Eshu Marneedi

John Gruber, writing at Daring Fireball:

MacOS 12 Monterey came out in 2021. So I think that means you can one-time purchase and download an older version of Pixelmator, if you’re running an older version of MacOS. But if you’re running MacOS 26 Tahoe, you’ll get the new Liquid-Glassified version of Pixelmator whether you get it as a one-time purchase or through a Creator Studio subscription. I think? Update: That was wrong. It’s a little simpler than that, in that Pixelmator Pro is an outlier from the other apps in Creator Studio. The new version of Pixelmator Pro — version 4.0 — is only available through the Creator Studio subscription, and requires MacOS 26 (and iPadOS 26). The one-time purchase version of Pixelmator Pro is version 3.7.1 — the existing version, last updated two months ago — and that’s the version you get from MacOS 12 through MacOS 26 if you get it via one-time purchase. Pixelmator Pro is the only app in Creator Studio where the new version is exclusively available through the Creator Studio subscription.

Gruber cites the “About Apple Creator Studio” knowledge base article on Apple’s website to explain the confusion. I definitely think something fishy is afoot here, but I’m not sure whether I fully agree with the conclusion (though time will tell). Here’s the relevant quote from the article:

Final Cut Pro, Motion, Compressor, Logic Pro, MainStage, and Pixelmator Pro are also available as one-time purchases for Mac on the App Store. If you previously purchased one of these apps and you also have an Apple Creator Studio subscription, you can use either version of the apps. You can have both versions of these apps installed on your Mac. To make it easier to distinguish versions, the apps in Apple Creator Studio have unique icons…

Minimum system requirements for each Apple Creator Studio app are as follows: All Mac apps require macOS 15.6 or later, except for Pixelmator Pro, which requires macOS 26 or later.

This part of the article is indeed confusing because it says there’s a difference between the Apple Creator Studio versions and the one-time-purchase ones, yet it doesn’t further elaborate on them later in the article. When Apple says Pixelmator Pro requires macOS 26 Tahoe, does it mean the one-time-purchase version or just the Apple Creator Studio one? Gruber believes it’s the latter, but the wording is ambiguous. Apple could just mean that Pixelmator Pro is an exception to the rule, and that both the one-time-purchase and Creator Studio versions will require macOS Tahoe (but why?). If Gruber is correct, however, this would mean Pixelmator Pro is the only Creator Studio app whose one-time purchase version will no longer receive updates (also, why?). I think the likelihood of this is slim. Again, we’ll find out for sure on January 28, when Apple Creator Studio officially launches.

As I wrote in my initial Apple Creator Studio reaction article, I don’t mind the Liquid Glass design in photo editing apps. I actually think Acorn, by Flying Meat, does this particularly well: photo editors are the unique case where I want the user interface to recede into the background. The photo should be front and center. I would mind it a lot more if Final Cut Pro received the Liquid Glass design, but it doesn’t, which is good. But if the Liquid Glass design — and future updates — are restricted to a Creator Studio-exclusive Pixelmator Pro 4.0, I think that would be enough for me to pay $130 (well, $30 with my student discount) for the bundle. (I’m a sucker for Pixelmator Pro, an app I regard so highly that it’s one of my “essential Mac apps” listed in the colophon of this blog.) Of course, that would also require me to put up with the new asinine Pixelmator Pro icon, so I’m hopeful Apple just continues updating Pixelmator Pro for users who have already purchased it.

On Photomator’s lack of mention in the Creator Studio bundle, Gruber writes:

If Photomator did not have a future as part of Creator Studio, I think Apple would have used this moment to stop selling the existing version. They’d say that it too remains functional but is no longer being updated. But that’s not what they said…

My guess is that Apple and the acquired Pixelmator team are hard at work on a new Creator Studio version of Photomator, including a version for iPad, and it just isn’t finished yet. I’m more unsure whether they’ll keep the Photomator name (which I think is too easily conflated with the Pixelmator name) than whether they’re working on an ambitious update to the app to include in Creator Studio.

I have no little birdie insider information about that, just my own hunch. I just think that if Photomator didn’t have a future, Apple’s statement about it would say so, and they’d stop selling the current version.

Photomator is another one of my most beloved Mac apps. Combined with Pixelmator Pro, the two have become my de facto replacement for Adobe’s products, which I find to be poorly designed and priced predatorily. (Pixelmator Pro is to Photoshop what Photomator is to Lightroom.) Before the Creator Studio announcement earlier in the week, I was most nervous about Pixelmator Pro being sunsetted, to the point where I’d even forgotten that Photomator was perhaps a more likely target for Apple’s ire. And lo and behold, Apple didn’t mention Photomator in its Creator Studio announcement, which led me to believe that it would fade away soon, just like Pixelmator (the iOS photo editor Apple discontinued on Tuesday, not the company). As a Photomator user, this makes me anxious.

It is still quite possible that Apple will discontinue Photomator entirely and roll it into the Photos app on iOS and macOS. But it’s a 50/50 whether or not this happens; Gruber could also be correct that a new version of Photomator is coming later this year, whenever it’s ready. Many years before Photomator, I was an avid Aperture user, which, much like Photomator, was a beautiful, Mac-native Lightroom competitor for editing photos in your iPhoto library. People have wanted Apple to revive Aperture for years now, and Apple, in a way, did: it bought Photomator, a modern-day version of Aperture. It would truly be a shame if Apple flubbed a prime opportunity to make a true competitor suite to Adobe’s products: a video editor (Final Cut Pro), an audio editor (Logic Pro), a Lightroom competitor (Photomator), and a Photoshop competitor (Pixelmator Pro). All for $130 a year — with options to outright purchase the apps separately — and that would truly be the deal of a lifetime. Apple is close, and I really don’t want it to fumble the ball here.

George Hammond and Cristina Criddle, reporting for the Financial Times:

OpenAI is introducing advertising on ChatGPT, as the $500bn start-up seeks new sources of revenue to fuel its continued expansion and fend off fierce competition from rivals Google and Anthropic.

The San Francisco-headquartered company announced on Friday that it would begin testing adverts on its free chatbot and cheapest paid offering.

The ads will appear at the bottom of ChatGPT answers in the coming weeks. OpenAI said the marketing messages will be clearly labelled and will appear if relevant to the query.

Sam Altman, OpenAI’s chief executive, in 2024: “I kind of think of ads as like a last resort for us for a business model. I would do it if it meant that was the only way to get everybody in the world, like, access to great services, but if we can find something that doesn’t do that, I’d prefer that.”

It’s safe to say that OpenAI hasn’t found that in the year since Altman gave this quote. But unlike most others in the artificial intelligence space who learned this news on Friday, I don’t think Altman was lying in this interview. I really do think OpenAI hasn’t found a way to monetize its users and is now looking for a way to make it big in an initial public offering. Today, ChatGPT loses money on every free user, and its only business strategy to stop hemorrhaging cash has been upselling people on $20 and $200 monthly subscriptions. That freemium business model doesn’t work when a company wants to make a profit — there’s no room for losing money.

Come to think of it, OpenAI really isn’t doing great these days. The company is under tremendous legal pressure after a series of high-profile deaths involving ChatGPT. The working population of the United States has largely soured on AI, despite using it heavily, wary that it might contribute to job loss. The tide is turning on AI, and people are directing their attention to its negative effects. Meanwhile, Anthropic’s Claude Code and flagship Claude Opus 4.5 model have demanded attention across the internet. People are vibe-coding — using AI to write software with little to no human intervention — various apps, games, and websites for themselves, then raving about the results on social media. Fortune 500 companies run on Claude these days — even Apple uses it internally, and even Elon Musk, one of the most antisocial figures in technology history, can’t help but admit Anthropic has done something incredible in the professional AI space.

On the consumer side of the market, Gemini has shot up in the rankings. As of mid-January, it claims the top spots on the LMArena benchmark, a website that pits models head-to-head against each other and asks users to choose the response they prefer. Gemini’s pre-training technology is simply best-in-class, and both Gemini 3 Flash and Gemini 3 Pro are leading frontier models that put OpenAI’s GPT-5.2 to shame. Where Anthropic has a distinct advantage in post-training, Google is superior in pre-training. GPT-5.2 has decent pre-training — though seemingly not close to Gemini — but OpenAI’s post-training is horrible. GPT-5.2’s responses sound like a human resources video about sexual harassment in the workplace. They’re consistently below awful and just plain unhelpful. Gemini doesn’t have as much of a personality as Claude, but it makes up for it in pre-training, and it shows. Comparing Gemini’s responses to ChatGPT is just embarrassing for Altman’s company.

OpenAI hasn’t been with it since at least the launch of OpenAI o1, the first reasoning model from the GPT-4o line of large language models. That was the last groundbreaking, class-leading model from OpenAI — it was the first model to use a chain of thought, and the result was significantly better responses. Gemini was lackluster at the time, and programmers flocked to o1 before Claude Code, the command-line utility. Since then, OpenAI has been on a downward trajectory, clearly indicated in both its model quality and apparent desperation to turn a profit. So now we’re here, at the beginning of 2026, and OpenAI announced two last-ditch efforts to make more money for an IPO: the new ChatGPT Go plan and advertising. (ChatGPT Go was available in poorer countries, like India, to lower the cost for more usage. The fact that it’s now available in the United States is either a recession indicator or proof that OpenAI isn’t doing well financially.)

I don’t think this is going to go well for OpenAI. When the AI bubble inevitably bursts, I imagine the company will plunge into bankruptcy. The “bubble” I speak of is more of a financial situation than a cultural one. Currently, the AI industry effectively works on glorified gift cards. Microsoft gives OpenAI a gift card for free Azure credits to run ChatGPT. Nvidia gives OpenAI a gift card to buy more graphics processing units. OpenAI gives Cursor a gift card for access to ChatGPT. I think the Cursor example is perfect for this: neither company is profitable. Whenever this bubble bursts, every company that isn’t already profitable will go bankrupt, since the gift cards will all expire. At some point, OpenAI will need to pay Microsoft real United States dollars to get access to servers. Microsoft will need to pay OpenAI real money to use ChatGPT models in Copilot. This is not very good news for the severely-in-debt OpenAI, which is working with way too many gift cards and way too little cash.

