Apple 的 WWDC 退守、Liquid Glass 与一个问题，Meta 投入 148 亿美元推动 AI 重启

Hello, and welcome back to another episode of Sharp Tech. I’m Andrew Sharp, and on the other line is Ben Thompson. Ben, how are you doing?

I’m doing okay, Andrew. This is the first week of the transition back to Wisconsin for the summer. It’s been a busy one.

Always an adventure.

A little hectic. WWDC is this week, so I had to write about it. I literally just finished recording my Thursday interview, which I call out because I talked to Patrick McGee, who wrote Apple in China.

I’ve talked to other authors, and they might have good books, but I’ve never given this level of endorsement. Everyone listening to this podcast needs to read this book. It’s one of the best tech books ever. It’s one of the best books about Apple, and it’s one of the best books about China. It’s fantastic. Everyone can listen to the interview if they like, but definitely go get the book.

There you go. I can’t wait to listen to the interview, and I can’t wait to double back and discuss it at some point in the next week or two. Your offline endorsement came through on iMessage before the interview. You said it was the best book, that it may be the best book ever written about Apple, and that it may be the best book ever written about China.

I had purchased it and hadn’t gotten around to actually reading it, so that was the kick in the ass I needed to get on it over the next month or so. I look forward to diving into all of that.

I’ve also selfishly determined—I think it’s at number 7 on the New York Times nonfiction list—that I want to see what pull we have. Let’s see what pull we have.

Let’s—

Bump it up.

Let’s give it the Stratechery bump.

That’s right.

The Sharp Tech bump. Absolutely. Audience, you have your assignment.

It was actually a bad idea, because I think it came out like three weeks ago, so it’s probably in a natural, you know, coming down, so I’m going to be unable—

An app—

—to resist to—

Yeah—

—to stop that, but what are you gonna do?

Well, let’s spike it back up there.

That’s right. That’s right.

You know? I have faith in the Stratechery audience. For now, we are going to begin with WWDC this week. I’m coming to enjoy WWDC as sort of a mile marker for the podcast. I found myself thinking back to past recaps we’ve done.

Two years ago at WWDC, we were reacting to Apple’s grand plans for the Vision Pro and a future built around visionOS.

Oh, Vision Pro. I was trying to remember what that was. Yes, that’s right—the headsets.

Back then, I was gently but correctly skeptical of the whole affair, and then last year—

The Vision Pro is still cool. I put it on again because John Gruber had his live talk show, and it was broadcast in Vision Pro, so I wanted to check it out. Every time I put it on, I like it.

It was interesting, actually. I didn’t write about this, but the WWDC segment on Vision Pro was pretty cool. They added a bunch of stuff. A lot of it addressed things we were complaining about back then, including a better guest mode for other people and the ability for 2 people with Vision Pros to watch the same thing.

Mm-hmm.

Imagine that: watching a movie together with a headset on. There was a lot of enterprise stuff, and it does feel like Apple is saying, “Look, if you need it for a specific sort of task, companies will buy it for their employees.” That’s where Apple is definitely leaning into it.

And maybe they’ll come—

Maybe they’ll come to like it.

And maybe it’ll gain adoption that way.

Yep.

I think you were right on the money a year and a half ago with that take, and a year and a half later, Apple is right on time with some sort of direction for the Vision Pro.

Last year, we were focused on Apple Intelligence, of course, and how elegant Apple’s strategy was, because they have access to all the most important data that AI can utilize for customers. And, oh my God—

Is this a litany of Ls for me? Is this—

Apple is so brilliant.

Is this something that’s framing up?

No, look, I agreed with you back then. That WWDC knocked my socks off, and I was like, “Wow, Apple’s going to be the one that wins here.”

Obviously, that turned into a prolonged corporate catastrophe for them. This year, it seemed like the big idea was that there was no big idea—no revolutionary step forward. In your article on Tuesday, you called it one of the more encouraging Apple keynotes in a long time. So tell me why it was encouraging.

Because they overdid it. They went too far. There’s the Apple aura in general, and it kind of ties into this book, actually, so maybe it’s just top of mind.

A big theme in this book is how suppliers will work with Apple knowing they’re not going to make any money, because Apple is basically going to teach them and fund them to learn how to do completely new things that have never been done before. They can then take that knowledge to other suppliers or other customers and charge them much higher rates. So it pays off in the long run. That’s definitely an important theme.

But Apple had the benefit of the doubt last year. They actually had the benefit of the doubt with the Vision Pro. If anyone could pull this off, it’s Apple. The Vision Pro, again, was awesome. It remains awesome. It’s just $4,500 and weighs like 4,500 tons.

Well, look—

And there are no applications for it, but it’s still very cool.

It’s an amazing piece of hardware. There’s no question about that. There just hasn’t been much content, and there hasn’t been much software. There are literally 2 or 3 times a year when I’m like, “Man, I would love to have a Vision Pro.” The talk show is one such occasion.

The Apple Vision Pro is the exercise equipment of computing devices.

It just sits there in your living room—

No, what I mean is, after you exercise, you’re like, “Wow, I feel great. I should exercise more often.” In the entire lead-up to exercise, you’re like, “I don’t want to do this. This is terrible.”

Mm-hmm.

Although it’s probably worse than that, because you don’t even think about it, right? It’s the treadmill down in the basement in storage.

I was going to say, it just sits there taunting you as something that you spent thousands of dollars on, and it would be useful to use, maybe, and yet it just sits there making you feel guilty the entire time.

It does. It does. But yes, last year was AI. Apple gets the benefit of the doubt. They present a story, and this is probably a weak point for me, because I do look at things structurally and at where companies are.

This idea that Apple owns the device, has access to your private data, can provide an experience that no one else can, and then hand off broader tasks to the large foundation models—it’s elegant. They can maybe set up an auction to have access on the device. It’s like a replay of the Google Search deal. They’re the AI aggregator.

All these ideas that we talked about make elegant sense. The problem with all of the strategic thinking—and I put myself in this basket, critiquing myself—is that you still have to execute. I think if you listen to our podcast, and if you go back to what I wrote, we said, “Assuming they pull it off.”

