Starting today, end-to-end encrypted RCS messaging begins rolling out in beta for iPhone users running iOS 26.5 with supported carriers and Android users on the latest version of Google Messages. When RCS messages are end-to-end encrypted, they can’t be read while they’re sent between devices. Users will know that a conversation is end-to-end encrypted when they see a new lock icon in their RCS chats. Encryption is on by default and will be automatically enabled over time for new and existing RCS conversations.

I hope this leads to a future where all RCS messages are end-to-end encrypted, but I doubt it. Currently this E2EE RCS depends both on the carriers (of both parties) in a direct chat, and the software running on their devices. The carrier list is pretty broad, but as far as I can tell, it still doesn’t include Google’s own Google Fi.

But the indication for this is subtle. You have to read the small print metadata in each chat to see if it’s encrypted. The message text remains the same shade of green. If it’s a group chat and even one single member isn’t on a phone and carrier that supports E2EE RCS, the entire chat will not be encrypted.

With iMessage, all chats are always E2EE, and always have been. iMessage has been exclusively E2EE since it was created. With RCS you have to look in the metadata small print to check. That’s better than not supporting encryption at all, but my recommendation is to assume all RCS chats are not encrypted unless you double-check every time.

Other than bug fixes, encrypted RCS is the biggest new feature in iOS 26.5.

1. iPhone 17
2. iPhone 17 Pro Max
3. iPhone 17 Pro

And Apple’s phone in spot #6 was not the iPhone Air, alas, but the year-old iPhone 16.

(Via Horace Dediu.)

Though the Mac software introduced the same Liquid Glass interface seen in iOS 26, the design language hasn’t translated as smoothly to the larger displays and different input methods of desktops and laptops. Part of the reason is that Liquid Glass was created with more modern hardware in mind: the crisp OLED displays that are used on iPhones, some iPads and Apple Watches. The software also will be well-suited to the more glass-centric iPhone 20 coming in 2027.

Most Macs, in contrast, still rely on industrial designs introduced several years ago. The current look of the MacBook Air debuted in 2022, while the latest MacBook Pro and iMac designs date back to 2021. Macs also continue to use LCD displays, which don’t render translucency, shadows and glass effects as effectively as OLED screens.

If you’ve used Tahoe, you’re likely familiar with some of the quirks — particularly the transparency effects and shadows that can make lists and other text-heavy areas harder to read.

Trying to argue that the differences between LCD and OLED displays have anything to do with MacOS 26 Tahoe’s UI problems is like arguing that the reason your undercooked poorly prepared food tastes like shit is that it was designed to be served on higher-quality dinnerware. A nicer display is just a nicer display. A bad UI is a bad UI. Shitty undercooked poorly prepared food is shitty undercooked poorly prepared food.

You can actually see MacOS 26 Tahoe on an OLED display using Sidecar with a recent iPad Pro. It doesn’t help. You can also see iOS 26 and iPadOS 26 on devices that don’t have OLED displays. It looks fine. The notion that anything on MacOS 26 Tahoe was optimized for OLED displays makes no sense — there are no MacBooks or Apple desktop displays that use OLED. OLED MacBooks are purportedly coming at the end of this year or next year, but by the time that happens we’ll be mid-cycle for MacOS 27. Lastly, Apple just came out with the new $3,300 Studio Display XDR, using Mini-LED not OLED technology, in March. Even the future of Mac display technology is only partially OLED.

Last year’s Liquid Glass UI redesigns for iOS and iPadOS 26 were pretty good. The Liquid Glass redesign for MacOS 26 was pretty bad. That’s it. It has nothing to do with display technologies.

I’m happy to see Gurman report that the upcoming MacOS 27 release sports a revised UI, but you don’t need a weatherman to know which way the wind blows. Apple revises their UIs almost every year. Given the obvious problems with Tahoe and the pervasive criticisms from UI critics this year, it’d be absolutely flabbergasting if MacOS 27 did not reflect some noticeable changes.

Elsewhere in this week’s column, Gurman writes:

If a new Vision Pro-like device does end up coming together, I wouldn’t expect it for around two more years at least given the hardware resources are so concentrated elsewhere.

I suggest taking Vision headset product timelines from Gurman with a few grains of salt. In mid-October 2025 Apple announced and began shipping the second-gen Vision Pro, with a speed bump from the M2 to M5 chip. But in January 2025, Gurman wrote:

One thing missing from this 2025 road map is the Vision Pro. As of now, I don’t believe there will be a new headset from Apple shipping this year, though there theoretically could be an unveiling ahead of a release later. Signs point to a second-generation model coming in 2026 with an M5 chip.

Worse, in April last year, Gurman not only whiffed again on the second-gen model with M5 being released later in the year, he actually suggested that the M5 speed bump revision was cancelled:

So the company is pushing forward and is currently working on two new models, I’m told. Though Apple had previously considered doing a more basic refresh of the current hardware (changing the chip from the M2 to upcoming M5), it’s now looking to go further.

That exact “previously considered” product shipped just six months after Gurman wrote that. Signs point to Gurman having terrible sources — or just making shit up — regarding Apple’s Vision Pro hardware roadmap.

If you’re building B2B SaaS, especially AI, you quickly need enterprise features like SSO, SCIM, and audit logs. Your developers shouldn’t waste cycles rebuilding that infrastructure. Free them to focus on what sets you apart. WorkOS gives you production-ready APIs for auth and access control that integrate directly into your product. Trusted by over 2,000 companies, including OpenAI, Anthropic, Cursor, and Vercel.

Build faster with WorkOS.

Meta is installing new tracking software on U.S.-based employees’ computers to capture mouse movements, clicks and ​keystrokes for use in training its artificial intelligence models, part of a broad initiative to build AI agents that can perform work tasks autonomously, the company told staffers in internal memos seen by Reuters.

The tool, called Model Capability Initiative (MCI), will run on work-related apps and websites and will also take occasional snapshots of the content on employees’ screens, according to one of the memos, posted by a staff AI research scientist on Tuesday in a channel for the company’s model-building Meta SuperIntelligence Labs team.

I love this. Anyone who works at Meta knows who they’re working for. I hope they’re all as creeped out by this as I would be. What I’d really like to know is how far up the chain does this go? If they have any honor, every single employee at the company, right up to Zuckerberg, would get this MCI spyware. What’s good for the goose is good for the gander. I presume that’s not the case, and there’s a two-class system where if you’re high enough in the org chart, you don’t get it. I would love to hear from any little birdies about this, but I don’t have many little birdies in Menlo Park. (Alan, baby, help me out?)

Also:

Meta is planning to lay off 10% of its workforce globally starting on May 20 and is eyeing additional large cuts later this year.

If I worked there I’d raise my hand to get a buyout, but, I’d never work there in the first place.

Horwitz, by the way, won a 2026 Pulitzer for his investigative coverage of Meta.

Prolost Watches is an iPhone app for managing your watch collection. It’s part database, part journal; designed for the detail-obsessed mind of the watch fanatic. As you log each day’s choice of watch, insights are revealed. Wear logs trace a path on the map. Events from the past are resurfaced at opportune times. Finances mange themselves as you buy and sell. Your entire collection lives in your pocket, and you get to enjoy all your watches, even the ones you’re not wearing. [...]

Prolost Watches is a one-time purchase. There’s no subscription, no ads, no account, and no server. Your data is secure and private, and never leaves your device. Pre-order now for US$14.99, with expected release on June 16 at US$19.99.

I’m friends with Stu, I have my own little obsessively curated watch collection, and, for some of my interests in life, I love keeping a log of activity. So I jumped at the chance to beta test Prolost Watches. And it turns out, my watches are not one of those things I want to track in a log or database. I just want to continue doing what I’ve been doing ever since I went from owning only one watch to two: I pick the one I’m in the mood for that morning and I wear it for the day.

I feel the same way about sleep tracking. I’m fortunate in that I sleep great every night. I’ve been sleeping better this past year than I have in my entire life. So while I have my Apple Watch(es) set to track my sleep overnight — because why not collect the data? — I don’t look at it or worry about it most days because all it tends to do is add a little stress to my mind over something I ought not have even an iota of stress regarding. (I like wearing an Apple Watch while I sleep not for the sleep tracking but because it’s easy to read in the middle of the night in the dark.)

But, that’s me. I obsessively track other things that most of you would think are a bit nutty to track. You don’t get to pick your obsessions, but you know what they are. If you’ve got a few watches and you think you’d be interested in tracking how often you wear each one, you should already have the above link to Prolost Watches open in a tab.