Here’s how I imagine the next five years will go: OpenAI will IPO, the bubble will burst, its stock will crash, and the company will become a group of patents and debt. Some company, whether it be Apple, Google, or someone else, will acquire OpenAI’s debt and technology, roll it into their existing model infrastructure, and OpenAI will be reduced to a Wikipedia page we tell children about in the year 2042. I have no reason to believe this won’t happen. OpenAI’s actions clearly allude to the fact that it is desperate for cash, and companies in a mad scramble for revenue in a bubble don’t tend to fare well outside of that bubble. Meanwhile, Google is profitable, and while its AI ventures are loss leaders, Google Search is well equipped to take over ChatGPT’s prominence. As a secondary and less ambitious prediction, I believe the separate Gemini website and app will be integrated into Google’s AI Mode, which would replace the “AI Overviews” that currently have primary placement in Google Search.

Google really does seem primed to usurp the consumer AI market. Its deal with Apple to handle pre-training for the models that power Apple Intelligence is a possible center of revenue — similar to the infamous search deal between the two companies — and AI Mode increasingly looks like the future of consumer AI. Google is quite literally the most popular website on the internet. With the way Google DeepMind keeps chugging away at pre-training, Gemini’s models will eventually become so good at web search and agentic work that, combined with Google Search’s prominence and deals with other websites, Gemini models will become the primary way people interact with generative artificial intelligence on the web. Meanwhile, Anthropic seems ready to capitalize on the enterprise and professional markets — the company makes most of its money selling enterprise subscriptions to Claude. I don’t imagine that’ll change. OpenAI is nowhere in this equation.

Apple Newsroom:

Apple today unveiled Apple Creator Studio, a groundbreaking collection of powerful creative apps designed to put studio-grade power into the hands of everyone, building on the essential role Mac, iPad, and iPhone play in the lives of millions of creators around the world. The apps included with Apple Creator Studio for video editing, music making, creative imaging, and visual productivity give modern creators the features and capabilities they need to experience the joy of editing and tailoring their content while realizing their artistic vision. Exciting new intelligent features and premium content build on familiar experiences of Final Cut Pro, Logic Pro, Pixelmator Pro, Keynote, Pages, Numbers, and later Freeform to make Apple Creator Studio an exciting subscription suite to empower creators of all disciplines while protecting their privacy…

Apple Creator Studio will be available on the App Store beginning Wednesday, January 28, for $12.99 per month or $129 per year, with a one-month free trial, and includes access to Final Cut Pro, Logic Pro, and Pixelmator Pro on Mac and iPad; Motion, Compressor, and MainStage on Mac; and intelligent features and premium content for Keynote, Pages, Numbers, and later Freeform for iPhone, iPad, and Mac. College students and educators can subscribe for $2.99 per month or $29.99 per year. Alternatively, users can also choose to purchase the Mac versions of Final Cut Pro, Pixelmator Pro, Logic Pro, Motion, Compressor, and MainStage individually as a one-time purchase on the Mac App Store.

The bundle itself is quite unremarkable, and it has been rumored for a while. I think it’s a pretty decent deal, especially compared to Adobe, its chief competitor. Subscribers get access to all of Apple’s Pro Mac apps, plus their iPad variants where applicable, along with some ChatGPT-powered features in the iWork apps for $130 a year. And anyone who doesn’t want another subscription can still purchase the Pro Mac apps on the App Store at their same prices and own them outright. This is the option I chose many years ago, and I’ve been satisfied with the decision since. Unless one needs the iPad versions of the apps, I suggest most people buy the apps they want outright if they use them regularly and forgo the subscription. They’ll pay for themselves in a few years. The market for this subscription is mostly people who only want access for a few months at a time or really need the iPad apps, which still remain subscription-only. (The new Pixelmator Pro iPad app is pretty neat, but more on that shortly.)

All of the Pro Mac apps, with the sole exception of Pixelmator Pro, have forgone the Liquid Glass design. They remain visually identical to their prior versions. I think this is good for Pro-focused Mac apps, since the Liquid Glass design on the Mac really is more of a distraction than an affordance. Liquid Glass looks nice in photo editing apps — which is why Pixelmator Pro presumably has it — but otherwise, it’s probably useless. Just look at Xcode: the only Liquid Glass update it received is to its windows’ corner radius. Making the text editor glassy, for instance, would just be an unnecessary nuisance.

Lest the casualties be forgotten: Pixelmator, the Pixelmator company’s iOS photo editor, is now discontinued, just a year after Apple’s acquisition of the company. Here’s what I wrote in November 2024, as the acquisition was announced:

This is the beginning of the death of a beloved product.

I would be ecstatic to be wrong. I really do love Pixelmator Pro, and I want it to become even better, more ingrained into macOS, and for it to thrive with all of Apple’s funding, just like Beats did. I loved Aperture, and if Apple fused all the features from that bygone app with Pixelmator and Photomator, I’d be happy. But even if Apple did all of that — even if Apple cared about loyal Pixelmator Pro users — it would slap a subscription onto it and eliminate the native macOS codebase because Apple itself cares more about the iPhone and iPad than it does the Mac.

I was partially correct, though my nervousness about Pixelmator Pro turned out to be unfounded. Pixelmator really was beloved, and I wanted it to thrive as a separate app, but alas, it is a relic of the past. And Apple did slap a subscription onto Pixelmator Pro and made the iPad app only accessible via that subscription. I still frown upon independent app acquisitions, and think that if Pixelmator remained independent, it would thrive under the current Apple design climate.

When I say, “current Apple design climate,” I don’t mean Liquid Glass or macOS 26 Tahoe. I’m referring to the truly deplorable iconography all of the refreshed apps have been treated to. Words cannot adequately describe my sheer disgust, hatred, and shock at how asinine these icons are. They look as if Xiaomi was contracted to rip off Apple’s design language. The Final Cut Pro icon, currently a rainbow slate, has been replaced by a purple line art illustration of the same slate. Who at Apple Park decided purple was a fitting color for this icon? Shouldn’t it be polychromatic? And what is the Pixelmator Pro icon even supposed to be? It’s an opaque red squircle with a hollow squircle floating above it, and below the two squircles are three dots connected by lines. What even is that? I’ve been looking at it on and off for hours now — what is that?

And the new Motion icon is quite literally an unfinished McDonald’s logo, but pink instead. Maybe it’s supposed to spell out “M for Motion,” but the trajectory of the figure (it’s presumably being drawn) doesn’t look like the second arch is high enough for it to be an M. Again, what is that? What is this?! I have a strict no-profanity rule on this website, but I have never wanted to disobey it until today. I have no constructive criticism to add here. I can’t comprehend what was happening in the minds of the people who drew these icons. They should all be fired and never allowed near a graphics design program for the rest of their lives. A 14-year-old could draw better icons than this — I could draw better icons than this. What is MainStage even supposed to represent? It looks like the icon in Apple Health for the pill reminders feature, for crying out loud.

But that Pixelmator Pro icon, for Chrissake. I think that truly might be the worst icon in macOS Tahoe. I don’t know how it was approved — I don’t even think it was approved. ChatGPT made that icon or something. It can’t be real. I can’t believe this will ship in a few weeks.

A correction was made on January 13, 2026, at 10:36 p.m.: An earlier version of this post said Photomator by the Pixelmator company was being discontinued. This is inaccurate; Pixelmator is being discontinued. I regret the error.

Nikita Bier, X’s head of product, in a now-deleted post:

The most definitive measure of posting ability is your follower-to-total posts ratio (essentially: how many posts did it take you to get where you are today)?

We should probably start showing this instead of follower count to signal account quality.

The problem with Bier is that he only sees dollar signs everywhere. If it’s not a way to sell more ads via increasing daily active users or active minutes, he doesn’t want it. He doesn’t care about his users or their thoughts; he cares about making money. Scrolling through Bier’s account, this is immediately apparent. Here’s a selection of quotes:

• January 11, teasing the upcoming Smart Cashtags feature: “X is the best source for financial news — and hundreds of billions of dollars are deployed based on things people read here.”
• January 10, on the new algorithm: “The algorithm was rebuilt from scratch by the xAI team and now runs on +20K GPUs at the Colossus data center… Time spent is up 20% and follows are up even more. But there is still so much more to improve.”
• January 9, about the new Priority notifications tab: “You’ll now get a notification anytime someone follows you with more followers (or has >150K)”
• January 4, Bier-ing: “X has hit the ground running in 2026: It is the highest engagement start-of-the-year in X’s history — with yesterday being among our biggest days.”
• December 15, on the new xAI-powered algorithm: “X is on track for the best December in the app’s history. There are still some rough edges to polish before the holidays, but the algorithm team is making huge strides with the timeline.”

There’s a clear theme here: Bier only cares about boosting time spent on the app. Time spent is up 20 percent! Follows are up even more! You’ll get a notification when someone famous replies! Highest engagement start-of-the-year! Yesterday was one of our biggest days! December was the best in the app’s history! This is exhausting, if not patronizing, but it makes sense coming from him. Bier’s claim to fame is a hot-or-not app called Gas — and one that came before it, called TBH (stylized as “tbh”) — that was acquired by Discord and shot up to the top of the App Store rankings because teenagers loved to engage with it. That’s Bier’s cup of tea: making useless apps for teenagers without fully developed prefrontal cortices.

Back to X. Elon Musk strutted into Twitter headquarters with a fundamental misunderstanding of why Twitter was broke. He blamed it on institutional bloat, which undeniably existed, but employee salaries and “proudly woke” T-shirts weren’t driving the company into the ground. Neither was the application programming interface, nor were the numerous deals with publishers and researchers to bring Twitter data into more contexts, such as to fuel election misinformation research. Twitter was so strapped for cash because it put user experience ahead of any metric. Its one profitable year, 2020, was a complete fluke, and it returned to unprofitability as soon as the news cycle slowed. It could have been profitable if it tried to artificially extend the 2020 wave into the Musk era, but it didn’t — it focused on authenticity, not money.

X/Twitter is full of people shouting into the void until something works. Viral tweets are written by people with 56 followers who wake up one morning to the disbelief that their shower thought got 200,000 likes overnight. That’s the beauty of Twitter, and that’s why people keep coming back. Tens of thousands of people use Twitter as their journal, posting quick little quips or reactions to things from their day. There’s an entire account dedicated to a woman posting pictures from her walks and what she ate for dinner — that account gets thousands of likes every day. She’s not a celebrity or a “high value” account by any of Bier’s metrics, but she and people like her make Twitter the quirky, wonderful place that it is. Not everyone is Taylor Swift, with very few tweets and very many followers. Boosting only such accounts proves a fundamental misunderstanding of social networking.