Right.

But we didn’t say it with our chest, or at least I wasn’t saying it with my chest. It was sort of, “Yeah, assuming they pull it off, and they’re Apple, they’ll pull it off.”

In retrospect, there wasn’t enough acknowledgment that what Apple was selling at last year’s WWDC was, in fact, ahead of the state of the art in AI capabilities.

Right. That’s exactly it.

No, that’s why I’m not giving myself an excuse. I should take the L. That was John Gruber’s point when he wrote something that was right out of the city of Cupertino. He opened by being mad at himself. He was like, “How did I—They couldn’t even demo this stuff. Why did I assume that they could pull that off?”

When you put it in the context of what they were promising, it was an experience that OpenAI couldn’t deliver, that Anthropic couldn’t deliver, and that Google couldn’t deliver.

Mm-hmm.

Particularly the Google one, because Google has access to your private data. Why was Apple going to pull this off?

And it’s even worse than that, because it’s not just the general comparison to the AI companies. Apple has historically and culturally been bad at this kind of thing. They’re not good at dealing with a gazillion edge cases or failing gracefully.

Mm-hmm.

Apple is good at building finished products. They build operating systems that have to ship. They build phones. An even better example is that, in September, that phone is going to be in a box, on a shelf, in a customer’s hand, and you don’t get to go back and iterate it.

And the OS has to be ready at that point.

“Oh, we can change it to be better.” That’s right. That’s right.

Yeah.

It’s just a very different way of working, and I think companies that are good at one of those ways of working are intrinsically bad at the opposite. And again, this is why I have to take the L. I’ve written exactly that about Apple, that they’re bad at those sorts of things. I just gave them too much of the benefit of the doubt, in violation of my own previous thinking and principles, and in violation of the broader context of the industry.

Mm-hmm.

And in retrospect, given that, for Apple to get up there—

“You were right. Let’s take a beat.”

They say, yeah—okay, this: “We look forward to talking about XYZ in the coming year.” At least they admit it up front. And then everything else was stuff they could do, and everything that they showed, they demoed. It was just sort of a, “Look, before you try again—”

Mm-hmm.

You have to reset and get your house in order. This felt like a getting-your-house-in-order sort of keynote. And yes, there was the big UI redesign, and it’s going to be a mess. People are going to complain about it. This happens with UI redesigns.

I’ll withhold judgment on it. You probably should withhold judgment, too; it might even be a couple of years. iOS 7 was really rough. It wasn’t until iOS 9 or 10 that it really settled in. This is going to be the nature of it. But this is also what Apple does. Apple makes user interfaces, right? They’re in their wheelhouse.

Yeah.

And that’s not very exciting. It’s not very sexy. But Apple needed to be not exciting and not sexy because they went so far off the reservation in terms of execution over the last year.

Among the things they announced, they are in fact opening up on-device AI models to developers, exactly as you’ve been recommending for the past couple of months on this podcast. Now that it’s official, can you remind people why exactly you endorse this approach for Apple?

The on-device model is interesting. Just to be clear—and this goes to the obvious things that still need to be said, and should have been more strongly said last year—an on-device model is going to be vastly inferior to your ChatGPT experience—

Anything in the cloud.

—that’s running in the cloud on massive NVIDIA GPUs. That’s just the physics of computing. It’s running on your phone in your hand, on a battery. That’s not comparable to a $50,000 NVIDIA GPU, or a whole host of them in the cloud.

Mm-hmm.

And it was fundamentally flawed for Apple to anchor so strongly on that because they’re basically committing to delivering a worse experience because it’s not capable enough. They’re dressing it up in privacy and these sorts of things. It’s also a big cost savings. It’s like, “Oh yeah, it’s because it’s using the user’s power, right? It’s using the user’s chip that’s already there—”

Yeah.

—and I’m sorry, Apple, you don’t get to save money and sell it as a feature, right?

Well, when we say Apple anchoring on that, they were going to deliver all the on-device AI, and it was all going to have Apple’s name on it, and then perform worse than anything that’s operating in the cloud. That’s the core problem with that.

There is a point that’s worth bringing up about that, which I’ll come to in a second. But the thing with letting developers do it is that, in the most idealized version of the App Store, you have the one-person shop, and they’re just shipping an app on the App Store. The App Store is amazing and lets them build a business doing whatever it might be.

A great example is Marco Arment. A lot of people listening to this might be listening on Overcast. He makes—

Mm-hmm.

—a podcast client. Someone like that can’t use AI in the cloud. It’s not compatible with their business model, to say the least. And even the expense and expertise required to do something compelling, if you’re only one person, you’re not going to pull that off. But Marco, for example, one thing he would like to offer is transcription of podcasts. We’ll see how it works once he gets access to these models, but he could theoretically offer it now that it’s free.

Yeah.

In this case, the free isn’t Apple saving money. The free is that developers can do stuff they couldn’t do before, or couldn’t afford to do, when everything had to be in the cloud. Because they have access to a local model, they have access to the user’s power to pull these sorts of things off.

And the great thing from an Apple perspective is, suppose Marco builds this and it stinks. Marco’s got to deal with the bad reviews and people being grumpy.

People complain about Overcast.

That’s right.

Yeah.

It’s not Apple’s problem, right? If it’s awesome, suddenly, oh yeah, the podcast players on iPhone platforms have transcription. The ones on Android, maybe they do if you have a high-end Android. If you have a low-end one, it’s going to be iffy, right? You have this whole universe—

Mm-hmm.

Android covers everything, so you don’t get a consistent user experience. You don’t know what you’re targeting. Apple only sells a few iPhones, and they’re pretty high-end, so you know if you get one, you’re going to get a better experience. Their upside here is that you have differentiated apps on your platform.

And again, maybe the vast majority of the use cases—maybe it will turn out these on-device models are just insufficient for everything. That’s fine. There’s no harm done. People are going to experiment, and they’re going to find weird things that you never thought to use AI for, but you can experiment because it’s free.