Interesting too, is how Maschwitz made Prolost Watches:

Bitrig has changed a lot since I used the iPhone version to create the ill-fated version of Drinking Buddy. It’s now a native Mac app that allows prompt-base creation of native SwiftUI apps for iPhone, as well as iPad, Mac, and Apple Watch. It has a built-in simulator, and can preview your apps on your device as well. If Lovable (which I used to create the shipping version of Drinking Buddy) is at one end of the spectrum — easy for anyone to use with little experience, and Claude Code running in the terminal is at the other, Bitrig is in a sweet spot right in the middle: a little nerdy, but with some well-considered creature comforts that, in my case, made it mostly a delight to craft and refine a complex app.

The iPhone version of Bitrig got swept up in Apple’s infuriating but unsurprising crackdown on iOS vibe-coding apps a few months ago. It’s still in the App Store, but hasn’t been updated in over five months. The Mac app is in active development. It’s kind of bananas that the iPhone is a nearly 20-year-old platform and you still can’t use an iPhone app to make iPhone apps. And when developers, like Bitrig, found ways to build atop LLM capabilities to make iPhone apps that can make iPhone apps, Apple put the kibosh on it.

(I think it’s a total coincidence that the Febreze ad seems roughly color-matched to the sky. But scroll down in the Bluesky thread for some links to the absolute genius campaign from Cerveza Cristal beer, with spots specifically designed to integrate into Star Wars when it was broadcast on commercial TV in Chile. “Your father wanted you to have this when you were old enough” might be the funniest TV commercial ever made.)

Update: Here’s Todd Vaziri’s 2026 HD remaster of the original jump cut, for your comparison. Enjoy.

Those around the league quickly honored that legacy during Monday night’s slate of games. Tributes rolled in from across the league, with various play-by-play announcers deviating from their typical routines to give a nod to Sterling’s distinct style.

It started with the Yankees and TV man Michael Kay, who called Aaron Judge’s first-inning home run exactly as Sterling would have: “It is high, it is far, it is gone!” Kay said, continuing: “Aaron Judge, a Judgian blast! Here comes the Judge!”

The Yankees also tipped the cap to Sterling by playing a recording of his iconic post-win call over the loudspeakers in Yankee Stadium once New York wrapped up a 12-1 win over the Orioles — “Yankees win, theeee Yankees win!” — something Judge and manager Aaron Boone each said they hoped could continue to be done moving forward.

The Yankees will wear a commemorative patch for the remainder of the season. I’ve got my beefs with Hal Steinbrenner, but the organization still knows how to do stuff like this right.

Sterling called 5,426 regular-season Yankees games and 225 postseason games. According to this tally, there are only three teams that have even played in at least 225 postseason games in franchise history. He called 5,060 consecutive games from September 1989 to July 2019 — a span that included every at-bat of Derek Jeter’s career and every inning of Mariano Rivera’s. He called five seasons for the Atlanta Braves before getting a real job.

To put that longevity in perspective, how’s this for a factoid:

John Sterling called nearly 3.0% of all games in MLB history — this includes all games, for all teams, even those prior to the first ever radio broadcast of a ballgame on Aug. 5, 1921.

(Vin Scully, the best there ever was, called over 4 percent of all MLB games ever played.)

I listened to Sterling call a lot of games. He never once made it boring.

Every AI-generated application creates the same problems: Where does it run, and how is it deployed, secured, and managed?

The app needs a database. It needs user authentication, role-based permissions, and audit logging. It needs backup and recovery. It needs to work on Mac, Windows, iPad, iPhone, and the web. It needs to run on infrastructure your organization controls, whether that is in the cloud or on your own hardware.

This is what FileMaker already is. One unified platform with data, logic, interface, security, and cross-platform delivery, built together from the start. We’ve been solving this problem at scale for over 40 years.

“Mac, Windows, iPad, iPhone, and the web”, huh? Feels like there’s a somewhat popular platform missing from that list. Can’t quite put my finger on it. Oh yeah.

FileMaker 2026 is coming soon. This release focuses on resiliency, productivity, and infrastructure, including native disaster recovery and business continuity capabilities enabled by two new features: FileMaker Server Remote Backup and Standby Server. Additionally, it lays the groundwork for agentic development. We will share specifics in the coming weeks.

Later this summer, following the release of FileMaker 2026, we will deliver the first developer previews of our agentic coding functionality.

I have to admit I very seldom hear about FileMaker, and I’ve never once heard of Ryan McCann before. But it must still be a significant business. Tim Cook certainly doesn’t seem like the sentimental type.

Apple today announced that Chief Financial Officer Luca Maestri will transition from his role on January 1, 2025. Maestri will continue to lead the Corporate Services teams, including information systems and technology, information security, and real estate and development, reporting to Apple CEO Tim Cook. As part of a planned succession, Kevan Parekh, Apple’s Vice President of Financial Planning and Analysis, will become Chief Financial Officer and join the executive team.

That continued leadership role wasn’t ceremonial. Maestri still has an executive page, which reads:

Luca Maestri is Apple’s vice president of Corporate Services, reporting to CEO Tim Cook. In this role, he oversees a range of functions, including information systems and technology, information security, real estate and development, Caffè Macs, and Claris.

I find the mention of Caffè Macs to be utterly charming. (And the mention of Claris to be at least a little interesting.)

Apple has removed more desktop Macs from its online store as the global memory shortage continues. Mac mini models with 32GB and 64GB of RAM are no longer available for purchase, nor is the M3 Ultra Mac Studio with 256GB RAM.

The M3 Ultra Mac Studio is now available only in a 96GB RAM configuration, with higher-tier options eliminated. Both M3 Mac Studio and M4 Max Mac Studio models have delivery estimates of 9 to 10 weeks.

Last March, Apple was hit with a class action lawsuit after delaying the launch of the “more personalized Siri” that was first announced at WWDC 2024. Apple agreed to settle the case in December, and the full settlement terms are now available. Apple is set to pay $250 million to settle the lawsuit, equating to an estimated $25 per device. That number could reach up to $95 per device, depending on how many users submit claims. [...]

As part of the settlement, Apple is not admitting any wrongdoing. The company continues to assert that “it acted in good faith and in a manner reasonably believed to be in accordance with all applicable rules, regulations, and laws.” In a statement to 9to5Mac, an Apple spokesperson said:

Since the launch of Apple Intelligence, we have introduced dozens of features across many languages that are integrated across Apple’s platforms, relevant to what users do every day, and built with privacy protections at every step. These include Visual Intelligence, Live Translation, Writing Tools, Genmoji, Clean Up and many more.

Apple has reached a settlement to resolve claims related to the availability of two additional features. We resolved this matter to stay focused on doing what we do best, delivering the most innovative products and services to our users.

A $25/device settlement sounds about right. Apple ran ads showing features that still haven’t shipped. That they honestly intended to somehow ship those features, as promised, doesn’t mean the ads didn’t wind up being false.

The Pentagon said President Trump’s Iran war has cost the United States at least $25 billion, driven primarily by the military’s use of munitions, the Financial Times reports.

The New York Times had an interesting piece trying to put that number in context (gift link):

$25 billion is similar to:

• The annual budget of NASA.
• Spending on military aid to Israel after Oct. 7.
• Spending by U.S.A.I.D. before it was disbanded.
• The cost to expand Obamacare subsidies for one year.

These are all comparisons to other aspects of the U.S. federal budget. It’s interesting also to use this in comparison with the current moment in tech:

• OpenAI’s latest valuation of $852 billion (I love the 2) equals 34 Iran Wars.
• Anthropic’s $380 billion valuation equals 15 Iran Wars.
• Apple’s current four-year U.S. manufacturing commitment of “more than $500 billion” equals 20 Iran Wars.
• Google expects to spend between $91–93 billion in capital expenditures this calendar year, mostly related to AI infrastructure. That’s a little over 3 Iran Wars this year.
• Larry Ellison currently has a net worth of $220 billion. That’s just short of 9 Iran Wars. But since the start of the war on February 28, his net worth has grown $46 billion. That’s about 2 Iran Wars during the time of the actual Iran War thus far.

My muse for the session was this quote from Walt Disney: “We don’t make movies to make money; we make money to make more movies.” To me, that’s it. That’s the thing.

Merlin and I were talking about independent writers and podcasters, because that’s what we were (and remain), but the concept applies just as perfectly to independent developers. This came to my mind after reading (and linking to) David Smith’s description of the new Pedometer++ today. Not just what it does, but why he spent six years making it. That’s the sort of productive obsession that fascinates me.

Ice water is always refreshing, but it tastes better when you’re on a road trip to hell. It feels like the world of software is bifurcating quality-wise. This whole thing about Adobe’s new craptacular “modern” UI language (a.k.a. “Spectrum”) exemplifies one side of that bifurcation — the bad-and-getting-worse side. Software that is the product not just of an ignorance of long-established principles of interaction design, but of a willful disdain for those principles. What Adobe is now shipping is just inexplicably bad UI, ignoring literally decades of great work and long-mastered concepts — a lot of which work was pioneered by Adobe itself!