The harsh yet utterly rewarding quality of Twitter inspires authenticity. Very few people post on the platform to actually get attention. They’re there to have a good time, find some friends, and engage in interesting conversations. It’s people slaving away posting 10 tweets a day with zero likes just to have one go viral that makes the platform special. Twitter is not filled with entirely “high-quality” content (i.e., celebrity tweets; high engagement), but it is filled with interesting, authentic perspectives. It’s quite literally TikTok but for text. Its users skew slightly nerdy, introverted, and tech-savvy. Twitter has a culture, one where a nobody posting an image out of context becomes a viral sensation. Just scroll through Know Your Meme. No social platform — not Bluesky, Mastodon, Reddit, or even Threads — has been able to fully replicate this authenticity.

User minutes or daily active users are terrible metrics to determine if people are having a good time on a website. People will spend a lot of time reading tweets they hate. They’ll engage with content that makes them angry. Nikita Bier, again, is too stupid to understand this. Or perhaps he’s acting maliciously, forcing rage bait down people’s throats to boost engagement using the algorithm he claims he doesn’t work on, yet also says he does. Funny character.

A return to form

I usually write a few columns every night during the Consumer Electronics Show at the beginning of the year. This year is an exception, not out of laziness but because a post-vacation cold hit me like a truck just as the show began. It turns out my body wanted the authentic Las Vegas experience while being situated on the opposite side of the country.

Either way, I have been intently following the developments out of Las Vegas this year. These days, I like to think of CES as less of a serious trade show where all of the biggest consumer technology companies announce their latest products for the year, and more of an attraction for the press. It’s a large theme park, where the theme is artificial intelligence and the rides are exhibits, large and small. People go to CES to network and have fun, not to seriously report on technology that ought to change people’s lives in the coming months.

With that brief prologue complete, here is an unorganized list of some of my favorite “rides” from the CES show floor this year. Some may ship, most won’t, but that’s the thrill of CES.

Here’s Allison Johnson at The Verge on the Clicks Communicator, a dumb phone-esque device aimed at supplementing smartphone use:

Clicks, the company known for its keyboard cases, didn’t just launch a combination MagSafe power bank and slide-out keyboard accessory. It also launched a whole-ass phone. The Communicator leans hard into Clicks’ BlackBerry DNA, with its full keyboard and Curve-esque design. The prototype units I got to play with weren’t functional, but the keyboard keys worked, and boy did they feel nice. The interchangeable back panels are sleek, and I’m personally campaigning for a fuzzy tennis ball optic yellow option.

The Communicator isn’t necessarily trying to put your main smartphone out of a job. Its creators envision it as a companion to your daily driver, something you leave the house with when typing out an email is going to be your focus more than scrolling through a feed of vertical videos. You know, like when you’re running between meetings at a convention for four days. But you don’t have to use it that way; Clicks cofounder and marketing lead Jeff Gadway told me that they’ve been surprised by the number of people expressing interest in the phone as a primary device. And you know what? Those people might be on to something.

The Communicator runs Android but isn’t a typical Android phone. It’s quite literally a dumb phone that happens to run modern apps, like Telegram or WhatsApp, as opposed to being limited to SMS and normal telephone calls. The custom Android skin displays all of one’s text messages across apps in one “message hub,” with easy ways to filter contacts and reply. The hardware keyboard purports to make typing and keyboard shortcuts easier, and the device includes a camera and headphone jack. It’s a clever little device, albeit for a niche market. That isn’t a bad thing at all — its quirkiness is the feature. I like how it doesn’t even try to replace the smartphones people carry in their pockets anyway — Clicks’ founders have correctly realized people love their phones, but might also be interested in a secondary device strictly for communication.

It’s like if Beeper, the messaging aggregator founded by Eric Migicovsky that landed in the spotlight for bringing iMessage to Android, made a hardware device. I think it’s a neat idea, much like the regular Clicks Keyboard cases, which aim to bring a bit of nostalgia to a boring smartphone landscape. Some products are just good for the soul, and Clicks seems to have a great knack for making delightful little things.

Karissa Bell at Engadget reports on the Switchbot Onero H1, a humanoid smart home robot that the company claims can help with basic household tasks:

CES 2026 isn’t the first year we’ve seen a wave of interesting robots or even useful robots crop up in Las Vegas. But it’s the first year I can remember when there have been so many humanoid and humanoid-like robots performing actually useful tasks. Of those, Switchbot’s Onero H1 has been one of the most intriguing robot helpers I’ve seen on the show floor, especially because the company says that it will actually go on sale later this year (though it won’t come cheap).

Up to now, Chinese company Switchbot has been known for its robot vacuums and smart home devices. Much of that expertise is evident in Onero. The unexpectedly cute robot has a wheeled base that looks similar to the company’s robot vacuums, but is also equipped with a set of articulated arms that can help it perform common household tasks.

I understand but disagree with the premise of humanoid robots. Humans are inherently limited creatures, and we’ve designed a physical world to work alongside our limitations. Our living spaces and tools are designed to be used by our hands, which lack precision, and our bodies, which are large and cannot move freely in all directions. A robot vacuum doesn’t drag around a stick vacuum because that would be inefficient — a regular vacuum cleaner is designed with these human limitations in mind. To that end, arms and legs that emulate humans are destined to be slow and clunky. They just take a lot of work to get right. As Bell notes, the robot took two minutes to grab a piece of cloth and put it in a laundry basket feet away.

I don’t think the people behind this company are anything less than smart. But this is the best their robot could do in an ideal demonstration. They could have hard-coded the task and perfectly placed the clothes in the path of the robot so it would finish in a few seconds. But they did try to prove a point, and that point backfired: Humanoid robots are far from complete, and when they finally are, they’ll be inefficient.

This might sound like it’s in jest, but I’m serious: While browsing the internet, as one would while they’re sick, I stumbled across this startup that installs laundry chutes inside people’s homes, so from any room, they can toss their clothes into the chute for it to be deposited into the laundry room through pipes running in the walls. (If I could find this video, I’d link to it here, but alas, I don’t have it saved.) As preposterous and expensive as that sounds, I think it’s a better solution than this robot that takes two minutes to move a few feet. That doesn’t require “AI,” and it probably needs less engineering time, too.

Examples like robot vacuums and this weird laundry chute thing from the internet prove why humanoid robots will never be the most eloquent way to automate. (Also, it’ll be a miracle if this particular robot ever ships at all.)

Jaime Richards wrote about Samsung’s creaseless foldable display for TechRadar, which is probably a prelude to the foldable iPhone coming later this year:

Samsung just won CES 2026. If you ask me, all the other vendors and tech makers can try again next year, because nothing else at this year’s CES is going to be quite as cool as Samsung’s creaseless folding OLED display.

Rewind: what does any of that mean? To put it simply, Samsung has developed a folding display that forms a completely smooth surface when unfolded. Up until now, folding displays have always had a crease – sometimes deep, sometimes minimal – at the point they fold over. It’s long been seen as a necessary compromise, an acceptable price for the expansive screen space a folding phone offers.

One of the marquee features of the rumored foldable iPhone is that Apple has worked with suppliers for years to ensure the device has no crease in the middle of the display. I agree with this insistence — the crease is distracting and unsightly, cheapening the look and feel of an expensive device. Samsung is rumored to supply this display, so it wouldn’t surprise me if this exact technology is touted during the iPhone event come September. I also find it rather suspicious that Samsung pulled this unit mere hours after unveiling it for the press with no explanation. It’s not like it’s in a real device or anything — but what if it is? My hunch is that Apple caught wind of this panel at CES and told Samsung to stop displaying it.

The display itself looks quite convincing. While it still clearly looks like flexible plastic — Samsung insists it’s not plastic — there is no visible distraction on the display. It looks worthy of being in an Apple product, so long as it’s only looked at from the front and not from the side when it’s closed. It is a thick, unsightly panel, and I think that’ll be one of the foldable iPhone’s biggest marketing snags Apple will have to somehow contend with in the fall.

Andrew Hawkins, The Verge’s transportation editor, has some news about Sony and Honda’s Afeela vehicle, a CES staple:

Sony and Honda’s joint venture, Sony Honda Mobility (SHM), said it will start customer deliveries of the $90,000 Afeela 1 electric vehicle in the US in late 2026. The company also showed off an SUV concept that it said would inform a production model in the US for as early as 2028.

The new Afeela Prototype 2026 looked remarkably similar to the Afeela 1 pre-production unit, with short overhangs, a long wheelbase, and an overall larger footprint. No other details about the vehicle were released. SHM CEO Yasuhide Mizuno called it an “early-stage concept.”

I’ve been reporting on this story since 2022, so I felt I had to circle back this year. My official stance on this has been simple: I’ll believe it when I see it on the road. When cars are about to ship “later this year” — an oft-reported phrase in the tech media landscape — they’re usually spotted on the road in some kind of wrap disguise, being actively tested. Especially self-driving ones, like the Afeela purports to be, since all of this software is new and must be certified before it’s released. The Afeela has never been seen in public outside of CES, and the joint venture company seems to just disappear for 11 months right after the show is over, just to come back again for more attention in January.

This is classic vaporware, a term used to describe products that are unveiled but only exist conceptually. I find it even more galling that Honda and Sony have a new concept sport utility vehicle before their current $90,000 sedan even ships. And the presenters onstage seemingly had no comments about the stage of autonomy the vehicle would have on Day 1. I don’t think it’ll come even remotely close to Tesla’s Full Self Driving, a technology the company has been working on for over a decade. I can’t wait to come back to this topic next year and grade my predictions: this car won’t ship, and even if it does, it won’t drive itself.

I’ll end on a more philosophical note: CES 2026 focused heavily on robotics, a small but noticeable contrast from the last two years’ generative artificial intelligence-filled shows. This is a good thing — while CES 2024 brought ChatGPT to every consumer product and CES 2025 advanced silicon for AI inference and training, CES 2026 took CES back to its roots. CES is a show first, and AI chatbots make business people, not tech people, happy. They’re profit centers inflated by an AI bubble that’s about to burst at any minute. They don’t make for good presentations or flashy demonstrations to the press, and they don’t appear well on camera.