Well, strategically, that’s so much smarter for Apple because you’re weaponizing the long tail to your benefit. And the one thing that has come through with the last couple of WWDCs is that Apple, left to its own devices and its own ideas, doesn’t really have any idea what to do with AI. Image Playground is a punchline.

It’s something.

But again, every time they come back to Image Playground, I’m like, “This is a mess. Who does this exist for?” And then merging emojis and stuff like that. There were some groan moments during WWDC, and I trust a universe of developers to come up with cooler, more experimental ways to use AI and optimize it than I trust Apple at this point.

Yeah.

And so, on that basis alone—

Unfortunately, Apple doesn’t trust them.

—it’s a no-brainer.

Um, now the one thing that—

But now they do.

Well—

Sort of.

So—

The way Apple does general-release features, whether it be a new API or a new capability, is that they usually do it themselves first. They dogfood it. They have to build with it.

What does “dogfood it” mean?

It means you use it yourself. Say there’s an API here.

Okay.

There’s an API to use this language model. Apple was working against that API for the last year, year and a half, to build the features that they wanted to build, if it’s Polygon or whatever it might be. So they probably made changes to that and said, “Oh, this actually doesn’t work well for a developer doing XYZ.”

Then, a year later, it’s broadly available, and the problem with making something broadly available is that you have to support it forever. It’s very possible Apple actually always intended to make the local models available this year. It’s actually historically in their cadence to first do it themselves for a year.

Part of that is, yes, they want to get the benefit, but there are actually legitimate platform reasons for them to do it first. It was always going to be available to developers now. But that said, I’ll take the W. I started out with a string of Ls, so if we want to sit here and say, “Yes, Apple listened to Ben—”

There you go. Well, take the W. I feel like that’s a sensible strategy on your part. And as far as Apple’s strategy, I do look back: It was in keeping with their newfound control-freak ethos to not open the on-device models. Who knows whether this was all part of the plan or not, but the point is they’re finally headed in the right direction on that front.

Right. And to be clear, this isn’t going to change the world, right? These models do still stink. We should all be honest about that. I think there’s a bit where everyone’s like, “Oh, finally they’re doing the right thing.” Let’s be clear what is happening. But yes, anyhow, continue.

The right thing that may or may not move the needle—

Right. Exactly.

is, I think, the proper way—

Exactly.

to think about that—

Yes.

this week.

Yes.

Look, it was a boring WWDC, a sensible WWDC, though—

Sensible. That's the word.

relative to the last couple of years.

No, that's the word. That is the word. It was sensible. And you know what? Apple needed a bit of sensibility.

Indeed. A couple of other notes here. They're deepening their integration with ChatGPT, including ChatGPT animations into Image Playground. With iOS 26, Apple's Visual Intelligence feature will be able to analyze images and text on an iPhone screen. Users can then ask ChatGPT for more information about whatever they're looking at.

Sorry, just real quick: We have a little bit of Apple copying Microsoft here. Microsoft figured out that the year-version number makes everything a lot easier. It's actually a shame Microsoft got away from that. They were onto a good thing.

Well, is copying Microsoft actually a good thing? I think that might be its own little red flag, isn't it?

That's a good question. Partnering—well, they're copying Microsoft in partnering with OpenAI. This, I think, is really interesting.

Okay.

In some respects, this is much better than before. It's not every single time, “Do I do it on-device? Do I do it with ChatGPT? Do I ask the user?” It was just so much of a mess. If you're doing this Visual Intelligence where you take a screenshot or whatever, and it's on the screen, you can ask right there for more about it. It doesn't try to answer and then say, “Oh, I failed. Give it to ChatGPT.” It's asking—it’s like, no, it's right there. Ask ChatGPT.

Yeah.

Number one, the appropriate level of humility about your capabilities. Number two, if you're a regular ChatGPT user, suddenly the iPhone is that much more attractive. Because if you're—

Mm-hmm.

—signed into ChatGPT on your device, you do a Visual Intelligence query, you can go to your computer, open up the ChatGPT app, and continue the conversation. It's there. It is giving something very valuable to ChatGPT, but it's also taking something very valuable, which is, to the extent ChatGPT becomes a driver in “What phone should I buy?”, an iPhone is better than Android. Android's going to have—

Mm.

—all the Gemini stuff, right? Now, the level, the deepness, and the number of features is probably going to be faster because Google's fully integrated. They're doing it all. But if you're super committed to ChatGPT, and that matters in your phone choice, suddenly there's actually a compelling reason to choose an iPhone because of AI. It's not their AI; it's OpenAI's AI, but that is actually—I think this was underrated. People anchored on the—oh, Ben, you called it—the local-device models. I also called for replacing Siri. They're not quite at replacing Siri—

Yeah.

—but this is a very significant step, I think, in that direction, and hopefully it just accelerates.

Well, it also goes back to your take on Apple. For 20 years, their selling point has been delivering the absolute best experience to users, and users don't want to sit there and have ChatGPT be the backup option if the model fails on-device. Just give me ChatGPT. So it seems like, again, a much more sensible way to handle all of this.

Yep, and a much better partner than Google because ChatGPT is winning in the consumer space. That's the horse to bet on.

That's the thing. What users want is ChatGPT, you know? So don't pussyfoot around.

I have a take on—what is it? This is definitely tortilla chips, but I think I coined this in the context of some other food item. I can't remember. It was like a tortilla, but it was a regular food item that you think people would like. I'm like—

Mm-hmm.

—I love that food, but it is a flavor vessel, in that it was something that you dipped. I can't remember what it was. Tortillas are the classic flavor vessel.

Sure.

And this is a little humbling for Apple, but honestly, what I want is a ChatGPT vessel.

Yeah.

So to the extent that my phone becomes—

They're the tortilla chip.

No, Apple's the tortilla chip—

Yeah.

—and OpenAI has some tasty salsa. Give me some more.