The whole thing with MacOS 26 Tahoe is similar. To be clear, the UI crimes in Tahoe are deeply worrisome, but they are nowhere near as severe as those in Adobe’s Spectrum. But the problems with Tahoe are steps down the same fork in the road that Adobe took years ago. Spectrum is where Tahoe suggests that MacOS was headed under Alan Dye’s leadership: cross-platform sameness for the sake of sameness, with a complete disregard for longstanding platform nuances and idioms. In Spectrum’s case those platforms are MacOS and Windows and the web. In Tahoe’s case it’s MacOS and iOS.¹

The other side of the software fork is not deserted. It’s just populated, more than ever, by the products of small independent developers who obsess, first and foremost, over quality and artistic vision. Remarkable new software gems exhibiting spectacular UI design appear all the time. They’re just not coming from the biggest companies, the ones whose apps, alas, dominate not just our desktops and pockets but our entire culture today.²

There’s always been software with poorly designed user interfaces. Much of it has been successful financially, sometimes spectacularly so. I’d argue, in all seriousness, that that’s the story of Microsoft in a nutshell. What’s new today is poorly designed software from developers from whom we expect better. In the old days there were people who would argue that prioritizing good user interface design was a waste of time — like spending hours decorating cupcakes destined for kindergarteners who are simply going to mash them into their mouths. (Again: cf. Microsoft’s undeniable market success.) What’s new today is people holding up objectively bad interaction design and proclaiming it to be good, and the product of teams that purportedly prioritize “design”, when it’s clear they have no idea what they’re talking about. It’s one thing to make something poorly designed and shrug on the grounds that it doesn’t matter. It’s another thing to make something poorly designed and hold it up as good design.

We are justified to expect nothing short of insane greatness from Apple, and solidly good design from Adobe. In principle, all software ought to have well-designed user interfaces. That’s never going to be the case. But software for designers — Adobe’s raison d’être — absolutely demands to be well-designed itself, like how a book on writing must itself be well-written.

Perhaps I was wrong, though, to describe Adobe’s new UI as inexplicable. It’s just indefensible. The explanation for so much software going so rotten from a UI-design perspective is, the more I think about it, related to Nilay Patel’s “Software Brain” theory, which I’ve commented on both directly and indirectly. Here’s Patel’s definition of “software brain”:

The simplest definition I’ve come up with is that it’s when you see the whole world as a series of databases that can be controlled with the structured language of software code. Like I said, this is a powerful way of seeing things. So much of our lives run through databases, and a bunch of important companies have been built around maintaining those databases and providing access to them.

Zillow is a database of houses. Uber is a database of cars and riders. YouTube is a database of videos. The Verge’s website is a database of stories. You can go on and on and on. Once you start seeing the world as a bunch of databases, it’s a small jump to feeling like you can control everything if you can just control the data.

But that doesn’t always work.

You might think it counterintuitive that a movement obsessed with software would be spearheading a severe decline in the design quality of software, but in Patel’s definition, there’s no concept of software as art, as a practice, as a craft. Software brain is purely an obsession with software as a medium in and of itself. A means with no consideration for the end.

Framed in Walt Disney’s adage, software brain makes software only to make more money. The idea of making money in order to make more software — to afford the time and talent to craft it — does not compute. Framed in the metaphor that Steve Jobs used to close his introduction of the original iPad, and returned to again to close his final keynote at WWDC 2011, software brain is nowhere near the intersection of technology and the liberal arts. Software brain is so far down Technology Street that it’s no longer in the same zip code as Liberal Arts Avenue. Another way, perhaps, to define software brain is that it’s the utter rejection of Jobs’s maxim that “technology is not enough”. With software brain, technology is all there is.

I love going on wilderness adventures. I am rarely happier than when I am far off into the mountains without a soul in sight. As a result, I have spent a lot of time learning how to safely explore and navigate when I’m away from civilization. The most important habit I’ve found for not getting lost is to be very regular in checking your location as you go, and the best way I’ve found to do that is to have a map on my wrist.

For more than six years I’ve been working towards creating the best possible mapping experience on the Apple Watch. With yesterday’s launch of Pedometer++ 8, I feel like this design journey has reached a meaningful destination. I would contend that Pedometer++’s watchOS mapping support is the absolute best available on the App Store.

So I wanted to walk through the journey it took to get here.

“I love going on wilderness adventures” is how you start a post about an app update. Or at least my type of update to my type of app. I don’t have any desire for maps on my watch, but reading this makes me want it anyway. Enthusiasm is contagious. (See also: the Pedometer++ blog, and Stephen Hackett at 512 Pixels.)

WorkOS gives you production-ready APIs for auth and access control that integrate directly into your product. Trusted by over 2,000 companies, including OpenAI, Anthropic, Cursor, and Vercel.

Build faster with WorkOS →

If you do a little poking around in Adobe’s application bundles, a key reason for the jankiness of these user interfaces becomes apparent: it is because they are little webpages. These dialog boxes are HTML files that reference a chunky CSS file and oodles of JavaScript, and appear to be built with React. [...]

I was going to write about how this stuff should have been tried with people who actually use Adobe’s apps in a high-pressure environment, but I am sure it was and, also, it does not matter. Wichary has it right. These are fundamental principles of user interface design that Adobe is ignoring because its internal tooling has taken precedence.

I will quibble only with this line from Heer’s post:

Also, Adobe’s interface has always been unique and not quite at home on either MacOS or Windows.

You have to go back to the 1990s and classic Mac OS, but Adobe’s best apps used to have exemplary native UIs. Apps like Photoshop helped push the state of the art in Mac UI forward. Tabbed palettes were a revelation. Fire up, say, Photoshop 3.0 on MacOS 7.6 and see what I mean.

Also worth noting is how much this new “modern” UI isn’t just subjectively ugly, it’s objectively breaking the habits and expectations of users with literally decades of experience with Photoshop — users who, like me, remember when Adobe’s UI wasn’t just merely tolerable but actually good. It’s insane when you think about it.

How did Adobe lose that good sense of yore? Two ways. Gradually, then suddenly.

I may or may not write and publish a short e-book about Markdown sometime this year, most likely as part of a monthly focus. But l’ve written small parts of it already, as I do, and I figured it might be interesting for at least some readers. And so here’s an early draft of an introductory chapter that may or may not be called “On writing.” We’ll see.

It’s odd how things turn out in life. Thurrott’s and my careers are almost uniquely parallel, but have seldom intersected. This book would have been a very surprising outcome to me, if you’d told me about it 20 years ago. Sort of a fun outcome, though, and I must admit to being curious what comes of it.

Sam Altman was the president of Y Combinator for several years, and left to become the full-time CEO of OpenAI. The New Yorker quotes Y Combinator co-founder Paul Graham multiple times, in the context of Altman’s trustworthiness. (Some of those quotes are firsthand, others secondhand.) Graham’s role in the story — particularly his public remarks after publication — comprised an entire section in my own take on the New Yorker piece, wherein I concluded:

I would characterize Graham’s tweets re: Altman this week as emphasizing only that Altman was not fired or otherwise forced from YC, and could have stayed as CEO at YC if he’d found another CEO for OpenAI. But for all of Graham’s elucidating engagement on Twitter/X this week regarding this story, he’s dancing around the core question of the Farrow/Marantz investigation, the one right there in The New Yorker’s headline: Can Sam Altman be trusted? “We didn’t ‘remove’ Sam Altman” and “We didn’t want him to leave” are not the same things as saying, say, “I think Sam Altman is honest and trustworthy” or “Sam Altman is a man of integrity”. If Paul Graham were to say such things, clearly and unambiguously, those remarks would carry tremendous weight. But — rather conspicuously to my eyes — he’s not saying such things.

The thing that stuck in my craw is this: Does Y Combinator own a stake in OpenAI? And if they do, given OpenAI’s sky-high valuation, isn’t that stake worth billions of dollars?

OpenAI was seeded by an offshoot of Y Combinator called YC Research in 2016 — when Altman was running YC. In December 2023, the well-known AI expert (and AI-hype skeptic) Gary Marcus wrote the following, in a piece on Altman’s trustworthiness in the wake of the OpenAI board saga that saw Altman fired, re-hired, and the board purged in the course of a tumultuous week:

After poking around, I found out that “I have no equity in OpenAI” was only half the truth; while Altman to my knowledge holds no direct equity in OpenAI, he does have an indirect stake in OpenAI, and that fact should have been disclosed.

In particular, he owns a stake of Y Combinator, and Y Combinator owns a stake in OpenAI. It may well be worth tens of millions of dollars; even for Altman, that’s not trivial. Since he was President of Y Combinator, and CEO of OpenAI; he surely was aware of this.

