird i don't maybe apple is concerned about going too far deep with one partner and becoming beholden to them right um and they worry that if they go too far down the open ai rabbit hole that yeah they could either get you know screwed one way or another be it um because they feel that they have to use them all the time going forward and and they would rather sort of have these entities play off one another um and again like i would anticipate that in in a final state if they did do a partnership they would probably ideally want to use a few of them and that was obviously the rumor with with google as well integrating gemini um but if if they had to pick one that is surprising that they would go with Claude ahead of ChatGPT.

18:12So again, I'm just trying to ping pong why they might have gone with them or the rumor has it. I think they probably, if they've met with both leadership teams, I think they probably are a company that would absolutely go for Dario's safety branding. Like it's the safety company with the privacy company versus OpenAI, which is kind of like the chaos product company even though the product's working better but the the um the terms of the deal to me are also fascinating so uh you you wrote about this also uh that or that anthropic is asking for a multi-billion dollar annual fee that increases sharply every year now apple's biggest acquisition in history was three billion dollars so it's not exactly a company that is sort of ready to spend$2 billion a year or more on an integration partner.

19:06So what do you think is happening in that negotiation? Yeah, I mean, that was maybe the weirdest thing about that entire report was, you know, the notion that both Anthropic might even just be asking for that because they have to know that Apple, you know, will be super reluctant to do anything for the reasons you're exactly talking about, right? They just they historically don't spend money, even though they have all the money to spend. They don't spend money, certainly on acquisitions. But if this this could be really larger than their largest acquisition, which is wild. I mean, from the flip side, I think Anthropic probably, you know, assuming this is accurate, they're looking at it like, I mean, if we really are the main focal point of Siri going forward, that is going to absolutely blow up our servers to an extent we haven't seen, at least on the sort of user facing consumer side to date.

19:59And, you know, we see what happens or, you know, we hear reports about what happens when sort of these viral integrations happen on the ChatGPT side and how that brings down their servers, blows up the cost and everything. And so does Anthropic really want to be in that boat, you know, without any sort of life vest, you know, in the form of payments? That was always one of the most surprising things to me about the Apple and ChatGPT deal was the reporting that it was no money exchanging hands, right? Because it felt like, too, there was a real possibility that if one of those things went viral, that it was the weird situation where Apple would be causing such an influx of usage of ChatGPT, which at the time, at least, and presumably still mostly is running on Microsoft servers.

20:48And so basically, Apple would be, you know, causing these huge increases in cost to OpenAI and Microsoft by way of their deal. And if they didn't have any, I was always trying to like, think through, there must be a way that they're going to eventually pay them. And I do think that that's sort of what led down that, you know, at the time rumored and then rumored to be off potential investment that Apple was going to do in OpenAI. To me, that made all the sense in the world because that was a de facto way for them to pay for, you know, that integration. Basically, they would just be investing billions of dollars into OpenAI potentially.

21:27And obviously that didn't happen. But I almost wonder if something similar isn't at play here with Anthropic, where, look, they might say, we're not going to give you whatever,$2 billion a year, say, to use Claude. But, you know, you guys always need more money. You're taking investment. We'll happily invest, you know, in the company, knowing that their one thing that Anthropic seemingly wants is diversified investors, right? Because they don't just want to have Amazon owning so much and they don't just want to have Google owning so much. both of them own quite a bit right now. But if they could diversify that even further with Apple, I think that they might be interested in that.

22:03Another question of if Amazon and Google would be interested in that. But still, I could see it maybe going down that path. But otherwise, it's pretty shocking that they would ask for that from Apple. I think you're totally right about this. And if you are a client of Anthropic, whether you're using Claude off the shelf or whether using the API, one thing you know to be true is there are often lag times. And often, And if you're trying to tag into the API, it can be down. And I think a lot of people are saying, you know what, this is server capacity. So now imagine you plug that company in with Apple and then all of a sudden the product is good and demand goes through the roof.