Exactly. OpenAI is the dip that everybody is signing up for, and Apple—excuse me—you’re just the delivery mechanism, but that could be very profitable in its own right. All right, a couple more notes here, bullet points. They announced a new Games app, which perhaps lays the groundwork for Apple to bifurcate in-app purchasing on games and the purchases made on other apps. This is another crusade of yours, a 10-year-long crusade. It looks like they might go that direction, but perhaps they're too late at this point, given some of the adverse rulings that have been handed down. Any quick thoughts on that front?

I put it in there as a retreat. Sometimes the retreats are not a function of—

Forced.

Yeah, you're getting shot in the back of the head. It was interesting that all the—It looks like the Takings Clause is not going to carry the day, as it were. The appeals court denied Apple's request for a stay of Judge Yvonne Gonzalez Rogers' injunction, which basically said steering's gone. It is what it is, and I'm definitely fine with that. I think the steering thing was terrible.

What I would've been fine with all along is no steering provision for the vast majority of apps, but games are as locked down as could be. And again—

Right.

—that's the way the gaming industry has worked, right? Consoles have that model.

That's the way consoles work.

Yeah.

Yeah.

That's the way Steam works. The specifics might differ with the steering thing, but consoles for sure are super locked down, and Apple's insistence on it is not enough. And by the way, the games are kind of skeevy, right? I mean, sorry, game developers. I apologize every time I say it. I'm sure you're good people, but this whole, “Let's gamify things. Let's get people to find some digital whales.” You're calling them whales. That might be a bit of a red flag there. People who are just putting in tons and tons of money to get more gems or whatever it might be.

At the end of the day—and honestly, Epic falls in this too—Epic, it's cool that you got people to buy dances and clothes for your character or whatever it might be.

We're not shedding tears for the people who are shaking down 12-year-olds all over the world.

Right. And so this was always the solution to me, and that's where most of the money came from. Keep all the limitations there. Frame it in something like, “Oh, we're looking out for the kids,” or whatever it might be, even if you just want to keep all the money, and unleash it. Let people who want to actually sell on your devices, who want to have productivity software—Kindle wants to sell some books—let go. Let them do what they do.

Mm-hmm.

Now, the Games app is not there yet. It is, though, another app store. You can get games in the Games app. You can also get them in the App Store. But it is at least laying the foundation to potentially go in this direction, and in the U.S., it's going to be too late. You're going to have to allow the games to steer because you didn't do this when you should have.

Right, because you've lost so hard that you've now lost control. Speaking of productivity software, the other 2 notes: They're revamping iPadOS to allow people to use it more like a Mac with multiple windows, and then there are enhanced Continuity features that will help people work seamlessly across their Apple devices, with one example being the ability to monitor a mobile Uber Eats order from your Mac. Do you have any takes on those developments?

I can't fully evaluate that one because, as I mentioned, I'm not in Taiwan currently, where we made approximately 37 Uber Eats orders a day. It actually might have been a distraction—having that pop in my menu bar every time. You have not experienced the power of food delivery until you've lived in a very dense city with people on scooters. Want a boba tea?

It sounds fantastic.

It is. It's fantastic.

You've been selling me on that vision, and I'm not jealous of my friends in New York.

I've been here for 48 hours, and I'm going through withdrawal from not having Uber Eats. Believe me, it is rough out here—

Well, listen—

—in the suburbs.

Get out there. Speaking of tortillas, get out there. Get yourself some Chipotle and experience the majesty that the United States of America has to offer. Savor it.

It is true. I just need to go do Chipotle and it's God bless America. That is absolutely correct.

Do you have any takes on the iPad changes? People have been clamoring for that for about 5 years now.

I like—

Not me. I hate the windowing. To me, that’s actually what I stopped using. I use the iPad to do my drawings on Procreate, and that’s basically it. I actually barely even use it for watching sports anymore because the stupid thing’s never charged. There’s been some bug where it doesn’t sleep properly and the battery drains, and it—

Hmm.

That bug’s been there for ages. Maybe it’ll be fixed this year. Who knows? But I’ve kind of given up on the iPad. I wrote a lot about the iPad. I was very passionate about it. I think some of that is reflected or downstream or related to how Steve Jobs was very passionate about it. One of my favorite keynotes was the iPad 2 keynote, where he demonstrates GarageBand. The look of satisfaction on his face—“Now anyone can make music”—is Steve Jobs’ philosophy in a nutshell.

Right, the platonic ideal of what tech can be, and that sort of enthusiasm.

It was. And I wanted more apps like that. I was so frustrated with the App Store and its policies because it was foolhardy to try to make a complex app that took advantage of what it meant to be an iPad and took over the whole screen, because the business model wasn’t there. Even if you found your users, you couldn’t monetize them over time. This was before there were subscriptions and things like that. You just—

Mm-hmm.

You could sell one time, and that was it. Apple, you’re not doing even the bare minimum. You want to control everything, and you’re completely killing an entire part of your economy that would be great for you, because people who use productivity apps will buy the latest device with the most power, and they’ll upgrade all the time. You can’t be bothered to help them because you’re busy harvesting all this money on gems in games.

Yeah.

This is the frustration. And this is why there’s relief this week. Apple backed up on something. It just feels like they’ve been going to the extreme on so many things. Maybe this is a false flag. Maybe they just had no choice, and I’m going to look back on this article and say, “Wow, that was wishful thinking.” But you know what? If Apple’s an alcoholic and they turn down a drink one time, I’m going to hype the crap out of that. Good job.

It’s still progress. Yeah.

Yeah, seriously.

Standing ovation.

Absolutely. So was this a boring WWDC? Yes, and good. We needed some boringness. For the iPad, I feel like Apple just gave up, and they’re like, “Okay, well, let’s make the iPad more like a Mac, I guess. People like Macs.” And that’s been terrible. They’re trying to do the swiping—

Well, that’s what I’ve—

—edges and stuff.

There’s a very loud contingent who loves the processor on the iPad and loves working on the iPad and has been lobbying to turn it into a Mac for close to a decade now. It seems like they’re just offering a sop to those people. It’s hard to get too worked up in either direction on that front.