So it’s well known that Y Combinator owns some stake in OpenAI. But how big is that stake? This seems like devilishly difficult information to obtain. I asked around and a little birdie who knows several OpenAI investors came back with an answer: Y Combinator owns about 0.6 percent of OpenAI. At OpenAI’s current $852 billion valuation, that’s worth over $5 billion.

Graham and his wife Jessica Livingston are two of Y Combinator’s four founding partners. The fact that Paul Graham personally has billions of dollars at stake with OpenAI doesn’t mean that his public opinion on Sam Altman’s trustworthiness and leadership is invalid. But it certainly seems like the sort of thing that ought to be disclosed when quoting Graham as an Altman character reference. A billion dollars here, a billion there — that adds up to the sort of money that might skew a fellow’s opinion.

To win the artificial intelligence race, Google not only has developed its own technologies, but has also pumped money into prominent A.I. start-ups. And to preserve its competitive edge, Google has kept its ownership stakes in those start-ups a secret.

Court documents recently obtained by The New York Times reveal Google’s stake in one of those start-ups, Anthropic, as well as how its investment in the young company is set to change. Google owns 14 percent of Anthropic, according to legal filings that the A.I. start-up submitted as part of a Google antitrust case. But that investment gives Google little control over the company. The internet giant can own only up to 15 percent of Anthropic, according to the filings, and Google holds no voting rights, no board seats and no board observer rights at the start-up.

Still, Google is set to invest an additional $750 million in Anthropic in September through a type of loan known as convertible debt, according to the filings. The companies agreed to the convertible note in 2023. In total, Google has invested more than $3 billion in the A.I. company.

Anthropic’s latest funding round — a rare Series G — valued the company at $380 billion. So let’s say Google has invested $4 billion to date, and Anthropic really is worth $380 billion. Google’s slice of that would be worth a little north of $50 billion, quite the return on investment. And competitively, there’s a heads-they-win (with Gemini), tails-they-don’t-lose (with Claude) aspect. Maybe that’s not the best metaphor, since OpenAI would make it a three-sided coin, but still.

(Via today’s subscriber-only Stratechery update, where Ben Thompson noted this in the context of Google last week reporting a 30% increase in operating profit year-over-year, but an eye-popping 81% increase in overall profit. The difference was the growth in their investments, almost certainly Anthropic in particular.)

iOS App Store search is no longer about relevance. It’s about ad inventory. With Apple’s introduction of a second search ad, for any query where we weren’t #1, we’ve effectively moved down one position. [...] If you’re counting at home, roughly 70% of the interface is covered in ads. A casino ad, to boot.

That was a month ago. Two weeks later, he posted a follow-up, showing the effect on Think Tap Work’s apps in the App Store:

I wanted to share some updated numbers from our own apps. To isolate the impact, these numbers only include App Store Search impressions from iOS devices, comparing Mar 26–Apr 8 to the prior two weeks. In other words: how much visibility we’ve lost in search.

• Morpho Converter: 26% decrease
• Pop Out Timer: 23% decrease
• Attendant for Zoom: 30% decrease
• Participant: 3% increase. As previously mentioned, we advertise Participant using Search Ads. It’s not surprising that it might be getting a bump from this change.

The screenshot in his follow-up shows another casino ad, this time in a search for “Roblox”. Kinda gross.

Here’s Wikipedia on the “Zero-One-Infinity Rule”:

The zero-one-infinity (ZOI) rule is a rule of thumb in software design proposed by early computing pioneer Willem van der Poel. It argues that arbitrary limits on the number of instances of a particular type of data or structure should not be allowed. Instead, an entity should either be forbidden entirely, only one should be allowed, or any number of them should be allowed.

In Apple Notes, you can only have one main window open. In Apple Mail, however, you can open as many Viewer Windows as you want. Both are compliant with the Zero-One-Infinity rule. An app that allowed you to open multiple viewer windows — but no more than some arbitrary limit — would not be. ZOI is a very good rule of thumb.

I feel like a variation of Zero-One-Infinity is a good rule of thumb for ads, too. From the perspective of users — and probably developers — zero was the best number of ads for Apple to show in App Store search results. One was worse but acceptable. But now that they’re showing more than one, they’re on their way to infinity. They’ve started down the slippery slope. Remember when Google only showed one ad in search results?

Anyway, who’s looking forward to ads in Apple Maps this summer?

For decades, Japan’s Oreos weren’t made by Nabisco at all. They were produced domestically by Yamazaki Biscuits, under a licensing arrangement with what eventually became Mondelez International. This was, by most accounts, a reasonable arrangement. The cookies were local. The quality was consistent. Nobody was complaining.

Then Mondelez did what corporations do when things are working fine. The license expired, and Mondelez moved production of the Oreos it sells in Japan to China, exporting them to Japanese wholesalers and retailers. A cost decision. A spreadsheet decision. The kind of decision made in a room with no windows and a very good projector.

Sensitive Japanese consumers noticed quickly — the taste had changed. Into that opening stepped the Noir, inheriting the flavor the old Oreo had left behind.

Yamazaki Biscuits launched Noir in December 2017 as the successor nobody had officially asked for and everybody apparently wanted.

I have a great affinity for Newman-O’s, which I’ve previously described as “the cookies Oreos pretend to be”. Turns out though I’ve mostly sung the praises of Newman-O’s on my podcast and social media, not here on Daring Fireball. I love Newman-O’s, never tire of them, and will fight any man who argues that Oreos taste better. In fact, late last night, when a friend texted me with a link to this story from Adelstein, I was by sheer happenstance eating a few Newman-O’s. True story.

But now I’m fascinated by the existence of these Japanese rivals. A spite Oreo called Noir. They look and sound delicious, but they seem difficult to obtain in the U.S.

I’m angry. (Clearly.) We should all be angry in face of stuff like this. This is how people get fed up with software — because it feels unstable and deteriorates on its own without needing to.

I know I brought up that an existing power user base can be a huge pain in the ass, and I am a decades-old Photoshop power user. But this is different than other examples where the product needs or at least wants to evolve past its core audience or toward a different market. For Photoshop here, nothing I see indicates any change in course or clientele — and yet all of these good moments in UI that used to help me out no longer exist.

Plus, all those transgressions are solved problems. Those issues are not buried in pages of heavily litigated patents, or in seven collective brains of world-class interface designers whose driveways are presently occupied by cash-filled trucks sent over by frontier companies. This isn’t some long lost art that requires archaeologists to decipher. This feels like carelessness and laziness in face of basic UI engineering; in a likely internally-motivated effort to refresh the interface, the team threw an entire nursery worth of babies with the bathwater.

The before-and-after screenshots look like examples from a lecture on user interface design — if you swap them around make the new ones “before” and the old ones “after”. Better balance, better focus behavior, appropriate platform-native typography.

(Shades of Héliographe’s devastating critique of the history of the app icon for Pages: “If you put the Apple icons in reverse it looks like the portfolio of someone getting really really good at icon design.”)

Anthropic expects AI-powered virtual employees to begin roaming corporate networks in the next year, the company’s top security leader told Axios in an interview this week. [...] Virtual employees could be the next AI innovation hotbed, Jason Clinton, the company’s chief information security officer, told Axios.

Agents typically focus on a specific, programmable task. In security, that’s meant having autonomous agents respond to phishing alerts and other threat indicators. Virtual employees would take that automation a step further: These AI identities would have their own “memories,” their own roles in the company and even their own corporate accounts and passwords.

Unlike Anthropic’s ambitious prediction regarding the vertiginous rise in AI code generation, this one, I think we can say, has fallen flat on its face. This isn’t how companies are using AI — or at least they shouldn’t. But contra Axios’s year-ago headline (“Exclusive: Anthropic Warns Fully AI Employees Are a Year Away”), this wasn’t a warning. It was an advertisement — and exactly the sort of wink-wink-nudge-nudge software-brain “warning” that has tanked public sentiment regarding AI. It wasn’t an indication that Anthropic actually believed there would exist “fully AI employees” today, but rather that they wanted to build enthusiasm amongst the sort of ghoulish “let them eat cake” executives who really wish that they could “hire” fully AI employees.

But where I think Amodei’s remarks, quoted above, are facile is that it hasn’t played out as simply that lines of code that would have been written by human programmers are now generated by AI models. That’s part of it, for sure. But what’s revolutionary — a topic I’ve been posting about twice already today — is that AI code generation tools are being used to create services and apps and libraries that simply would not have been written at all before. It may well be that the total number of lines of code that will be written by people today isn’t much different from the number of lines of code that were written by people a year ago. But there might be 10× more code generated by AI than is written by people today. Maybe more. Maybe a lot more? And a year or two or three from now, that might be 100× or 1,000× or 100,000×.

In that near future, human programmers are likely still to be writing — or at least line-by-line reviewing and approving — code. But as a percentage of all code being generated, that will only be a sliver.