Robots have defined CES for decades because they do appear well on camera. They’re just plain interesting, and they also happen to include “AI,” the buzzword of the decade. Most of these robots won’t succeed, but as I said at the beginning of this article, CES is a showcase, not a serious presentation. Some do ship, however, and one of them is the Roborock Saros Z70, a normal robot vacuum with an arm attached to the top. Here’s Jennifer Pattison Tuohy, The Verge’s smart home editor:

Roborock has added an arm to its latest flagship robot vacuum. And this is no tiny appendage like the one the company debuted on its S8 MaxV Ultra at CES last year; it’s an actual articulating robotic arm. The arm rises from the middle of Roborock’s latest flagship bot — the Saros Z70 — and can extend out to pick up items such as socks and tissues while cleaning your floors. While it’s mildly terrifying and currently extremely slow — I can certainly see the potential in a robot vacuum that can clean up ahead of itself.

Products like these delight me and remind me what CES really is about. I own a Roomba i7+ robot vacuum, made by the about-to-be-defunct iRobot back in 2018, and even though it can navigate to rooms by itself, I still remove large items from the floor before I set the device to begin cleaning. This has just been a habit of robot vacuum owners since their conception. The newer Roborock vacuums have an “AI” vision algorithm that uses the camera at the front of the unit to detect large items and maneuver around them, but that doesn’t address the problem. This arm-fitted robot does.

No, the Saros Z70 won’t be the next hit to take on the robot vacuum market — if not for its impracticality, the nail in the coffin will be the price — but it’s emblematic of a gradual shift to CES normalcy. Perhaps the nostalgic gadget lover in me appreciates that.

Caroline Haskins, writing for Wired:

Elon Musk’s AI chatbot Grok is being used to flood X with thousands of sexualized images of adults and apparent minors wearing minimal clothing. Some of this content appears to not only violate X’s own policies, which prohibit sharing illegal content such as child sexual abuse material (CSAM), but may also violate the guidelines of Apple’s App Store and the Google Play store.

Over the past two years, Apple and Google removed a number of “nudify” and AI image-generation apps after investigations by the BBC and 404 Media found they were being advertised or used to effectively turn ordinary photos into explicit images of women without their consent.

It’s just preposterous to assume that allowing CSAM on the App Store or Google Play Store is a “concession” to the Trump regime in return for favorable regulatory treatment. This isn’t a golden gift display at the Oval Office; in fact, it has nothing to do with the president at all. Any suggestion that Apple and Google are keeping a CSAM generator, Grok, and CSAM publication platform, X, on the App Store and Google Play Store for regulatory benefit is complete nonsense.

I don’t know nor really care about the Google Play Store since Android is known to be a cesspool of pornographic — and probably illegal — apps, many of which can be sideloaded. But I do know that there’s an affair between Apple and X, beginning in 2022 when Musk took over the company, formerly known as Twitter. When Musk complained about the 30 percent In-App Purchase fee on the App Store, Tim Cook, Apple’s chief executive, personally invited Musk to Apple Park to sort things out. The result was that Apple would continue advertising on X despite practically every other major advertiser leaving the platform after Musk’s lax stance on content moderation. It took Musk agreeing with an antisemite for Apple to temporarily halt advertising on X before resuming last year.

My takeaway from this is that the reason CSAM remains on the App Store today is that Apple’s leadership clearly wants X’s business, irrespective of where that business comes from. When people sign up for X Premium to get access to the newly paywalled CSAM generation feature, Apple wants 30 percent of that revenue. The only logical conclusion is that Apple doesn’t care if that revenue comes from child abusers and pedophiles. It’s quid pro quo — Cook promised Musk that X would be welcome on the App Store as long as it keeps paying the 30 percent commission, and it doesn’t matter that Apple advertisements for the new iPhone show up alongside CSAM. Because ultimately, protecting children and punishing rapists is bad for Apple’s bottom line as long as it continues collecting a 30 percent commission from the purchases of those rapists.

Elizabeth Lopatto at The Verge never minces words, but I think her headline calling Cook and Sundar Pichai, Google’s chief executive, “cowards” is insufficient. They aren’t cowards. They’re not doing this because they fear retaliation. It’s child sexual abuse material. What honorable person could possibly retaliate against it? Cook and Pichai keep X and Grok on their app stores because doing business with child rapists is economically profitable for them. They’re not keeping CSAM on their platforms because they’re scared of something — they’re proud that CSAM makes them money. Prove me wrong.

Elizabeth Lopatto, writing for The Verge:

In October, Warner Bros. put itself up for sale, leading to a number of bids. The two we are concerned with are a bid from Netflix and another from two nepo babies: David Ellison and Jared Kushner. David Ellison is the head of Paramount, but most famous for being Larry’s son. Jared Kushner is most famous for being Donald Trump’s son-in-law, though he also got his start in business by taking over his felon father’s firm when Charles was in prison; his firm is involved in the financing.

Netflix won the bidding. Warner Bros. made an agreement to sell most of its business — the studio part — to the streaming giant for $83 billion, including debt. (That figure is a bit more than five times Paramount’s market cap.) Warner Bros. felt that spinning itself into two companies, the Netflix acquisition and the cable networks, gave shareholders a value of $31 to $32 a share, rather than the $30 a share Paramount was offering, according to The Wall Street Journal.

Nonetheless, Paramount announced a hostile bid to take over Warner Bros. for $30 per share, which puts the total at $108.4 billion, including debt. Apparently, Paramount has been after Warner Bros. for the last two years, even before Ellison père et fils entered the picture. But now, it’s backstopped the deal with the Ellison family trust, which includes a war chest of about 1.16 billion Oracle shares. The current offer from Paramount is not the “best and final” — at least, according to The New York Times — so this dumb fight is likely to drag on for a while.

Put anti-consolidation hesitation aside for a bit and take the deal at face value. Netflix is perhaps the only company that (a) is objectively tech-focused, and (b) wouldn’t kill Warner Bros. entirely. This makes it immensely appealing to anyone who cares about art and tech infrastructure. Netflix would let Warner Bros. operate independently, but it would integrate its existing intellectual property into Netflix, a world-class tech platform. Maybe it might raise prices, maybe this is bad for Hollywood, but the merger is sensible. David Zaslav, the Warner Bros. chief executive, is hell-bent on selling the studio because he doesn’t care about art. Netflix does, and it also cares about tech. As much as I’m against the deal overall, this is probably the best outcome for Warner Bros.

But why does Paramount want Warner Bros.? Paramount Skydance, controlled by David Ellison, is not a tech company. It obviously has ties to the elder Ellison’s tech company, Oracle, but it’s not like Oracle is particularly a leader in technology infrastructure either. The big deal in enterprise software in 2025 is cloud computing and artificial inference, and Oracle Cloud just isn’t a competent player in the space. People know Netflix, they generally like the service, and they know it’s a tech company. The same goes for Amazon Web Services, Google Cloud, and Microsoft Azure — these companies have proven track records of success in making technical advancements good for consumers. They’re also way ahead of Oracle Cloud. Oracle, at least to me, is stuck in 2019, making (important) database software and really not much else. David Ellison isn’t much of a technical or business mastermind, either: He dropped out of the University of California, Los Angeles, to fund a flop movie.

So Paramount Skydance’s leadership lacks technical prowess, business savviness, and artistic taste. How are these companies, Warner Bros. and Paramount, aligned? It’s becoming increasingly clear to me that the Ellisons run a political enterprise centered around vacuuming as much money out of the federal government during the Trump administration. Larry Ellison’s wealth and notoriety come from Silicon Valley, but his interests nowadays lie in Washington. He, much like Elon Musk, has made it his life’s work to shift public opinion to benefit Republicans, which in turn would lower his tax bill and reduce regulatory burden on his business. It’s a very low-energy attempt at forming a kleptocracy. But this strategy makes it clear that, much like Musk, Larry Ellison is not a tech person. He’s a politics-adjacent figure, and the younger Ellison’s desperate bid for Warner Bros. is yet another attempt to advance those political goals.

Paramount owns CBS News, and it’s become evident in recent weeks that the Skydance merger really happened just to get CBS under the Ellisons’ control. Paramount installed Bari Weiss, a conservative editor, as editor in chief of CBS to pivot the publication into right-wing news territory. These days, the CBS News website is dominated by a gargantuan blue banner linking to The Free Press, a conservative opinion website run by Weiss, with pieces like “How to Win a Pardon from Trump” and “Israel Had a Disorienting 2025” appearing prominently. The “This Week in Politics” section has nothing about President Trump’s illegal torture camp in El Salvador, CECOT — the “60 Minutes” segment that was supposed to cover CECOT was pulled last minute by Weiss — or any other important news story. It instead leads with the headline, “Karoline Leavitt announces she is pregnant with her second child.” Seems like nationally significant news.

The Ellisons’ handling of CBS News gleans some insight into why they’re now desperate to own Warner Bros.: to hijack CNN next. David Ellison is not a tech person, and neither is his father, but they’re both political hacks, and CNN is their next target. As Lopatto notes, Paramount quite literally doesn’t have enough money to finance a Warner Bros. merger, but the Ellisons are willing to do whatever it takes to control yet another American media company. This batch of Silicon Valley entrepreneurs is not a group of “tech bros” — they merely use their notoriety in the tech space to pivot to politics. They’re government lobbyists, not tech people. Understanding this distinction between “tech bros” (Mark Zuckerberg, Sundar Pichai, Jeff Bezos, Tim Cook, etc.) and political spenders (the Ellisons, Musk, David Sacks, etc.) is crucial to getting an idea of American politics. There are really two camps hiding under the same veil, and one is far more sinister. The former sees the administration as an obstacle to progress, the other sees it as an opportunity.

Tim Bradshaw, Stephen Morris, and Michael Acton, reporting for the Financial Times in November:

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.

It has taken me over a month to link to this article because I thought I largely didn’t have anything to say about it other than “finally.” And in one way, that was my initial reaction to this reporting. (I’ll choose to ignore Mark Gurman’s separate report at Bloomberg claiming this entire report is false; the Financial Times is a reputable newspaper.¹)

John Ternus, Apple’s senior vice president of hardware engineering, is rumored to be the most likely successor after Cook inevitably leaves, whether that be in 2026 or anytime after. But my biggest question is how much independence Ternus would have as chief executive. I would bet money that no matter when Cook leaves, he won’t truly abandon the company to retreat to a private island. Cook isn’t a normal billionaire, and I don’t think he has much of a life outside Apple. His primary work has been transforming Apple into the successful, multi-trillion-dollar corporation it is today. He isn’t a product person, nor does he have any technical experience — he just manages the books, and I think he wants to stay in that role forever.