Yeah, like I said, I’ve given up, so whatever. They want it, go ahead. But Apple, if you’re going to make the iPad into a Mac—and it’s starting to look an awful lot like one with these multiple windows, overlapping windows, and all these sorts of things—how about, given that we have an iPad that’s like a Mac, and you can use a pointer and all these sorts of things, and it’s locked down because it’s an iOS device, we loosen the reins on the Mac then? For those of us—

Hmm.

—who actually want to use a proper computer, stop trying to turn that into an iOS device with all the gazillion warnings or whatever. I was getting my tennis match from you—the incredible French Open. I was on a plane, so I couldn’t watch it.

Oh, Alcaraz.

Yeah.

Yeah.

Because of this whole Channels thing, I can get the file from your computer and put it onto my computer, so I can watch it later. This is a back-end Mac mini that’s running on a server. In my case, the storage is on a separate device, a NAS device. I’m in the Terminal, where you have to run this. By the way, I’m running this via my phone because I’m very eager to get this onto my computer.

It’s not running, and I can’t figure out why. I have to open up a screen-sharing application to the Mac mini, where there’s a pop-up warning saying, “Do you want to allow Terminal to access external devices?”

Oh, boy.

I’m in the Terminal. I don’t need to be asked if I want to access external devices. Stop treating me like I’m one of these children who wants to have windowing on an iPad. No, I’m just kidding, guys. But no, seriously, it’s very—

Paternalistic Apple, man.

Oh, it drives me bonkers. So yes, if you want to make the iPad—if you want to go down this road, again, all the iPad people are surely infuriated with me at this point. But if you want to make it like the Mac, then let the Mac be like the Mac and not like an iPhone. Thank you.

Mm-hmm. Fair enough. Personally, I love the iPad exactly the way it is, in between an iPhone and a Mac, and I have not been one of the people clamoring for windowing. I like the single-window approach on iPads. It’s like a throwback to technology 10 or 15 years ago, and it’s a great way—

It’s a throwback to using full-screen Word as a lawyer on your PC—

—and having an attention span, you know? Just focus on one thing. I don’t have any notifications on the iPad. It’s my favorite single piece of technology hardware I own, and I like it exactly the way it is.

But I’m happy that all the people with pitchforks clamoring to optimize the processor and do X, Y, and Z with the iPad can have a great time with their iPads over the next couple of years. And then the final issue here: the sexiest announcement at WWDC was this new OS and new UI called Liquid Glass.

Two emails on Liquid Glass. Rich says, “Ben and Andrew, as a disgruntled materials engineer, I just needed to put this correction on the record. There is no such thing as liquid glass. Glass is an amorphous solid formed by quickly cooling from the liquid state before the molecules form a long-range crystalline structure. Marketing BS will be the death of me and every other engineer out there.”

Justice for all the disgruntled and beleaguered materials engineers out there who may have been taking in WWDC on Monday.

I have bad news for Rich about the reality of everything that you’re seeing on your computers. Mostly, it’s not real, so I think it’s totally fine to call it Liquid Glass.

I just enjoyed the passion from Rich.

It was a good email. Yes, it was a good email.

So I had to give him a platform there. Kim says, “Andrew and Ben, I agree with Ben’s quote from Sebastian De With in his update on Tuesday. Apple’s new UI, Liquid Glass, really is like something we imagined the future to be from popular culture. In fact, when I first saw it, it reminded me of nothing so much as what a great Hollywood production designer would produce for the user interface that populates the world in a high-budget sci-fi movie from a major studio. This is neither praise nor condemnation, simply an observation. I’ll leave it to you guys to tell me what it means.”

So, Ben, what do you think of Apple’s approach and the new science-fiction UI here?

I don’t know. I’m going to reserve judgment. It’s one of those things you can’t know until you use it. I think I mentioned earlier that, in reality, you’re not really going to know what it’s like until we get a year or two in. For the next year, it’s going to be rough. If iOS 7 is any indication, there are going to be a lot of complaints. People are going to post the most egregious screenshots on Twitter of where it just completely falls apart, and that’s going to be the reality.

I’m not going to use it because I don’t install betas. I am a tech analyst. I am not a product reviewer. I don’t need to suffer with short battery life—

You’re not going to harm—

—your computing experience.

Well, it’s funny because it got a lot of attention. I was watching the event and came to a lot of the same conclusions you did in your article: this understated approach from Apple is probably the only viable move on the board for Apple at this point. It’s probably a step in the right direction if we’re ultimately talking about a bunch of incremental improvements to the user experience on various Apple devices, leaning on developers to help solve some of the AI problems, and pushing this Liquid Glass UI that, at least in the videos, looks like it’s going to create the same sort of tactile joy that all the best Apple products have always given customers. And so—

Right, but we learned from a year ago not to trust Apple’s videos.

That’s what I was thinking. I was sitting there watching it, and I thought, “Maybe this is the cost of some of the decisions that Apple has made over the last couple of years.” Normally, I would be pretty psyched about this, but watching it on Monday, I was like, “Will any of this actually work?”

I’m still pissed off about what they did to the Photos app, and I just don’t know.

None of this works as well as it used to. Your rant a couple of weeks ago about the settings situation—

Oh.

On the Mac—

Oh, God.

That generated some real passionate feedback and agreement among the listener base. I would say that even with Apple's strengths, they still have more to prove going forward here in order to realize some of the potential we saw Monday.

For sure.

And if Liquid Glass is not just teething pains but actually horrible, like the Settings app is horrible, that is interesting in its own right. It's not just that Apple can't do the stuff that it's traditionally bad at. What if it can't do the stuff it's traditionally good at?

Right.

So, like I said, that is going to be interesting. I did speak to John, and he really likes it. He's pretty persnickety about user interfaces, with the caveat that there are lots of rough edges around the corners, and he said, “They might have shipped it a year early.”

Mm-hmm.

It also felt like, “We have nothing to talk about, so let's ship this a year early.” That might have been going on here a bit. But in all those futuristic bits, there are people griping, “Do you know how much processing power it takes to do this?” It takes a lot.