Early in April we kind of got a number we can assign to this: 14×. GitHub COO Kyle Daigle posted on Twitter/X (alternative link):

Yup, platform activity is surging. There were 1 billion commits in 2025. Now, it’s 275 million per week, on pace for 14 billion this year if growth remains linear (spoiler: it won’t.)

GitHub Actions has grown from 500M minutes/week in 2023 to 1B minutes/week in 2025, and now 2.1B minutes so far this week.

No one at Apple is talking about it publicly (yet?), but judging by response times, App Store review is facing a similar deluge. And as for GitHub, yeah.

SightMD, a leading ophthalmology practice in the greater New England area, today announced a historic milestone in surgical innovation. Dr. Eric Rosenberg, DO, MSE, has become the first surgeon in the world to successfully perform cataract surgery using the Apple Vision Pro, powered by ScopeXR, a groundbreaking mixed reality surgical platform co-developed by Dr. Rosenberg.

The initial procedure was successfully completed in October 2025, and since that time, Dr. Rosenberg and his team have performed hundreds of additional cases using the platform, demonstrating both its scalability and real-world clinical impact.

Not being ready for mass-market popularity is such a different thing from not being ready for niche practical use cases. Would be a weird thing indeed if Apple “gave up” on this platform.

A colorful personality who engaged and entertained fans with a distinct conversational style, Sterling called 5,426 regular-season Yankees games and 225 more in the postseason from 1989 until his retirement in 2024. After initially stepping away from the microphone in April of that year, Sterling returned to call selected games late in the ’24 season, including each contest of the World Series.

At the time of his initial retirement, Sterling said that he considered himself to be “a very blessed human being,” noting that he had lived out a childhood dream of broadcasting on the radio for more than 64 years.

“It’s your medium. You do what you want,” Sterling once said. “You have to paint the picture, which I love doing.”

That’s baseball, Suzyn.

It’s official! I’m permanently banned from X for tweeting “TLDR: Fascism.” (appeal denied)

“TLDR: Fascism” was Durán’s two-word response to this 1,000-word essay from Palantir describing their vision for a “Technological Republic”. (Alternative link to essay if you don’t want to visit x.com.)

Getting perma-banned from Twitter/X by Elon Musk gives Durán a nice Streisand-effect boost to promote his upcoming new book, The Nerd Reich. If the book is even half as good as its title it should be a bestseller.

There is no point getting any more outraged or disgusted at Meta for firing these contractors than you already were in the first place. They had to fire them. The moment this investigative report was published in late February, the fate of Sama’s Kenyan operation was sealed. They were toast. The key to understanding this is that Meta runs a criminal enterprise. Most of the organized crimes Meta commits aren’t crimes against the legal code (although some are), but rather crimes against public perception and human decency. Remember what they did with Onavo, their VPN product? Was that illegal? Dunno. Was it outrageous? Hell yes.

Let’s just concede for the sake of argument that there’s nothing illegal about the way Meta was sending video footage from users’ AI Glasses to contractors in Kenya to review. I presume they’re still doing it today, just with different contractors, in a different computer cubicle sweatshop, perhaps in a different country. Nothing to cover up legally. But just the plain description of what they’re doing fills people with a visceral repulsion. However, people only have that visceral reaction if they know what’s going on. Part of the whole premise is that the whole thing has to be kept on the q.t.

If it said right on the box that when you use Meta AI Glasses, the footage might be reviewed by third-party contractors, and when that footage is reviewed, you — the user whose footage is being reviewed — won’t know it’s happening and won’t get prompted first for permission (because you’ve actually OK’d it in advance just by hitting the “Accept” button on the long dense terms of service that literally almost no one reads because such terms are written in impenetrable legalese), almost no one would buy them. And if it were more widely known that this is how these glasses work, there’d be more of a social stigma surrounding those who wear them.¹

That, I think, is the primary reason why the contractors were in Kenya in the first place, and their replacements (now that Meta has terminated its contract with Sama) are surely still in some third-world country. It’s not about the lower wages (but that doesn’t hurt). It’s about the fact that the entire existence of the operation is easier to keep quiet when it’s literally on the other side of the planet. It’s a goddamn marvel that the investigative reporters from those two Swedish newspapers found them.

Most illegal acts are scandalous, but many scandalous acts are perfectly legal. But all scandalous acts need to be covered up. The operation has to be kept quiet, has to be covered up, because it’s unacceptable. It’s outrageous. If this were more widely publicized, Meta would suffer on two fronts. First, it would become better known that there’s nothing artificial about some of what they call “AI” — it’s in fact powered by human intelligence, just in another hemisphere. Second, and related to the first, some of the interactions you have with Meta AI — including images and video you send it, and images and video captured by Meta AI Glasses — are reviewed by human contractors. People write things and show things to AI, thinking it’s kept private between them and a computer program, that they would never share if they knew it might be seen by human beings paid by the AI provider to refine the training and correct its mistakes. A lot of people only use these “AI” products because they have no idea what’s actually going on.

“Three may keep a secret, if two of them are dead.”
—Benjamin Franklin, Poor Richard’s Almanack

Anyway, enjoy the Meta AI built into WhatsApp and Instagram. And maybe keep a link to that report on Meta’s contractors in Kenya handy for anyone you meet who wears AI Glasses.

Here’s what I wrote when Apple first made this current commitment in February 2025, just weeks after Trump’s second term started:

Apple announced a similar plan four years ago — $430 billion and 20,000 jobs. In the announcement of that 2021 plan, Apple said, “Over the past three years, Apple’s contributions in the US have significantly outpaced the company’s original five-year goal of $350 billion set in 2018.”

So I don’t think this announcement is bullshit, at all. But I also don’t think what Apple has announced today is much, if any, different from what they’d be doing if Kamala Harris had gotten 1–2 percent more of the vote in a handful of states in November. The difference is that everyone is looking for quid pro quo with President Transactional back in office.

Apple first announced a plan in 2018, during Trump 1.0, to spend $350 billion over the next five years. Then in 2021 — midway through those five years, at the start of the Biden administration — they said spending was above that previously promised pace but they were announcing a new five-year plan to spend $430 billion. That plan would have run through 2026 (this year). But, again, right after Trump was re-inaugurated last year, before the period covered by the 2021 five-year plan was up, they announced the current $500 billion plan. The only difference is that this latest spending commitment is a four-year plan, not a five-year one (probably because they know Trump doesn’t give one shit what they do after he leaves office).

This isn’t a shell game or a scam. I believe Apple really has spent what they’ve said they were going to spend, and really plans to spend what they’ve committed to spend in the coming three years. If anything, as they said in 2021, their actual spending has probably exceeded what they committed to, during each of these periods, and will continue to. It’s very Tim Cook-ian and very Apple-like to underpromise and overdeliver. So I’d say it’s a shoo-in that when Apple announced the current plan to spend “more than $500 billion” in the U.S. from 2026–2030, they actually planned to hit that target with room to spare. So saying that they’ll throw the proceeds from any potential tariff refund check into the same fund doesn’t actually change a damn thing about their plans.

And if the pattern holds, they’ll announce a new four- or five-year plan for $600 billion (give or take) after the 2028 election, regardless who wins. There’s never any sort of accounting where they show that they spent exactly, say, $447 billion between 2021 and 2026, or $389 billion from 2018 to 2023. And there’s never going to be any exact accounting like that for what they’ll spend in this current “more than $500 billion” plan covering 2026 to 2030. There’s also no accounting for how much Apple spent last year on Trump’s invalid tariffs. Presumably, if they eventually get a refund check from the Treasury, we will know the exact number. But given that whatever they spent on Trump’s tariffs had only a negligible effect on their earnings last year, we can presume that the money they’re committed to spending on U.S. manufacturing and job creation from 2026 to 2030 remains about $500 billion, and it’s really all just projects that they would have spent the exact same amount of money on if Kamala Harris were now in the White House — just like how they committed to spending $430 billion when Biden was president.

The whole thing is just presented in such a way to make it look like they’re doing what Donald Trump would like them to do, when in fact it’s just exactly what Apple wants to do anyway. That’s what makes it genius. It’s win-win-win. It’s what Apple wanted to do anyway, it pleases Trump, and it’s actually good for the American economy.

Well, Meta has fixed the problem. Chris Vallance reports for BBC News:

Meta is under pressure to explain why it cancelled a major contract with a company it was using to train AI, shortly after some of its Kenya-based workers alleged they had to view graphic content captured by Meta smart glasses.

In February, workers at the company, Sama, told two Swedish newspapers they had witnessed glasses users going to the toilet, and having sex.

Less than two months later, Meta ended its contract with Sama, which Sama said would result in 1,108 workers being made redundant.

Meta says it’s because Sama did not meet its standards, a criticism Sama rejects. A Kenyan workers’ organisation alleges Meta’s decision was caused by the staff speaking out.