22:40It just effectively, you know, destroys Anthropics ability to deliver its products. So can I ask you this? Like, what do you think it says about Anthropic that this is the biggest opportunity, biggest potential business opportunity of its lifetime to be like the de facto voice of AI on the iPhone, sort of surpassing that experience that we're both talking about with ChatGPT, but they just can't do it. I mean, assuming we're right about this, but they just can't do it because of servers. Yeah. I mean, and right. Right. So I thought I think about that as well, too, because basically Apple, again, assuming that the reporting is accurate here, Apple is handing them the golden ticket.

23:20Right. Like they for since the since the dawn of these LLMs, you know, basically ChatGPT has had the crown and has been able to sort of keep going and expanding, you know, their lead. Obviously, Google and Gemini are giving them a run for their money because they're putting it and baking it directly into search and Android and have incredible skill in Gemini or sorry, in Chrome to some extent right now. And so they're leveraging those strengths that they have in order to do that. But still, I think we would all agree that that ChatGPT has been sort of the leader in the clubhouse, you know, at least when it goes to sort of consumer facing versions of this stuff.

23:57And basically, Apple, if they were to give this to Anthropic, I mean, would they shoot ahead of ChatGPT in terms of actual usage? I don't know if that's the case. I do think that they would shoot ahead of them in like simple queries like we're talking about, you know, like what's the weather and that type of stuff, assuming that Apple outsourced that as a part of that deal. But, you know, we're talking about they would be in the hands of billions of of consumers overnight doing these types of queries. And so to your question of like, like, what's the reluctance to do that? Or why are they so reticent to that?

24:34I do wonder if part of it is, I think recently they gave some, some quotes in different interviews, talking about how they're sort of okay with seeding the, you know, the chatbot market, right, to, to ChatGPT. And it feels like that's not the path that they want to go down anymore, at least the main primary focus there, they want to do coding, and they want to like focus on their strengths. So it's almost like if they already feel like they made that decision internally and now Apple's potentially handing them, again, this golden ticket to sort of shoot back into the forefront of consumer, maybe there's some internal reluctance to sort of go down that path a little bit.

25:12Yeah. And it just sort of goes to like where you can make money in AI today. And I think there's like two different areas. There's the product side where you have ChatGPT and then you have the API side and Anthropic is doing quite well there. I don't know, would you call this like a product or an API play because you're going to give a version of Claude to Apple? I guess it would still be a product play. But I'm curious like what you think about the state of the AI business today, because obviously we have like a lot of applications that are out there that are doing well. But the API business to me still seems like a bit of a question mark, even though we've seen Anthropoc posts that like they went from like 1 billion annual recurring revenue, I think to four in like just a couple months.

25:53And yeah, seems like they are mostly API. So what do you think is the state of play right now? Yeah, I mean, that's still it's feels like a pretty risky business, right? Because it's so unlike what OpenAI has done with ChatGPT, where they have a brand now and like a known entity, the APIs like those are relatively not necessarily easy, but they're easier to just swap in and out different, you know, different tools and technologies as they become more readily available. And then, yeah, the cost will get sort of driven down over time, you know, as that happens, is there's more competition in the market.

26:27So it does feel like that's a bit, yeah, more vulnerable of a state to be in. And I mean, even look at the vibe coding space that they're in, right? It's like, they, if I have it right, they are predominantly powering Cursor right now, right? And some of the, and the one as well that OpenAI is buying, which they famously are now cutting them off windsurf yeah um from the newest uh versions of the models and uh the question becomes how much does cursor uh you know the company any sphere start to build their own stuff and obviously they're doing some of that internally um but you know that's a real risk point of course for for anthropic um and so it becomes like a point where do they really have to more rely on building, you know, cloud code, making sure that that's its own product.

27:18And that's the product side again, versus sort of going down the full on API path. But it's still like so much of the stuff is moving so fast still. And it feels like, you know, we weren't even talking about, quote unquote, vibe coding tools a year ago. And now, you know, maybe we were a little bit, but now that's all that anyone seemingly wants to talk about, certainly on the business side and the fundraising side. And so there will be something else that comes up in the next six months, some other layer of AI that everyone wants to talk about is the new thing from the model perspective. Yeah, this is a good moment for me to plug the Vibe Coding episode we have coming up.