Yeah.

And that's why Apple can do it, because they make their own processors, make their own graphics, and have all the subroutines to do it efficiently. That's great. They're doing what they're good at.

Yes.

Hopefully, they're still good at it.

We'll see if they're still good at it.

Yeah.

Yes. For every Apple event, our colleague Dominic Angulo runs a prediction contest with 10 questions about what will or won't be announced, and all the members of our tech group chat participate. For the second consecutive year, I emerged victorious as the winner—

Tech guy Andrew Sharp.

—of the prediction contest.

You've just been sandbagging us for years. It's amazing.

That's right.

Congratulations.

Lest there be any doubt—

Yes.

—about who's the normie and who's—

I'm stacking the Ls.

—you know, the expert on this show.

You're stacking the Ws. It's true.

I tied with ChatGPT, so I guess ChatGPT is also the expert there. Speaking of AI, Meta and Scale AI: I have to read a story from The New York Times.

“Meta is preparing to unveil a new artificial intelligence research lab dedicated to pursuing superintelligence, a hypothetical AI system that exceeds the powers of the human brain, as the tech giant jockeys to stay competitive in the technology race, according to four people with knowledge of the company's plans.

“Meta has tapped Alexander Wang, 28, the founder and chief executive of the AI startup Scale AI, to join the new lab, the people said, and has been in talks to invest billions of dollars in his company as part of a deal that would also bring other Scale AI employees to the company.

“Meta has offered 7- to 9-figure compensation packages—9 figures—to dozens of researchers from leading AI companies such as OpenAI and Google, with some agreeing to join, according to the people.”

Some of these researchers are getting Jordan Poole money to join up with Meta. Very impressive stuff. The numbers that have been reported around this deal are $14.8 billion for a 49% stake in Scale AI and an agreement to install Alex Wang to a leadership position inside Meta. We've got a couple of emails on this, but generally speaking, this was the other big news in tech this week. What do you make of what's happening here?

My first takeaway is that I've told this story on the podcast before about this moment I had when Apple bought Beats. It seemed like this huge deal, and there were a gazillion stories about it. Of course, I was writing about it, so it was very exciting. Then I realized that Apple bought them for $3 billion, and Apple was worth, I think at the time, around $600 billion.

Mm-hmm.

It just really didn't matter at all. It wasn't a big deal. I think there's a bit of that with this AI stuff. Meta's spending, what, $65 billion or $70 billion on CapEx this year? All the tech companies are in the high eight figures here. In that context, $14.8 billion is a lot of money—let's be clear.

Yeah.

But in the context of spending all that money on GPUs that are going to be unusable and obsolete in a few years, in the grand scheme, it's not that much. That's just crazy to think about. It's the same thing, by the way, with these packages they're offering to these high-end scientists. If you're going to spend all that money on the hardware, you need really good software, and the software in this case is the actual AI researchers that go into that.

The human—

—researchers that go into that.

Yeah.

I think a meta-observation—no pun intended, accidental pun—is that what is interesting is how crazy these numbers are.

Totally. Totally. I always think about that when people start wringing their hands about what Apple is spending on Apple TV+ programming.

Right.

It's a drop in the bucket. Meta—I just looked this up—had Q1 revenue of $42 billion in Q1 2025. They have the money to spend, and they're spending it in all sorts of other ways on AI. So, yeah, I think that's good context for the numbers, which do still look pretty astounding. It's not clear to me how additive Scale AI is to—

Well, that's the part—

—all the other investments—

That's the part why this is weird, okay? Apple buying Beats for $3 billion: They get Beats headphones. We get it. We kind of get the angle. They get what became Apple Music from the software side. They get a nice hardware business that they can sell in Apple stores. It all sort of makes sense.

Nothing about Meta getting 49% of Scale AI makes sense. Scale AI started out as a data labeler. They generate data for LLMs. They were a big early customer for OpenAI. OpenAI is doing more and more of this internally, probably in part because of something like this. Meta is a customer, Google is a customer, and you need data.

You need data. With the reasoning models, a lot of the initial stuff was just people sitting down and doing math problems, documenting all the steps, and doing that at scale. Scale AI likes to talk about those parts because they're hiring. “Oh, yeah, we hired a Ph.D. candidate in Pennsylvania,” and they make some money on the side. They're also running sweatshops in the Third World to label stuff.

Right.

It's a necessary business. It's also a business that just died. If you're OpenAI or Google, are you going to continue using Scale AI when what's to stop Scale AI from giving the data to their 49% parent, and suddenly you're helping fund your competitor?

Mm-hmm.

This can't be an acquisition for Scale AI and its capabilities, because the nature of the acquisition kills the value of what you're acquiring. That can't be what's going on. So what is going on?

There has been—I talked about this a little bit when I interviewed Mark Zuckerberg. This was kind of a weird interview. I was annoyed that he also did an interview with Dwarkesh Patel. Dwarkesh is a great guy and a great interviewer, but usually Mark Zuckerberg will do one interview in a particular period. I'm like, “Oh, he's doing a fleet of them.” That's kind of annoying.

When I talked to him, he seemed kind of unfocused. It was just—we talked about throwing spaghetti against the wall, right? Just talking about all this sort of stuff that AI can do.

Right. There's sort of a spray-and-pray approach to Meta AI development right now.

Well, particularly in that interview, whereas it had felt almost more cogent to me previously. Then I heard through the grapevine that, yeah, we knew a lot of the Llama people had left. Llama 4 was a mess, not very impressive. They fudged some numbers on a test to make it look better than it was. I heard that Mark was looking for someone to reset their AI initiative and head the team.

Mm-hmm.

I guess that's going to be Alexandr Wang. He's not an AI researcher. Those are the guys they're trying to hire from DeepMind and OpenAI, and whoever else it might be, and giving them all this money. But the head—Sam Altman isn't an AI researcher, but he's the person leading the charge and actually figuring this all out.