To that end, I’d be legitimately shocked if he didn’t become the chairman of the board of directors, a position currently occupied by Dr. Arthur Levinson, who is 75 years old. As noted by Joe Rossignol at MacRumors, Apple’s corporate guidelines say that a member of the board cannot stand for re-election after they turn 75, effectively forcing Dr. Levinson to step down at Apple’s next shareholder meeting in 2026, when shareholders elect the next chairman. This gives way to two possibilities: (a) someone else, not Cook, takes the position of the chairman and remains in that position for the foreseeable future, or (b) Cook, who is 65, assumes the role next year, stepping down as chief executive. The latter option seems immensely more likely to me, knowing Cook’s personality and role at Apple.

Many commentators in the space have said they believe a primary motivation for Cook choosing not to retire is to appease the Trump administration, a stance I generally agree with. But this gives me pause: As chairman of the board, Cook would assume a more outward-facing role than Levinson, which nobody but the most involved of commentators is familiar with. Cook would attend dinners at the White House to honor murderers, encourage anti-First Amendment action to protect kidnappers, and engage in other unscrupulous activities that have certainly contributed to Apple’s besmirched public image. But he’d leave the day-to-day, more technical operations to Ternus, who I assume is less familiar with the aimless politics Apple engages in these days.

That’s a perfectly reasonable plan, and I think it’d fly by Apple’s shareholders easily — it might even nudge Apple’s stock up, knowing how behind the company is in the technical space — but it lessens my “finally” sigh of relief I signaled at the beginning of this post. Ternus would undoubtedly bring change to Apple’s core products. His leadership has transformed the Mac lineup from an aging, decaying product line that did immense damage to Apple’s reputation. Today, the Mac is a fantastic, class-leading line of personal computers, a claim I couldn’t make earnestly six years ago. Apple’s hardware is easily the company’s best, most competitive, and most meaningful product(s), and I trust Ternus would bring that same care to the rest of Apple’s offerings, especially software.

But this restructuring wouldn’t address one of my biggest complaints with Apple’s core leadership: the company has lost its soul. Its environmental work is effectively on hiatus to support the kidnapping and deportation of American citizens. The features Apple builds into its products to protect journalists and activists are undermined by Cook’s wining and dining with Mohammed bin Salman, the ruler of Saudi Arabia, who murdered a journalist. Ultimately, Apple is at a point in its history where it must make tough decisions, and with Cook — who would no longer be involved in day-to-day operations the second he becomes the chairman of the board — shadowing over Ternus, he’d be unable to make those decisions independently.

When I wrote about Cook’s fecklessness in August and even earlier before, I said he had to go. That means that Apple has to get rid of him for good.

From Senator Dick Durbin, Democrat of Illinois:

U.S. Senate Democratic Whip Dick Durbin (D-IL), Ranking Member of the Senate Judiciary Committee, and U.S. Senator Lindsey Graham (R-SC) today introduced the Sunset Section 230 Act, which would repeal Section 230 two years after the date of enactment so that those harmed online can bring legal action against companies and finally hold them accountable for the harms that occur on their platforms.

“Children are being exploited and abused because Big Tech consistently prioritizes profits over people. Enough is enough. Sunsetting Section 230 will force Big Tech to come to the table take ownership over the harms it has wrought. And if Big Tech doesn’t, this bill will open the courtroom to victims of its platforms. Parents have been begging Congress to step in, and it’s time we do so. I’m proud to partner with Senator Graham on this effort, and we will push for it to become law,” said Durbin.

One of my favorite words of 2025 has been “slopulism,” a portmanteau of “slop” and “populism.” I don’t think all American populist movements are slopulism (Mayor-elect Zohran Mamdani, Democrat of New York, is an example of a good populist campaign), but many of them are. Slopulism particularly manifests itself in anti-tech, anti-artificial intelligence sentiment not rooted in fact. It’s a truth that generative artificial intelligence was trained without the permission of writers and artists. It is false to say generative AI is some kind of great catastrophe to the environment. Inference is remarkably cheap and efficient, and scientists are working to make pre-training more sustainable every day. It is true to say AI data centers do not contribute to local economies; it is false to assert they’re useless. Populism versus slopulism.

To that end, Durbin and Graham’s legislative joke is entirely slopulism. Section 230 of the Communications Act of 1934 gives platforms legal immunity over what their users say on those platforms. For instance, if a person encourages someone to commit suicide on X or Instagram, the deceased person’s family cannot sue the platforms for any wrongdoing. They can sue the other person, but the platforms are shielded. It is a hard, important line between user speech and company speech. Per the First Amendment, it is legal to say anything on the internet, and Section 230 maintains that right by giving platforms the liberty to moderate speech however they want.

Some platforms, like 4chan, the anonymous image board, refuse to do any meaningful content moderation unless the speech is explicitly illegal, e.g., child sexual abuse material. Other platforms, like Snapchat or Discord, engage in more active content moderation. But the commonality between all of these platforms is that these moderation decisions belong to the platforms themselves. They’re legally protected from most civil lawsuits, allowing a high degree of free speech on the internet. (And yes, contrary to people like Elon Musk, the internet is predominantly a free place.) This is all thanks to Section 230.

If Section 230 is removed, anyone — whether malicious or well-meaning — could sue platforms for their content moderation decisions. This is highly unprecedented and would result in a major crackdown on free speech on the American internet. Platforms would begin heavily censoring user-generated content in an attempt to prevent lawsuits, to the point of employing automated systems to instantly remove a person’s account if they’re deemed even slightly risky to the platform. Overnight, all users would become the platforms’ legal liability.

Platforms must be given some immunity against accountability because a poor moderation decision shouldn’t be punished like a crime. It would be like punishing a gun company for every single gun-involved homicide in America. As much as I don’t like the firearm lobby, that’s complete lunacy. It goes against the very core of the First Amendment. It should not be illegal to run communication platforms in the United States, and if it is, those platforms will no longer be used for any intellectual debate or legally murky conversations. Suing companies is trivial in the United States, yet communication platforms on the internet have been shielded from this lawfare to promote freedom of speech. Only a tech-illiterate person would risk that sanctity the internet has historically enjoyed.

John Keilman, reporting for The Wall Street Journal:

The company that makes Roomba robotic vacuums declared bankruptcy Sunday but said its devices will continue to function normally while the company restructures.

Massachusetts-based iRobot has struggled financially for years, beset by foreign competition that made cheaper and, in the opinion of some buyers, technologically superior autonomous vacuums. When a proposed sale to Amazon.com fell through in 2024 because of regulatory concerns, the company’s share price plummeted.

It owes $352 million to Picea, its primary contract manufacturer which operates out of China and Vietnam. Nearly $91 million of that debt is past due, according to iRobot.

Outlining its restructuring plan Sunday, iRobot said that Picea will receive 100% of the equity interest in iRobot, which the company said would allow it to continue operating.

The Federal Trade Commission, headed by Lina Khan, the former FTC commissioner, effectively blocked Amazon’s acquisition of iRobot in 2022. While I’ve expressed that I was generally a fan of Khan’s leadership, her stance toward acquisitions missed the mark. The idea behind blocking this acquisition was to protect consumers, but that was nonsensical, knowing (a) Amazon had no prior business in the robot vacuum market, and (b) iRobot was spiraling toward bankruptcy already thanks to increased competition from Chinese manufacturers. If the government didn’t actively help iRobot, it would have only been a matter of time before it finally kicked the bucket, which is exactly what happened on Sunday.

But the real icing on the cake for the Biden administration is that iRobot sold itself to a Chinese company, the very enemy it sought to destroy. That embodies the very essence of the Biden administration: trying to fix a problem and ending up making it catastrophically worse. For the record, I don’t see a significant problem with the leading robot vacuum maker being Chinese. Roborock makes great products — way better than iRobot — and it has found success in the market. It’s competing fair and square. But from the U.S. government’s perspective, letting the country’s most powerful enemy usurp a growing market for an entirely self-inflicted reason is just embarrassing.

The FTC did drag Meta into court for effectively the same reason (illegal acquisitions), but it did such a terrible job of proving Meta illegally acquired its monopoly — which it definitely did — that it lost the case under the Trump administration earlier this year. That proves sheer incompetence in the FTC: that it’s spending too much time on cases that don’t matter and not enough on the ones that do. The Biden administration had no survival instinct — at some points, it was too reactionary, and in others, it wasn’t reactionary enough. The result is a once-great company falling to a foreign competitor because the U.S. government sealed its fate years ago. I pity iRobot.

Andrew J. Hawkins, reporting for The Verge on Thursday:

At an “AI and Autonomy” event at the company’s office in Silicon Valley on Thursday, Rivian unveiled its own proprietary silicon chip, as well as a number of forthcoming autonomous features that it says will enable it to eventually sell Level 4 autonomous vehicles to customers. That includes equipping the company’s upcoming R2 vehicles with lidar sensors.

Rivian also said it will launch a new AI-powered voice assistant as well as a foundational “Large Driving Model” trained similarly to large language models like OpenAI’s ChatGPT that will “distill superior driving strategies from massive datasets into the vehicle.” And it said it would wrap everything up in an Autonomy Plus subscription service for a new potentially lucrative revenue stream for the company…

It’s safe to say Rivian’s R1-series of vehicles is much better than any Tesla on the market for the price. While expensive, Rivian sport utility vehicles and pickup trucks feel like luxury cars with beautiful interiors, all with comparable specifications to Tesla’s Model X and Cybertruck. But Tesla still has Rivian beat in the software department because of the company’s famed yet highly controversial Autopilot suite of driver assistance features. Autopilot’s traffic-aware cruise control and auto steer work on practically every clearly-marked road in the United States and abroad and are class-leading. While many companies compete with Tesla in the electric vehicle market, including Rivian, none of their autonomous driving systems come close to Autopilot’s versatility and reliability. Tesla has offered and perfected Autopilot for over a decade, so it’s no surprise. (Rivian’s current system only works on select U.S. highways.)