There’s no mystery here. The “standard” that Sama didn’t meet was keeping their mouths shut about the dignity-shredding nature of the entire operation. Like that fact that it even existed, let alone the gross privacy-invasive footage they witnessed. The deal wasn’t just for Sama’s workers to do the work, it was to do the work and keep it on the down-low. Go to Meta’s AI glasses website and try to find the part where they warn you that footage is subject to review by teams of contractors in third-world countries, Mechanical Turk-style. If you look hard enough, you’ll find oblique allusion to “review may be automated or manual (human)” in their legal small print, but their large-scale human review of video footage and recordings isn’t part of the brand or marketing image for their glasses.

During a complicated question from J.P. Morgan analyst Samik Chatterjee about product margins, Parekh unusually half-answered the question and then stopped and “turned it over to Tim” so that Cook could read an obviously prepared statement about tariffs, which included this bit:

In terms of applying for a refund of tariffs paid, we’re following the established processes, and we plan to reinvest any amount we receive back into U.S. innovation and advanced manufacturing. These would be new investments and would be in addition to our prior commitments in the U.S.

This is the sort of politics Cook will continue to be plying from the boardroom. Sure, Apple’s going to try to get its tariff money back. But it’s going to do so using the perfectly normal and established process, and if it does get billions back from the U.S. government, it double-promises to reinvest that money in the United States, above and beyond its already stated commitments. Trump Administration, take note.

The kind of logic puzzles I enjoy most are ones where, when the puzzle is posed, there’s no obvious solution. But once you see the solution, it seems profoundly obvious. Jason Kottke last week linked to 1D-Chess, a game from Rowan Monk that’s like that. Once you find the solution you can’t unsee it. (Don’t give up and peek at the posted answer!)

The question of tariff refunds is like that. Two months ago the Supreme Court ruled 6-3 that Trump’s obviously illegal tariffs last year, were, in fact, illegal. They left as an open question, however, whether importers who paid those tariffs should get refunds from the federal government. Apple, obviously, is one of those importers. The logic puzzle is this: if it turns out that Apple is eligible for a refund, how do they collect it without infuriating the petulant Donald Trump? Cook just spelled out the answer. Take the money but commit it all to their longstanding plan to spend $500 billion over the next few years to U.S. manufacturing efforts, a program they’ve maintained through the Trump 1, Biden, and now Trump 2 administrations, but which Cook has made dog-and-pony shows out of during both Trump terms to, as Trump himself describes it, “kiss his ass”.

That’s so obvious, now that Cook spelled it out, that it doesn’t even seem like a puzzle.

Update: “More on Apple’s Logically Elegant Tariff Refund Puzzle Solution”.

Sponsored by:

• Squarespace: Save 10% off your first purchase of a website or domain using code TALKSHOW.
• Drafts: The Swiss Army knife for text on all your Apple devices. Act now to get your first year of Drafts Pro for 50% off.
• Finalist: A daily planner for iPhone, iPad and Mac, built on proven paper-based planning methods. Use this link to get six months free.

On any given day, Los Angeles’s Hollywood Boulevard teems with tourists and street performers clustered near the area’s many landmarks. But in recent months, the strip has been set abuzz for a new reason.

Throngs of mostly adolescent boys and young men have been rushing the Church of Scientology’s international headquarters on the famed street.

The so-called “speed runs” appear to be bids for social media valor — clips of the raids have amassed millions of views on TikTok — as much as they are an outgrowth of public intrigue surrounding the church, which has previously drawn accusations of being a cult. Some users have developed blueprints for the building based on information gathered from videos of the raids.

I’m not condoning it, of course, but this video of one guy doing it sure was fun to watch.

Update: Now the kids are speed running Palantir too.

“Today Apple is proud to report our best March quarter ever, with revenue of $111.2 billion and double-digit growth across every geographic segment,” said Tim Cook, Apple’s CEO. “iPhone achieved a March quarter revenue record, fueled by such extraordinary demand for the iPhone 17 lineup. During the quarter, Services achieved yet another all-time record, and we were excited to introduce remarkable new products to our strongest lineup ever. That included the addition of the iPhone 17e and the M4-powered iPad Air, along with the launch of MacBook Neo, which is captivating customers all around the world.”

Record results for the January–March quarter for revenue, profit, and iPhone revenue in particular. And iPhones were constrained by supply. Tim Cook led the analyst conference call because he’s still CEO, but John Ternus joined for the first time. Worth a listen to just their opening remarks.

Over at Six Colors, Jason Snell has the usual assortment of graphs charting Apple’s numbers. Spoiler: up, up, up. And he has his usual transcript of the analyst call. From that transcript, this bit from Ternus’s prepared statement made me smile:

I want to echo Tim’s sentiment about our shareholders, especially those who have been with us for many years. Thank you so much for your confidence in our company. As you know, one of the hallmarks of Tim’s tenure has been a deep thoughtfulness, deliberateness, and discipline when it comes to the financial decision-making of the company, and I want you to know that is something Kevan and I intend to continue when I transition into the role in September.

Translation: “In case you were worried, we intend to keep making money hand over fist when I become CEO. I’m not coming into the job to spend the company’s money like a drunken sailor.”

Apple has all but given up on the Vision Pro after the M5 model failed to revitalize interest in the device, MacRumors has learned. Apple updated the Vision Pro with a faster M5 chip and a more comfortable band in October 2025, but there were no other hardware changes, and consumers still weren’t interested. [...]

The Vision Pro has been unpopular since it first launched, and Apple only sold around 600,000 units in total. Insider sources told MacRumors that Apple has received an unusually high percentage of returns, far exceeding any other modern Apple product.

Apple has apparently stopped work on the Vision Pro and the Vision Pro team has been redistributed to other teams within Apple. Some former Vision Pro team members are working on Siri, which is not a surprise as Vision Pro chief Mike Rockwell has been leading the Siri team since March 2025.

This report comes as news to everyone at Apple working in the Vision Product Group (VPG). Nothing about the future of the platform has changed recently. When it was a secret project, prior to unveiling, it was called the Technology Development Group (TDG) inside Apple. Then, when Vision Pro was unveiled, it became VPG. And then at some point the hardware went under Apple’s hardware group (led by John Ternus) and the software under the software group (led by Craig Federighi). So there have been changes, yes, but only the sort of changes that are natural when a product shifts from being a secret to being one of Apple’s regular non-secret platforms.

As for poor sales, I think it’s unquestionable that Vision Pro sales — and general enthusiasm — have been a disappointment. What momentum they had out of the gate has seemingly petered out. But the optimistic scenario inside Apple was not all that high. The best-case scenario was surely a bigger number of units than they’ve actually sold, but not that much more. There’s no realistic scenario where Vision Pro was an out-of-the-gate hit like, say, the iPad was. It’s an all-new device in an all-new product category that starts at $3,500 and costs more like $4,000 if you need corrective lenses. Before it debuted, there were multiple reports from multiple sources that suggested (a) that Sony could only manufacture a maximum of 900,000 displays per year, capping dual-display Vision Pro headsets at 450,000 per year; and (b) that Apple itself “expected to ship fewer than half a million headsets in the first year of its release, according to people involved in its supply chain” (per Wayne Ma at The Information).

Look at how Apple unveiled the second-gen Vision Pro with M5 — it was the definition of “low key”. I don’t think there was a single person in Cupertino — not one — who looked at first-generation Vision Pro sales and thought, “I know what will turn this around in a big way: a second-generation speed bump where the M2 chip is upgraded to an M5!” That speed bump in October was not intended to make a huge difference. It was just a signal that they’re still at it. Speed bumps are good. (And it probably helps, not hurts, margins because the M5 is used on Macs and iPads too, and no other product still in production uses the M2.) Rather than anyone — literally anyone — at Apple being surprised that the October second-gen M5 update did not meaningfully change the sales trajectory, I think the entire company would have been flabbergasted (and caught flat-footed on supply) if it had.

This sentence from Clover’s report is doing a lot of work:

Apple has apparently stopped work on the Vision Pro and the Vision Pro team has been redistributed to other teams within Apple.

There’s only one Vision hardware product to date, and that product, through two generations, is named Vision Pro. If Clover is saying that no one is working on a third-generation revision of the Vision Pro product we know today, maybe that’s correct. I don’t know. I certainly hope it’s correct. I think it was fine for Apple to do one new-generation speed bump of the original hardware. But going forward, they clearly need to do something significant for the next hardware. Ideally, two things: a much more appealing “Vision Pro” and a lower-priced “Vision Air” or just plain “Vision” or, hell, a “Vision Neo”. Take a new crack at the high end with a lighter-weight higher-resolution Vision Pro and open up new markets with something starting at under $2,000.

But I don’t think anyone is reading that sentence from Clover’s report that way. It implies — along with the headline — that Apple is just giving up on the whole platform. That’s how everyone is reading it, and it’s clearly what the article, and especially headline, implies.