It appears that's how it works, and this is one of the world's most expensive acqui-hires, where they're kind of buying him out. He has his company, it's successful, and now he's going to be an executive at Meta. It's not all just throwing him money. Now Scale AI is going to do work for Meta that Meta would have paid for at some point.

Mm-hmm.

So it's not like they're just flushing all that money away. They were going to spend some amount of it anyway, sort of like an Azure–OpenAI relationship. But yeah, it's a pretty strong sign—not necessarily of desperation, but of a real reset: We have to pull out all the stops because we're not good enough. That's my takeaway from it.

Yeah. We got this email from Mario. He says:

“Ben and Andrew, regarding Ben's note on Meta and AI, I would posit that generative AI is potentially more disruptive to Meta than Ben suggested in his Wednesday piece.”

Mario links to a tweet noting that ChatGPT's mobile app usage is now up to 20 minutes per day per user. He continues:

“While you allude to competition for attention with OpenAI in the last paragraph of Wednesday's Daily Update, I think the subject demands far more attention than has been given it in Stratechery. So far, Meta's presentation of AI through its consumer services has been very forced. I've spoken to numerous PMs at the company who feel that this approach has been the wrong approach, but the product decision was made at the very top. Given the relative lack of utility of Meta AI compared to ChatGPT, Gemini, and Claude, I think the proper lens through which to view Meta's recent AI decisions is that they stem more from fear than opportunity. That's not to say there are not great opportunities for Meta. I, too, feel that they should benefit manyfold from this new way of interacting with computing. At this moment, however, they risk harming a crucial vector of top-line growth for their core business should they not get this right.”

That email was interesting to me in 2 respects. First, the OpenAI number is pretty impressive and will inevitably get bigger as we go. Independent of any Meta commentary, I appreciate Mario highlighting that stat, and it occurs to me that they probably are a long-term threat. If the market is attention, Meta is probably right to view OpenAI as a competitive threat. What do you think about that read on what's happening here?

Well, number 1, it is a good email. If I state in a sentence that they're competing for attention with OpenAI, I count that as established on Stratechery. I trust my readers not to need more than that. I think it's pretty clear. So, Mario, come on. You have to give me a little more credit.

You don't need to do the LLM thing.

No, no.

Yeah, exactly. An LLM blows that into—

I don't need to pad it.

—3 paragraphs.

That's right. I gave you the sentence.

Yeah.

No, actually, I was thinking as I wrote that sentence this morning that this is something I should have been clearer about previously. So I actually think Mario's overall point—

Mario, you should've sent the email a day earlier, then you could've really—

Hmm.

—called me on it. Unfortunately, you waited for me to have already said it.

You could've quoted the email on Stratechery.

That's right.

Who knows what would've been possible—

Too bad.

What might've been.

Mario—

That's right.

—24 hours earlier.

No, but I do think that is true. That is a threat. In general, I feel like I'm more optimistic and see more potential for AI on Meta than Meta seems to. What Mario said here—that they're operating from fear—feels exactly right. When you're fearful, you're in a defensive crouch. You're not looking at the opportunities.

One other thing: The whole Mark Zuckerberg interview did kind of annoy me, because I thought I went there in person. They're like, “Oh, we should do it in person.” You think, “This is going to be the big interview about this event.” He does a bunch of other ones, and I'm like, “Well, that was a waste of time.”

But then they wanted to wait until after earnings to publish it, which was Thursday for me, and that was good. Then people come back and say, “Oh, yeah, he just repeated what he said on earnings.” Actually, what he said on earnings was different, because it was post-interview. His articulation of Meta's AI opportunity, I thought, was poor and limited. I'm like—

Hmm.

“What about this? What about that?” Particularly the importance of AI to the metaverse and all those sorts of things. He didn't even bring that up, and it was there in the earnings call, which was published before the interview. But he changed his spiel because I'm—

Yeah.

—the one correcting him about his insufficient listing of the potential benefits of AI.

Pushing him.

And this is why I'm giving Mario a hard time: This is a great point, this fear coming from a position of fear. I think he's right. They don't—

Mm-hmm.

Again, I feel nervous because I wrote about how AI should make Meta the most valuable company, and they're not operating like an entity that is going to achieve that. Worse, they don't seem to be fully aware of what they can do.

This concept that I put forward—actually, let's get to this next question, because this next question leads directly into it. It's from Joseph.

I was going to say, it's all related here. Joseph says:

“Ben and Andrew, undiscussed in Meta's AI struggles is the role of AI head Yann LeCun. He has been a vocal skeptic that LLMs are not going to be real intelligence. While Facebook AI Research has done some incredible work, from a recruiting perspective, I wonder if Yann's outlook has held the company back. It reminds me of The Onion's ‘Decision '80’ joke comparing Jimmy Carter's ‘Let's talk better mileage’ messaging versus Ronald Reagan saying, ‘Kill the bastards.’ What researcher wants to work for an organization whose leadership is scoffing at frontier LLM work when Sam Altman is pitching, ‘Get in, loser. We're building the God machine’? I'm not sure Alexander Wang solves this problem either. Scale AI provides a vital service for AI, but it's not the source of hard research and innovation. So is there a harder problem here that money can't solve?”

What do you think, Ben?

100%. Facebook isn't going anywhere until they fire Yann LeCun. I completely believe that. He's a legend, to be clear, and I also kind of agree with him. I'm a skeptic about LLMs achieving AGI.

Human-brain thinking.

Human-brain thinking, yeah.

Yeah.

I actually agree with his points. The problem is—and again, he has won the Turing Award. He helped invent neural networks. He's a brilliant guy. The problem is, number 1, to Joseph's point, why would you want to go work with him if you're excited about this space? Their most famous person is pooh-poohing the entire thing, number 1.

Number 2, it doesn't matter for this entire vision I paint for Facebook. They don't need AGI. They need LLMs as they exist today to achieve all these tremendous business outcomes. The most pie-in-the-sky outcome that I've talked about for Facebook is that every single pixel on a Facebook property could be advertised against.

Yeah.