Newer Tesla models can be equipped with a more advanced package of features called Full Self-Driving, which enables supervised point-to-point autonomous navigation, meaning the car will do everything from pulling out of a parking space to changing lanes to finding a parking spot at the destination. Full Self-Driving has improved considerably since its launch — which has suffered numerous embarrassing “delays” (the feature was never ready, contrary to Elon Musk, the company’s chief executive) — yet it still makes concerning mistakes and must be intently supervised at all times. Despite Full Self-Driving’s fundamental design flaws, it is a significant step ahead of legacy vehicle manufacturers like Mercedes-Benz and BMW, with whom Tesla directly competes, as well as Rivian, which, until Thursday, had no plan to implement a similar feature. (It’s worth noting Alphabet’s Waymo cars work unsupervised reliably, but in limited cities.)

Rivian’s Thursday announcements came in three parts: silicon, hardware, and software. Beginning with silicon:

The centerpiece of this new effort is the tiny chip with a 5 nanometer process node called the Rivian Autonomy Processor. Taiwan’s TSMC will produce the chip for Rivian. The company says that it “integrates processing and memory onto a single multi-chip module,” and is being used to power the company’s third generation computer. Rivian says the chip’s architecture will deliver “advanced levels of efficiency, performance, and Automotive Safety Integrity Level compliance,” referencing a risk classification system for safety-critical automotive electronics…

Tesla’s custom Samsung-fabricated silicon is class-leading and more or less enables Full Self-Driving’s dominance in the field. Taiwan Semiconductor Manufacturing Company is picking up the slack for Rivian, and that matters considerably. Current Rivian models don’t have silicon powerful enough for a feature similar to Full Self-Driving, and by including the Rivian Autonomy Processor in R2 models beginning next year, Rivian is strategically readying itself for better software. Next, hardware:

Rivian will use a variety of sensors to power its autonomous driving, including lidar. The company plans on integrating lidar into its upcoming R2 vehicles to help with redundancy and improved real-time driving. Waymo and other robotaxis use lidar to create 3D maps of their environment, while Tesla famously does not. Some automakers have said they would use lidar in future production vehicles, but that turned out to be easier said than done. Volvo, for example, recently dropped lidar for its EX90 SUV.

As Hawkins writes, Waymo’s success can largely be attributed to the massive lidar sensor array that sits atop every Waymo vehicle on the street. Similarly, while this might be controversial, I blame the Tesla Robotaxi and FSD’s failures mostly on Tesla’s insistence on cameras to power its software. Tesla calls its camera array Tesla Vision, and the result of this system is that Tesla vehicles make considerably more mistakes on the road than their lidar-powered Waymo counterparts. They’re so bad that while Waymo works without driver supervision in Austin and San Francisco, the Tesla Robotaxi still has a Tesla employee in the driver’s seat ready to take over in case FSD makes a fatal mistake. To that end, I’m grateful Rivian has gone with lidar for its hardware as opposed to a Tesla-like vision-only approach. Finally, software:

Rivian also outlined a series of advanced features coming to its cars in the future, including hands-free driver assist, also known as Level 2 Plus, and eyes-off driving, also known as Level 3. Early next year, Rivian plans on rolling out hands-free driving for its second-generation R1 vehicles that will function on 3.5 million miles of roads across the US and Canada — a big leap over the 135,000 miles it covered earlier this year. And the feature will be available on more than just highways. In a video, Rivian demonstrated its hands-free system on a variety of roads, including across the Golden Gate Bridge, up the steep hills of San Francisco, and along the Pacific Coast Highway.

From my understanding, the “Level 2+” autonomy level coming to R1 cars next year is roughly analogous to Tesla Autopilot from six years ago, whereas the Level 3 system is more equivalent to FSD. As I said earlier, Rivian is undoubtedly many years behind Tesla — and even more years behind Waymo — but Thursday’s announcements are the first steps toward catching up. The main thing Rivian drivers miss when coming from a Tesla is Autopilot, and the new system should aim to close that gap. FSD is still in its infancy, and I don’t blame Rivian for wanting to take its time with it. Lidar should presumably speed the process up — it’s just easier to work with than vision-only models — but for now, the Level 3 system is in the distant future. I think Rivian owners will be patient enough to keep waiting, especially if all R2 and R3 models ship with the hardware necessary from the factory, but I must call that eventual reality vaporware for now.

I’m highly bullish on self-driving cars: they’re safer, better drivers than humans and alleviate a major stress point, especially for Americans. I just want more competition in the currently heavily-dominated-by-Waymo-and-Tesla market.

All’s well that ends well, isn’t it?

A quote from Steve Jobs, presented at Apple’s September 2025 event. Image: Apple.

Bloomberg on Wednesday reported that Alan Dye, Apple’s head of user interface design for over a decade, would depart to work at Meta Reality Labs. The news was confirmed by Tim Cook, the company’s chief executive, who said Stephen Lamay, a longtime designer for the company, would assume Dye’s role. Bloomberg’s report was heavy on the “Meta rules, Apple drools” narrative¹, but if anything, that’s a reflection of what Dye has done to Apple during his tenure. The average IQ of both companies has increased. (Thanks to Twittgenstein on X for this wonderful adaptation of this quote.)

It’s safe to say I am not a fan of Dye’s work. I, much like Jason Snell at Six Colors, refrain from excessive personal attacks in my writing, but this is one of the few exceptions. Dye has overseen many inventive projects at Apple, namely iPhone X’s gesture-based navigation system and iPhone 14 Pro’s Dynamic Island, both designs that I have commended on numerous occasions. But that concludes the list of thoughtfully designed interfaces Dye has produced.

macOS 11 Big Sur was a clear design regression from previous versions of macOS. It obscured clarity to “make more room for content,” an adage Dye has used ad nauseam to the point where it has become meaningless filler. It hid nearly all important context behind some action, whether it was moving the cursor over a segmented control for more context, swiping to reveal more actions in a menu, or increasing button spacing so much that it became a nuisance to use the mouse. macOS Big Sur objectively made the Mac a worse computing platform, even if it brought feature and design parity across Apple’s operating systems. Apple, under Dye’s leadership, has failed to comprehend the sanctity of the Mac operating system: that it is fundamentally different from iOS and must be treated with a different level of precision.

Under Dye’s design leadership, the Mac has shipped with lowest-common-denominator apps pulled from iOS and transplanted onto a larger screen. My favorite example is Home, which is perhaps one of the worst first-party apps ever designed on any version of macOS. It is hilariously pitiful, so much so that it lacks support for even basic keyboard shortcuts and requires dragging with the mouse to change the brightness level of a light bulb. Another regression applies to the Home app on all platforms: tapping a device tile navigates to the detail view for the device, but tapping the icon switches the device on and off. This is not visually indicated anywhere, and it isn’t even consistent across device types.

This abject laziness and incompetence isn’t limited to the Home app. The double-tap-to-invoke-Siri gesture introduced in iOS 18 is so prone to accidental triggers that any reasonable person would conclude it was simply never tested by Apple designers. The Safari redesign in iOS 15 and macOS 12 Monterey was perhaps some of the most embarrassing design work from Cupertino since the infamous 2013 Mac Pro. In iPadOS and macOS, it was so hard to see which tab was selected that someone made a Safari extension to mark the selected tab with a colored sliver just so it was legible during the beta process. And the iOS version hid the refresh button underneath a context menu for over half the beta period until the uproar online was so loud that Apple was forced to change it.

And none of this considers Liquid Glass, which is so unbearable in some places that I had to write an article documenting everything that was wrong with it. Legacy macOS app icons are now destroyed by a gruesome gray border, encapsulating them in a rounded rectangle. The gorgeous tool-inspired designs that once made the Mac whimsical and fun are now extinct, replaced by nondescript, mundane app icons that don’t even pay homage to the original versions. Liquid Glass is still unreadable in many places, and Apple knows this: If a notification appears on the Lock Screen, iOS dims the Lock Screen wallpaper so the text is legible. And Dye’s solution to this conundrum was not to go back to the drawing board and rethink the Liquid Glass material, but to add a truly hideous Tinted option that looks like Google designed it.

Liquid Glass is completely nonsensical on the Mac. It nonsensically mirrors content in sidebars, it nonsensically moves elements to the bottom of the screen like in Music, and it nonsensically changes the corner radii of windows depending on their navigation structure. Every year, Dye rounds window corners even further despite the fact that no Macs ship with truly rounded corners. Why must windows be rounded this severely, and why is every single window’s radius different across the system? There is no consistency, no taste, no respect for the craftsmanship of the operating system. macOS has lost every ounce of class it once had, and it has been whittled down to a disorienting mess of iOS-like controls mixed with designs that feel like they’ve taken inspiration from Windows Vista.

Alan Dye is objectively horrible at his job, and it is a great boon to Apple that his tenure is over.

As Steve Jobs said, “Design is not just what it looks and feels like. Design is how it works.” Dye loves this quote so much that it was prominently featured at the beginning of the September iPhone event, and he has no right to love it. Dye is a Jobs cosplayer, not a protégé. He takes ideas from Apple’s post-Jobs Jony Ive era and applies them in all the wrong places. It’s like handing a wild animal a machete — he has all the ability to design some of the world’s most used and beloved operating systems, and none of the talent.² Jobs and Ive, Apple’s former chief designer, were such a great duo because they complemented each other so well. Ive would have these outlandish design ideas, and Jobs would rein them in. Jobs knew how to make good technology, and Ive knew how to make it beautiful.

Apple lacks technical leadership in the C-suite. Cook probably couldn’t figure out how to exit Vim to save his life. The situation in Cupertino is so bad that Luca Maestri, Apple’s former chief financial officer, not only had the power but the final say in rejecting a technical team’s request for graphics processing units to train artificial intelligence models. Not the leader of the company, not a member of the technical staff — the leader of the accounting department with a degree in economics. Therefore, it comes as no surprise that Dye’s shenanigans went completely unchecked. Craig Federighi, Apple’s software engineering chief, had the power and qualifications to put Dye in check, but he simply failed. Federighi’s failure to oversee software design (Dye) and engineering (John Giannandrea, Apple’s head of machine learning, who just recently announced his retirement) will go down as one of the most catastrophic missteps in Apple’s recent history.

As John Gruber writes at Daring Fireball, Dye has no technical experience, not even in designing user interfaces or computers. He’s a fashion executive who worked for Kate Spade, a clothing design brand. Someone like that either needs a technical supervisor (Federighi or a Jobs-like figure) or must be relegated to a lower-level position working on design prototypes. It is galling for Apple that he was appointed to such a prestigious role, and the steeply declining quality of Apple software, especially in this decade, is proof that he doesn’t fit there.