I don’t think that’s true, at all. There’s a VisionOS 27 update coming at WWDC and new hardware in the works. Not just AR glasses, but immersive Vision headsets. There are, I believe, as many people at Apple working on VisionOS software and immersive content today as there ever have been. It’s full steam ahead. The pressure is on, I’m sure, but there’s no doom and gloom. The Apple folks in the Vision group aren’t oblivious.¹ They actually know the roadmap, and they know just how much work is between where the platform is today and where it needs to be for it to be a meaningful contributor to Apple’s bottom line. But they’re there, working on it. I don’t know who told MacRumors what (and their sourcing is just “MacRumors has learned”), but I know for a fact that it is not true that the teams working on the Vision platform have “been redistributed to other teams within Apple.”

It’s a strange thing for MacRumors to state so categorically something I believe has no truth to it whatsoever. And if there is some truth to it, it’s not what the article implies, which is that the whole thing has been shut down, somehow without the world knowing until now. Just two weeks ago John Ternus and Greg Joswiak were interviewed by Mark Spoonauer at Tom’s Guide, and both spoke of a bright future for spatial computing. Joz describing Vision Pro as a product pulled into the present from the future is a good way of emphasizing the yet regarding a product — and category — that’s not there yet. Apple executives know how to give a non-answer answer to a question they don’t want to answer honestly. (Exhibit A: Tim Cook “squashing” rumors that he was about to retire ... one month before he announced he was stepping aside as CEO.) The way Ternus and Joz were talking about the platform, and immersive content, this month was not lacking in enthusiasm. It was asking for patience.

It’s certainly possible that this Vision thing ultimately isn’t going to work out and Apple will throw in the towel on it. But that hasn’t happened, and if it does, it’s not going to come out of nowhere as a story on MacRumors for the people in VPG working on it. When Apple threw in the towel on Project Titan (the car project) in February 2024, an all-hands was held to break the news, led by then-COO Jeff Williams and Titan project lead Kevin Lynch. The team didn’t learn it from a fucking leak.

In 2024, the port — which manages the Oakland airport — changed the name from Oakland International Airport to San Francisco Bay Oakland International Airport, hoping to entice travelers by emphasizing the hub’s proximity to SF. At the time, the number of people flying into Oakland was declining after a brief post-pandemic rebound, and the airport was losing routes.

The effort largely failed, while having the secondary impact of annoying San Francisco leaders, who swiftly sued, arguing that the name would confuse travelers. In 2025, the port swapped the two cities within the name to produce “Oakland San Francisco Bay Airport.”

San Francisco didn’t like that either, but the parties entered mediation in December and have now settled. The new name is fine, as long as “Oakland” always appears before “San Francisco” in all materials and the airport does not add the letters SF to its code, OAK.

The Standard ran this under the cheeky headline “Little-Known Bay Area City Will Keep San Francisco in Its Airport’s Name”, which is a little funny, but I don’t see the need to punch down like this. Nobody calls the city “San Francisco” anyway. Everyone just calls it “San Fran” or “Frisco”, either of which names are acceptable.

Today the first witness was sworn in in Musk v. Altman: Elon Musk. I was surprised by how flat he seemed.

This is not the first time I’ve seen Musk in court. During his defamation suit, he turned on the charm and the jury responded by finding him not guilty. Today he looked adrift and unprepared. The only times he showed real animation were when he was bragging about how much he’d done for OpenAI. [...]

What did Musk do at OpenAI? “I came up with the idea, the name, recruited the key people, taught them everything I know, provided all the initial funding. Besides that, nothing.” He paused for laughter, and one or two people obligingly chuckled. But most of the courtroom was silent. I thought he sounded petulant. “I could have started it as a for-profit and I chose not to,” Musk said.

Elon Musk petty, petulant, smug, awkward, and unprepared? Where’s my fainting couch?

At another point, Musk was asked to explain who former OpenAI board member Shivon Zilis was. “Shivon was the, um, my chief of staff and, uh, you know,” Musk said. One person in the gallery — presumably familiar with the fact that Zilis is the mother of a few of Musk’s kids — burst out in loud laughter. But the jury looked puzzled.

In an earlier post today, I pointed out that it’s at least slightly awkward that NYT reporter Mike Isaac is live-tweeting the proceedings on Twitter/X, a platform owned by Musk (and where he famously has rigged the algorithm in his own favor), but, well, when you own an empire as sprawling as Musk’s, pieces of it are bound to collide.

It’s the same with women who are the mothers of Musk’s children.

Musk is asking that Altman be removed from OpenAI’s board, that the company convert back to a nonprofit, and for the return of allegedly “ill-gotten gains” — some $150 billion — which Musk says would go to OpenAI’s charitable trust. Outside legal experts say that Musk is unlikely to win all or even much of this. His argument is confusing: OpenAI has certainly evolved from a nonprofit lab to a revenue-chasing, consumer behemoth, and a chorus of critics has alleged that it has deviated from its original mission of ensuring that AGI benefits humanity. But Musk himself appears to have insisted that OpenAI couldn’t keep up as a nonprofit — for instance, in early 2018, he wrote an email to OpenAI leadership saying that merging the firm with Tesla “is the only path that could even hope to hold a candle to Google.” And even before he sued, Musk launched a rival for-profit company, xAI. “Mr. Musk’s lawsuit is a pageant of hypocrisy,” William Savitt, a lawyer for OpenAI, told the jury today, later adding that Musk had “sour grapes.” [...]

The trial makes the AI boom seem sordid and small. In his sworn deposition, Altman wrote that Musk used to message him complaints that he wanted more credit for the success of OpenAI and took offense at not being included in an anniversary photo. [...] Musk, for his part, has said that he would drop his lawsuit if OpenAI changed its name to “ClosedAI.” Yesterday, as jury selection began, Musk began furiously posting on X and repeatedly called his co-founder “Scam Altman.” Before the start of opening arguments today, Gonzalez Rogers admonished Musk and Altman for their social-media use, asking them to limit their “propensity” to post about the trial; both meekly assented, “Yes.”

It all seems so petty and spiteful, but this is a real federal lawsuit with the entire future of the biggest startup in history at stake.

On the first day of testimony in a landmark trial between Elon Musk and OpenAI’s Sam Altman, two notably different tales were offered of how OpenAI evolved from a nonprofit artificial intelligence lab into one of the most influential tech companies in the world.

In Mr. Musk’s telling, OpenAI’s shift was one of the greatest heists in history — a nonprofit ripped from its promise of altruism by the greed of Mr. Altman, who founded OpenAI with Mr. Musk and a group of A.I. researchers more than 10 years ago. In OpenAI’s recounting of those early days, however, it was Mr. Musk who was the voracious capitalist. And when the lab’s other founders refused to go along with his plans, he left in a huff.

“This lawsuit is very simple: It is not OK to steal a charity,” Mr. Musk said Tuesday on the witness stand in an Oakland, Calif., courtroom. If Mr. Altman and OpenAI are allowed to continue with their plans, he added, “It will give license to looting every charity in America.”

A nine-member jury, seated a day earlier in federal court by Judge Yvonne Gonzalez Rogers, will hear from tech moguls, former OpenAI board members and employees in what is expected to be a monthlong trial. The jurors’ decision could shift the balance of power among A.I. companies, with Mr. Musk seeking $150 billion in damages and an order that OpenAI, now valued at about $730 billion, unwind its for-profit plans.

Gonzales Rogers, you will surely remember, presided over the Epic v. Apple case.

Isaac is live-tweeting the testimony and goings-on on Twitter/X. Here’s his thread of posts yesterday. Isaac is a terrific reporter, and I enjoy following his extemporaneous notes. It’s a little weird though that he’s posting these on Twitter/X, a site that is privately owned by one of the parties in the lawsuit. Musk’s empire is so sprawling that separate pieces inevitably collide.

Sunday’s crossword puzzle in the print edition of The New York Times Magazine contains a grid that does not match the clues. The correct version of the puzzle can be found in the news section of Sunday’s print edition of The Times. The puzzle on our app is correct.

Maggie Duffy, writing for Vulture:

Some solvers who, like Wegener’s wife, complete the Sunday puzzle in the print magazine (often with pen) complained on crossword forums and social media, saying they were “nearly in tears,” some with fears of “sudden onset dementia” or, worse yet, ineptitude.

For Irene Papoulis, a former writing instructor at Trinity College, the puzzle is typically a source of pride. “It didn’t even occur to me that it could be their mistake,” she told me. “I just blamed myself.” When Mike McFadden, in New Jersey, couldn’t crack it, he had a similar reaction. “I thought something was wrong with me,” he told me. “I didn’t think that they would have an error.” It nagged at him all day. At a function on Saturday, he couldn’t bring himself to mention it to his brother-in-law, a fellow solver; he was still too upset.