You should be able to recognize, “Wow, that green shirt Andrew is wearing is pretty cool.” AI should be able to recognize that. The makers of that green shirt should be able to have a bid in there: If my item shows up, I want to bid on it and make it clickable, and they can make money from it. This is doable with what exists today.

We're not trying to build superintelligence, is your point there?

Exactly.

For the record, this is part of what's confusing about the messaging around this Scale AI acqui-hire situation. Why is there a superintelligence unit being developed at Meta here? Are you doing that solely to recruit talent from DeepMind and some of these other labs? I don't understand.

Again, Scale AI isn't doing research. They're doing data stuff. I mean, I guess he—

Yeah.

—they need a leader, and I guess he thinks that Alexander can be that leader. But yeah, in part it's about getting Yann LeCun to work for Facebook. That was a huge coup, and probably it's like, “Well, if we get rid of him, that's going to make it hard to recruit also.”

But when I say that they need to fire him, that has nothing to do with his takes. It has to do with the fact that he's the wrong person with the wrong takes for Meta and what their opportunities are. They have huge opportunities, to Mario's point, but they're in a defensive crouch. They don't seem to even see them, and then they're layered with just this cloud of doom and gloom that is immaterial to their business.

The reason I'm optimistic about Meta AI is not because I think Meta is going to be first to artificial general intelligence. I think it's because LLMs slot into what they already do in a very compelling way.

Right. Potentially additive in a number of different areas.

Right. This is the whole thing about sustaining innovation versus disruptive innovation. Adding AI to all their stuff—yes, to the extent that—and by the way, OpenAI stealing time is not disrupting Facebook.

That’s just—

Okay.

Them beating Facebook, okay?

Disruption is structural changes to the market and the way people are competing.

Structural changes that hurt the bottom line, and your bottom line—

Mm-hmm.

This is just—

So this would just be classic losing in Meta’s case—

Yes, exactly.

If people are in the ChatGPT app all day.

And it’s a shame because I think LLMs are additive to every aspect of their business. Now, again, if someone else can use it better, that’s just getting beat. And so—

Mm-hmm.

They’re getting beat. And that’s a problem. So it’s a good thing that Mark Zuckerberg is throwing money around and making changes. People should be cheering this. This isn’t a metaverse sort of thing. Maybe if he’s really serious, it’s not just getting rid of Yann LeCun; it’s getting rid of the metaverse.

Yeah.

There’s some money to spend.

It takes some checks. Yep.

He should be spending this money. And the expectation, to use the Clay Christensen framework, is that they do spend this money, because that’s his point about sustaining innovation: the incumbents recognize why it’s important. They do spend to get it done. And so we’re—

Mm-hmm.

Seeing them doing that. Now the question is, is talent so scarce? Is the dream of AGI so large that going to work for a mere company reaching billions of people is not sufficient? Maybe that’s the case, but yeah.

Well, maybe that’s why you need to pay a 9-figure premium to get some of the best researchers in the world in the door at Meta. I mean, that statistic alone is pretty amazing to me, but when you look at what these companies are worth and what they’re spending on all the other AI development every year, then perhaps it’s not that crazy to pay somebody $100 million to come work for Meta.

Stuff’s definitely much more of a mess than it seems. That’s my biggest takeaway. You heard stuff, and stuff seemed a little odd. This is so bizarre from any normal business analysis that the only takeaway can be: stuff is a mess. It’s a much—

Well, clearly.

Bigger mess than it seems.

Yeah. I mean, there have been a couple of years where the investment has been substantial, and to get a couple of years in and have to reset this fundamental part of what Meta has tried to do, that sort of speaks volumes about how lost they are right now. And perhaps Alex Wang can help right the ship going forward. But I think the line in Mario’s email, “They risk harming a crucial vector of top-line growth for their core business should they not get this right,” does seem clear that Zuckerberg is determined to get it right. Just what that looks like and how it’s executed is a complete mystery right now, and it seems like they’re sort of fumbling in the wilderness as they try to see their way through here.

Yep. It does seem so. And I think, as long as I’m taking L’s this episode, that’s the theme. Perhaps I should have seen that coming to a greater degree.

Mm-hmm.

All the reasons to be optimistic were structural. The early Llama models were really good, and I probably should have talked more about all the talent that they lost. Mistral is basically ex-Llama engineers by and large.

Yeah.

And, yeah, that’s a big problem.

Well, great in theory, an absolute catastrophe in practice. That was the theme of Apple Intelligence and WWDC last year, and perhaps that will be the theme of Meta and its AI adventures this year. Who can say?

Rough.

Yeah. What the next 12 months holds?

Not great for me. Not great for me. We’ll see how it goes.

Mm-hmm.

I did want to reemphasize that Yann LeCun is a real pioneer in the field. I’m not insulting him personally, and I actually kind of agree with his take.

Mm-hmm.

Just to double down on the point for Meta specifically, I think Joseph’s exactly right. It’s just a big issue.

Yeah, because they need to be pushing AI into products and doing things beyond AGI and superintelligence work right now.

That’s right.

All of that made sense to me. But on that—and honestly, all the points you’ve made—

I’m—

Have made sense to me.

I’m just paranoid that’s going to get clipped now, and me repeating myself to add the same context I already added 3 times isn’t going to stop it from being clipped. It’s just another L for me right there.

Yeah. Well, I took your comment in good faith. I understood what you mean. And again, you were very coherent throughout this podcast, so great work by you after a wild 72 hours. I can’t even imagine how jet-lagged you are right now, but—

Oh, no, I’m fine. Just by virtue of not sleeping, you get over jet lag quicker than you would think.

Your system just shuts down—

That’s right.

I look forward to the interview with Patrick McGee.

Yes.

Go buy Apple in China. I’m actually dead serious. It’s great. It’s really good.

I know. Honestly, I feel like an absolute moron for not reading it over the last three weeks, so I will rectify that this weekend. And we will come back next week. Go get yourself some Chipotle. Celebrate being in the greatest country in the world, and I will talk to you on the other side.

Talk to you later.