As for Stephen Lamay, Dye’s replacement, I have no idea who he is. Maybe he’s a good designer, maybe he isn’t. But he does have technical expertise, something sorely needed at higher levels of Apple leadership. Ben Hylak, a former Apple designer who now works for an AI startup, says Lamay is “by far the best designer I have ever met or worked with in my entire life” — high praise from someone who worked under Dye and can now speak candidly. And if Gruber’s sources are to be believed, Lamay is universally liked internally. These are good indicators of competency: Apple employees, on many occasions, have criticized the leadership of Giannandrea and Robby Walker, a leader of the Siri team under Giannandrea. Both figures have left the company since. When an executive is despised internally, it isn’t a good sign.

The rumored “Snow Leopard”-style bug-fix update coming next year will be a true test of Lamay’s leadership. With technical guidance, he must bring Apple’s operating systems together, which are currently in an unstable state. They feel like a mélange of poorly integrated user interface concepts — especially macOS, whose design is unbearable to look at and use. Time will tell if Lamay is taking the pastiche route to Dye’s leadership, or if he throws all of this in the trash and works to restore Apple software to its former glory.

It doesn’t surprise me in the slightest that Dye has chosen Meta as his next employer. (And yes, I’m certain Dye chose Meta, not the other way around.) The two go together like a moth to a flame. Mark Zuckerberg, Meta’s chief executive, has been poaching top talent from his major Silicon Valley competitors since the beginning of this year, even offering pay packages of up to $100 million. Meta is short on talent, in both senses of the word: it does not have the inherent aptitude (via company culture) to make anything wonderful, nor the people interested in accomplishing anything spectacular. People work at Apple not for extravagant bonuses or work-life balance, but because they truly believe in the company’s mission. It’s unique in that sense. People who work at Meta only go there because they pay $100 million — the same goes for Dye.

Dye never truly believed in the Apple philosophy of design. I don’t mean this in the “Severance”-esque “there’s more to work than life” way, but that he doesn’t understand what makes Apple special. On Wednesday, as the news of his departure hit the internet, he posted to his Instagram story a quote from Jobs telling people to “not dwell on” a job for too long if you “do something and it turns out pretty good.” That is truly in dismal taste, almost like one last middle finger pointed toward Apple’s worship and respect for Jobs’ work. Nobody at Meta believes a thing Jobs said, but Apple employees — the ones Dye is leaving behind in Cupertino — certainly do, and using a Jobs quote in this way distorts the true meaning of Apple’s design work.

Meta Reality Labs, these days, makes AI wearable products, but it wasn’t too long ago that it was peddling the Fisher-Price metaverse. Zuckerberg was so bullish on the metaverse that he even renamed his company after it. What was once Oculus — talented makers of the finest virtual reality products — turned into Reality Labs, a division that is mostly focused on bringing Meta AI into the real world. Meta AI, however, is comically worthless. Llama, the company’s flagship large language model, does extremely poorly on all benchmarks compared to its competitors, and the technology is mainly used by elderly people on Facebook to reply to posts and share truly atrocious AI-generated videos. Alexandr Wang, Meta’s head of AI, whom it spent $14 billion hiring, truly nailed the Meta AI coffin shut.

Dye’s new job at Meta reminds me of Zuckerberg’s hiring of Wang, who has contributed virtually nothing meaningful to the company. Wang’s role at Meta is presumably quite significant: He’s the head of the Meta Superintelligence Labs, a sister division to Meta Reality Labs. (In Meta parlance, a “lab” is a division that specializes in developing new technology before handing it over to a consumer product team, like Instagram, for final implementation.) Dye’s role as chief designer for Meta Reality Labs would be roughly analogous to Wang’s, since design is one of the most difficult problems VR and augmented reality devices face. All of this leads me to believe Dye won’t be successful in this new role, despite the power he is given, similar to Wang.

Meta is a disoriented company without product taste or clear direction. It goes along with the market. Many of Apple’s biggest supporters who read my work have been quick to point out to me that Apple hasn’t suffered materially due to its lack of a successful AI product because its core hardware has been successful, and they’re right. The same goes for OpenAI, which pioneered the AI boom in 2022 with the launch of ChatGPT and has its eyes set on total domination in that field. Google has chipped away slowly at AI for the better part of two decades, and it too has found success there. All three of Meta’s most important competitors have decided on a path forward. Meta hasn’t converged on a field yet, and it never will — it started with social networking, then moved to the metaverse because it missed out on owning a major mobile platform, then haphazardly shifted to AI once it became apparent that it would be profitable. This isn’t a successful business strategy.

Zuckerberg is not entirely incompetent, but he’s certainly confused, and the sentiment within Meta is that everyone else is, too. They’re certainly proud of their work, but they never know what’s next, and there’s never a hint of altruism anywhere. Meta’s hiring strategy stems from a lack of direction, and Dye’s appointment is the latest example in Meta’s corporate messiness.

Allison Johnson, reporting for The Verge:

Samsung is officially announcing the Z TriFold, its much-anticipated foldable with not one, but two hinges. It’ll launch first in South Korea on December 12th, with a US launch planned for the first quarter of 2026. There’s no US price just yet, but it’ll cost KRW 3,590,400 (about $2,500) for 512GB of storage when it launches back home, so you should probably start saving your pennies nickels for this one.

The TriFold’s inner screen measures 10 inches on the diagonal, with a 2160 x 1584 resolution and a 120Hz adaptive refresh rate that goes all the way down to 1Hz. That’s a lot of screen. You can run three apps vertically side by side on it, and even use Samsung’s DeX desktop environment in a standalone mode without a separate display. On paper, the TriFold’s outer screen looks a lot like the one on the Z Fold 7. It’s a 6.5-inch 1080p display with a 21:9 aspect ratio.

I’m generally a purveyor of foldable phones — especially ones that open up like a book, as opposed to the ones that flip open — but I really don’t know how to feel about the Z TriFold. Terrible name aside, the device is essentially a tablet computer, but with none of the benefits of the standard Galaxy Z Fold. This device is too thick at 12.9 millimeters for any practical use and is too large to use out and about as a phone. It’s really more of a “tablet that folds for storage,” and I’m not sure how compelling a use case that is. I think it’s a gimmick.

With rumors of a foldable iPhone in full swing for next year, I’ve been thinking about the potential use cases for foldable iPhones. Up until recently, I’ve thought of the iPad as a unique but quirky device that doesn’t fit nicely in the age of foldable devices, but this year’s update to iPadOS has made me reconsider that view. The foldable iPhone will still run iOS, and thus, will carry all of its limitations. By contrast, iPadOS is now a capable — albeit still undoubtedly hamstrung — operating system, and it positions the iPad as more of a secondary Mac than just a larger iPhone.

So where does the foldable iPhone fit into all of this? For those who mainly use the iPad as a content consumption device, I can see a two-in-one device being a great fit. The same might apply to students if the foldable iPhone supports the Apple Pencil, but the Z TriFold doesn’t support Samsung’s S Pen stylus, so I’m doubtful Apple Pencil support is on the table. That covers most but not all iPad use cases, but at the rumored price of about $2,000, many customers will just opt for the standard iPhone and a much cheaper iPad anyway.

Again, I still like the idea of foldable smartphones. I think they have great practicality — the foldable iPhone would cover most iPad use cases! — but they’re just too expensive and don’t offer any new functionality that two cheaper devices couldn’t accomplish. That problem applies to both Samsung’s Z TriFold and the rumored foldable iPhone. I’m certainly bullish on some kind of foldable form factor being the future of mobile computing, but that time certainly isn’t now, and Samsung’s latest confusing contraption is more proof of that.

Anthropic, just before Thanksgiving:

Our newest model, Claude Opus 4.5, is available today. It’s intelligent, efficient, and the best model in the world for coding, agents, and computer use. It’s also meaningfully better at everyday tasks like deep research and working with slides and spreadsheets. Opus 4.5 is a step forward in what AI systems can do, and a preview of larger changes to how work gets done.

Claude Opus 4.5 is state-of-the-art on tests of real-world software engineering… As our Anthropic colleagues tested the model before release, we heard remarkably consistent feedback. Testers noted that Claude Opus 4.5 handles ambiguity and reasons about tradeoffs without hand-holding. They told us that, when pointed at a complex, multi-system bug, Opus 4.5 figures out the fix. They said that tasks that were near-impossible for Sonnet 4.5 just a few weeks ago are now within reach. Overall, our testers told us that Opus 4.5 just “gets it.”

Claude Opus 4.5 is easily the best large language model on the market for myriad reasons: It doesn’t speak corporate English, it’s phenomenal at programming, and it’s excellent at explanations. It is the only model that speaks remotely like a human being, and it is the only model that can write safe, efficient, and uncomplicated code. It appears my assertion that Gemini 3 Pro would be the “smartest model for the next 10 weeks” was a bit ambitious. But none of this surprises me: Anthropic’s models have a certain quality to them that makes them feel so nice to interact with. They’re candid, don’t try to be too clever, and push back when needed. I can’t believe there was a time when I discounted Anthropic’s competence.

Claude Opus 4.5 is, by all of the benchmarks, the smartest model for coding. LMArena, a website that asks people to blindly rank model responses, has it at No. 1 on the web development leaderboard, and it excels in all the benchmarks that Gemini 3 Pro previously owned just earlier in November. But I wouldn’t say its coding performance is that much better quantitatively than Claude 4.5 Sonnet or any of its competitors. If one gives a question to GPT 5.1 Codex Max, Gemini 3 Pro, Claude 4.5 Sonnet, and Claude 4.5 Opus, they’ll all conjure up more or less the same solution. The difference comes in how that solution is presented: Gemini is more verbose and messy, GPT-5.1 is terse and overcomplicates implementations, and Claude strikes a balance. Simon Willison describes this phenomenon well:

It’s clearly an excellent new model, but I did run into a catch. My preview expired at 8pm on Sunday when I still had a few remaining issues in the milestone for the alpha. I switched back to Claude Sonnet 4.5 and… kept on working at the same pace I’d been achieving with the new model.

With hindsight, production coding like this is a less effective way of evaluating the strengths of a new model than I had expected.