Some had such trust in the crossword that they believed the erroneous grid was purposeful. “I’m saying to myself, ‘Okay, maybe there’s some sort of scientific or mathematical trick,’” McFadden said. When I spoke with Will Shortz, the Times’ crossword editor, he said the Times does “so many tricks with the puzzles” that he could see how someone’s first thought would be “I wonder what they’re up to now?”

This is the first such mistake the Times has made in the 84 years that they’ve been printing a crossword puzzle. I came of age doing work in print — writing and editing The Triangle, the student newspaper at Drexel, and then spending a few years as a working graphic designer, at a time when print still ruled. There’s an inherent stress about going to press. Mistakes are forever. We once ran a headline at The Triangle that read “Headline Goes Here”. Once. Going to press is stressful but exhilarating. There’s an adrenaline rush that comes with giving the go-ahead to start a very expensive large-scale full-color press run. The stress focuses the mind.

Print, effectively, is hardware. Atoms, not bits. The web is literally software. If you make a mistake in software that results in incorrect mathematical results, you ship an update. If you make a mistake in a CPU such that it results in incorrect floating-point math, perhaps only in 1 out of every 9 billion calculations, people will remember the mistake 30 years later.

If The New York Times had run the wrong crossword grid on the web or in their app, they would have corrected the error quickly, few people would have encountered it, and fewer still would remember it. But by printing the wrong grid in the Sunday magazine last week, they made a mistake that some people will never forget (and some will never forgive).

Hardware brain is different from software brain. Software brain says Go faster; do more; the only mistake you can’t fix is having gone too slow. Hardware brain says Slow down; do less; focus; strive for perfection and never settle for less than excellence; mistakes are forever.

If his background in hardware means that incoming Apple CEO John Ternus has hardware brain, and will lead Apple accordingly, that suggests Apple will double down on zigging in the midst of a still-escalating AI hype cycle that has the rest of the industry zagging ever more frenetically. That feels right to me.

There’s no reason to assume that a change at the helm is imminent. Cook may be older than the CEOs of the other tech companies at the top of the S&P 500, but he’s hardly the oldest person running a major corporation. “If Trump or Biden can be president at 80, Tim Cook can be CEO of Apple for many more years. It used to be automatic that CEOs are moved out at 65,” says someone who knows him. “The world has changed.”

While Cook hasn’t given any indication how long he’ll remain in charge — other than telling Dua Lipa it would be “a while” — people close to him believe he’ll be CEO at least another three years. After that, they say, he’ll start a charitable foundation to donate the wealth he accumulated at Apple.

If Cook were to stay that long, people within Apple say, the most likely successor would be John Ternus, the hardware engineering chief. In a company whose success has always come from building category-defining gadgets, the ascension of a hardware engineering expert to the CEO job would seem logical. Ternus, who’s not yet 50, would also be more likely than other members of the executive team to stick around for a long time, potentially providing another decade or more of Cook-esque stability.

Ternus is well-liked inside Apple, and he’s earned the respect of Cook, Williams and other leaders. “Tim likes him a lot, because he can give a good presentation, he’s very mild-mannered, never puts anything into an email that is controversial and is a very reticent decision-maker,” says one person close to Apple’s executive team. “He has a lot of managerial characteristics like Tim.” Christopher Stringer, a former top Apple hardware designer, called Ternus a “trustworthy hand” who’s “never failed with any role he’s been elevated to.” Eddy Cue, the Apple executive known as Cook’s closest confidant, has privately told colleagues that Ternus should be the next CEO, according to a person with knowledge of the matter.

Linking to Gurman’s report, I wrote:

I wouldn’t have linked to this if not for the above line about Eddy Cue. If Cue is telling people that, that means a lot. No executive at Apple is more juiced-in company-wide than Cue. Cook’s first action as CEO was to promote Cue, and Cue was arguably just as tight with and trusted by Steve Jobs.

It was two more years, not three, but Gurman was the first to report that Ternus was the guy at the top of the list.

There was no significant additional reporting between Gurman’s May 2024 Bloomberg report until November 15 last year, when the Financial Times published a blockbuster story under the headline “Apple Intensifies Succession Planning for CEO Tim Cook”, with four bylines: “Tim Bradshaw, Stephen Morris and Michael Acton in San Francisco and Daniel Thomas in London”. Bradshaw is the FT’s lead Apple reporter, and it’s no coincidence his name was first among the four. The article gets right to the point at the start:

Apple is stepping up its succession planning efforts, as it prepares for Tim Cook to step down as chief executive as soon as next year. Several people familiar with discussions inside the tech group told the Financial Times that its board and senior executives have recently intensified preparations for Cook to hand over the reins at the $4tn company after more than 14 years.

John Ternus, Apple’s senior vice-president of hardware engineering, is widely seen as Cook’s most likely successor, although no final decisions have been made, these people said.

People close to Apple say the long-planned transition is not related to the company’s current performance, ahead of what is expected to be a blockbuster end-of-year sales period for the iPhone. [...]

The company is unlikely to name a new CEO before its next earnings report in late January, which covers the critical holiday period. An announcement early in the year would give its new leadership team time to settle in ahead of its big annual keynote events, its developer conference in June and its iPhone launch in September, the people said. These people said that although preparations have intensified, the timing of any announcement could change.

So, per the FT in November, Apple’s plan was to name Ternus as the company’s next CEO “early in the year”, after their Q1 results (January 29) but ahead of WWDC (June 8). The halfway point between those dates was April 4; Apple announced Ternus as the company’s next CEO on April 20. Every single word of the FT report, in hindsight, was exactly correct. I can’t think of a way that their November story could have been more prescient. It was a home run. A report for the ages, like when CNet and The Wall Street Journal scooped the Mac’s transition to Intel processors on the eve of WWDC 2005.

My own take, back in November when the FT report dropped, was that it had the distinct aroma of a deliberate expectations-setting leak, and was almost certainly accurate:

That “several people” spoke to the FT about this says to me that those sources (members of the board?) did so with Cook’s blessing, and they want this announcement to be no more than a little surprising. [...]

I would also bet that Cook moves into the role of executive chairman, and will still play a significant, if not leading, role for the company when it comes to domestic and international politics. Especially with regard to Trump.

Cook moving into the position of executive chairman and continuing to play a leading role as the company’s political ambassador was my own speculation, and that proved out. Easy money, making that prediction.

One week after the FT’s report, in his Bloomberg “Power On” newsletter on November 23, Gurman wrote:

In October, I wrote that the internal spotlight on Ternus was “intensifying,” and that barring unforeseen circumstances he would be the leading candidate. But I didn’t put a date on when a change might happen. Then, around midnight two Fridays ago, the Financial Times published a report with three central claims: Apple is “intensifying” succession planning; Ternus is likely the next CEO; and Cook is expected to step down between late January and June.

The first two points are anything but revelations if you’ve read Bloomberg coverage and Power On, or have simply been paying attention to the realities of Cook’s age and tenure. The timing, however, is another matter entirely. It’s a huge deal that the FT did this: A respected publication should only predict the CEO transition date for a company of Apple’s scale with a high level of confidence — based on people legitimately in the know.

This is where I have concerns. Based on everything I’ve learned in recent weeks, I don’t believe a departure by the middle of next year is likely. In fact, I would be shocked if Cook steps down in the time frame outlined by the FT. Some people have speculated that the story was a “test balloon” orchestrated by Apple or someone close to Cook to prepare Wall Street for a change, but that isn’t the case either. I believe the story was simply false.

Gurman must be well and truly “shocked” by this week’s announcements, because as it turns out, Cook is stepping aside exactly “in the time frame outlined by the FT”. The FT’s report was not “simply false”. It was, in fact, completely true. The Financial Times, which truly is a respected publication (with no black marks on its record, like, say, Bloomberg’s to-this-day-still-uncorrected “The Big Hack” fiasco), obviously did have a high level of confidence in Apple’s plans, because they were, in fact, briefed by people “legitimately in the know”. Gurman’s reading comprehension is questionable as well, because the FT did not report that Cook would “step down” between January and June. The FT report spoke only of “naming a new CEO” and making an “announcement” between January and June. That’s exactly what happened. Nor is anyone “departing” — but a change in leadership will occur in the middle of the year.

In January, Gurman reiterated his stance that the FT was wrong:

It’s just a question of timing. The Financial Times reported last year that the change would happen as early as the beginning of 2026. But let me be clear: This seems unlikely.

By pooh-poohing the FT’s completely accurate reporting as “simply false”, Gurman wound up poo-pooing the bed. Calibrate the grains of salt with which you take his other reporting on Apple executive goings-on accordingly. A humble correction and sincere apology to the Financial Times — and Tim Bradshaw personally — are surely forthcoming in this weekend’s edition of Power On.¹