I’m not saying the new model isn’t an improvement on Sonnet 4.5—but I can’t say with confidence that the challenges I posed it were able to identify a meaningful difference in capabilities between the two.

This represents a growing problem for me. My favorite moments in AI are when a new model gives me the ability to do something that simply wasn’t possible before. In the past these have felt a lot more obvious, but today it’s often very difficult to find concrete examples that differentiate the new generation of models from their predecessors.

I agree with Willison and think this is an astute observation. I haven’t been able to try out Claude 4.5 Opus in Claude Code — my preferred way of using artificial intelligence to write code, since it lets me abandon the hell of Visual Studio Code — because I only subscribe to Anthropic’s Claude Pro plan, not Claude Max. I’m yet to encounter a problem Claude Sonnet 4.5 couldn’t solve. Sometimes it has required extra guidance and a bit of backtracking or examples, but it has always gotten the job done. Perhaps Claude Opus 4.5 would format those responses better so I wouldn’t have to do any manual refactoring after the fact, or maybe it could accomplish the same thing with a less detailed prompt. But these aren’t reasons for me to spend five times the money on an AI chatbot.

Again, I maintain Anthropic’s models are the best on the market, just empirically. Whatever Anthropic’s engineers are up to, they’re amazing at post-training LLMs. Claude’s personality is best in class, its code is remarkably professional, and the models follow instructions well. OpenAI’s models are trained to be great consumer-grade busywork assistants. When you ask for feedback on writing, GPT-5.1 will just rewrite the text using the most insufferable corporate tone anyone has ever heard. “I really hear you — and I can help,” it emphasizes. Gemini will do the same more emphatically but with uninspiring diction. Claude does not rewrite; it tells you what is wrong with what you wrote. I don’t use LLMs for writing advice because I can confidently say I’m a better writer than any of these robots, but this is a common benchmark I use to test model personality.

Anthropic is, to put it lightheartedly, just built differently. Of course Claude 4.5 Opus is the best model on the market — Anthropic is the only AI lab left with taste.

Allison Johnson, reporting for The Verge:

Google just announced some unexpected and welcome news: Pixel 10 owners can now send and receive files with Apple devices over AirDrop. And equally interestingly, the company engineered this interoperability without Apple’s involvement. Google says it works with iPhone, iPad, and macOS devices, and applies to the entire Pixel 10 series. While limited to Google’s latest phones for now, Google spokesperson Alex Moriconi says, “We’re bringing this new experience to Pixel 10 first before expanding to other devices.”

When we asked Google whether it developed this feature with or without Apple’s involvement, Moriconi confirmed it was not a collab. “We accomplished this through our own implementation,” he tells The Verge. “Our implementation was thoroughly vetted by our own privacy and security teams, and we also engaged a third party security firm to pentest the solution.” Google didn’t exactly answer our question when we asked how the company anticipated Apple responding to the development; Moriconi only says that “…we always welcome collaboration opportunities to address interoperability issues between iOS and Android.”

When the feature was first announced earlier on Thursday, I was in disbelief and wondered how it worked. “Surely this must be some kind of collaboration, right? I was wrong, and Google indeed accomplished this by itself. How it did that is an interesting computer science lesson but irrelevant nonetheless. What is relevant is the striking parallel between this feature and Beeper, a company that reverse-engineered the iMessage protocol in 2023, allowing interoperability between Android and iOS. Beeper used a backdoor in the Apple Push Notification Service, commonly known as APNS, and made its solution available via a subscription. Apple promptly shut it down, but took no legal action. The resulting ordeal was a drawn-out cat-and-mouse game in the spotlight, with every technology blogger, including yours truly, having something to say about it. (As a writer, I enjoyed it, but eventually sided with Apple in the end.)

The Beeper Mini situation didn’t turn into an all-out war because Beeper is a tiny start-up with not nearly enough cash to fight a legal battle. (Beeper was eventually absorbed into Automattic, the company that makes WordPress.com and Tumblr, and Eric Migicovsky, its founder, now works on rebooting the Pebble smartwatch.) Mostly, the game was fought between Google and Apple proponents in a niche corner of the internet. This is not the same game, and I would be surprised if it ends any way other than a drawn-out fight. If Apple decides to pull the plug on Google’s unauthorized access to AirDrop — if such a thing is even possible — Google will no doubt retaliate somehow, either in the courtroom or online. (Remember “Get the message?”) If Apple can’t pull the plug because Google’s access uses Apple devices in a data center somewhere, it will send Google a cease and desist at least and a lawsuit at most.

The last possible result is the honeymoon ending: Google and Apple collaborate to bring AirDrop to Android. The likelihood of this is slim but possible, since both companies are embroiled in antitrust cases from the Justice Department and don’t wish to appear anticompetitive even in the slightest. (The latter matters especially to Apple, which is subject to investigation, even under the amiable-to-bribes Trump administration.) After the Beeper Mini ordeal, Apple added support for the Rich Communication Service, or RCS, in iOS 18, streamlining communication between Android and iOS devices. Those messages still aren’t end-to-end encrypted, since Apple uses the open standard which lacks encryption as opposed to Google’s which has it, but that’s coming as soon as Google adopts the new version of RCS. There’s precedent for collaboration, especially under consumer pressure. (I’m a proponent of the honeymoon ending because interoperability is good.)

This sets aside whether or not I think an antitrust investigation would actually succeed in court. I don’t think Google’s argument — that it can reverse a private company’s technology however it wishes without permission — would hold up in the eyes of any jury or judge, especially since Google has itself advocated it shouldn’t share its private search data with competitors since that’s proprietary information. The same logic applies in both cases. But it’s unlikely that a case would go to trial in the end, given the importance of Google and Apple to each other. They have a search deal worth billions of dollars, and they’re about to have an artificial intelligence deal to bake Gemini into Siri for some high price. These companies are reliant on each other, and it’s unlikely they’d fight in a courtroom. They would probably just settle.

Simon Willison, writing on his delightful blog:

Google released Gemini 3 Pro today. Here’s the announcement from Sundar Pichai, Demis Hassabis, and Koray Kavukcuoglu, their developer blog announcement from Logan Kilpatrick, the Gemini 3 Pro Model Card, and their collection of 11 more articles. It’s a big release!

I had a few days of preview access to this model via AI Studio. The best way to describe it is that it’s Gemini 2.5 upgraded to match the leading rival models.

Gemini 3 has the same underlying characteristics as Gemini 2.5. The knowledge cutoff is the same (January 2025). It accepts 1 million input tokens, can output up to 64,000 tokens, and has multimodal inputs across text, images, audio, and video.

I strongly agree with Willison: Gemini 3 isn’t a groundbreaking new model like GPT-4 or Gemini 2. I think large language models have hit a point of maturity where we don’t see such groundbreaking leaps in intelligence with major releases. The true test of these models will be equipping them with the correct tools, integrations, and context to be useful beyond chatbots. Examples include OpenAI’s acquisition of Software Applications Inc., the makers of the Sky Mac app; Gemini’s features in Chrome, Android, and ChromeOS; and Apple’s “more personalized Siri,” which is apparently due for launch any time between now and the world’s ending. That’s why Silicon Valley companies are hell bent on “agents” — they’re applications of LLMs that prove useful sometimes.

Back to Gemini 3, which, nevertheless, is an imposing model. It beats Claude Sonnet 4.5, GPT-5.1, and its predecessor handily in every benchmark, with the notable exception of SWE-bench, a software engineering benchmark that Claude still excels at. (SWE-bench tests models’ capability in fixing bug reports in real GitHub repositories, mostly in Python.) That’s unsurprising to me because Claude is beloved for its programming performance. Even OpenAI’s finest models cannot compete with Claude’s ingenuity, clever personality, and syntactically neat responses. Claude always matches the complexity of the program as it is. For instance, if a program isn’t using recursion, Claude understands that it probably shouldn’t, either, and uses a different solution. ChatGPT, on the other hand, just picks whatever is most efficient and uses as few lines of code as possible.

Gemini is quite competent at programming, but I don’t regularly use it for that. Gemini 3 Pro does not change this. It has historically been poor at SwiftUI, unlike ChatGPT, and I find its coding style to be unlike mine. It takes a very verbose route to solving problems, whereas Claude treats its users like adults. That’s not to say Gemini 3 is bad at programming, but it certainly is not as performant as Claude Sonnet 4.5 or GPT-5.1 with medium reasoning. Interestingly, Google has launched a new Visual Studio Code fork on Tuesday called Antigravity, with free support for Gemini 3 Pro and Claude Sonnet 4.5. I assume this will be Google engineers’ text editor of choice going forward, and it gives the newly acquired Windsurf team something to do at Google. Cursor should be worried — Antigravity’s Tab autocomplete model is equally as performant and it has great models available for free with “generous” rate limits.

Outside of programming, I found I used Gemini 2.5 Pro for analyzing and working with long text documents, like PDFs, the most. This is not just because of its industry-leading one-million-token context window, but because it’s trained to read the entire document and cite its sources properly. I don’t know what sorcery Google did to make Gemini so good at this, but OpenAI could learn. ChatGPT still writes (ugly) Python code to read bits of the document at a time, and often fails to parse text when it isn’t perfectly formatted. Claude’s tool calling, meanwhile, is nowhere near as good as Gemini or ChatGPT, and I seldom upload documents to it. In recent weeks, however, I’ve been uploading more documents to ChatGPT as I found that, despite its flaws, it was doing a slightly better job than Gemini only because GPT-5.1 is newer. Now that ChatGPT no longer has that advantage, I’m happy to go back to Gemini for my document reading needs.

Gemini 2.5 Pro was also the best for engineering-related explanations like physics, chemistry, and mathematics. ChatGPT got these questions right — and is much quicker than Gemini — but I appreciate Gemini’s problem-solving process more than GPT-5.1, even when set to the Candid personality. But again, in recent weeks, I’ve veered away from Gemini and switched to Claude for these explanations, despite Claude not rendering LaTeX math equations half the time, because I could feel Gemini 2.5 Pro getting old. (“Old” in the context of LLMs means untouched in three months.) Claude Sonnet 4.5 had more detail in its explanations and provided more robust proofs of certain math concepts, like ChatGPT, but with a more teacher-like personality. Gemini once again takes the crown for these kinds of explanations.

All of this is to say that Gemini 3 Pro is a great model, and I’m excited to use it again after the blockbuster launch of Gemini 2.0 Pro. It’s just that its predecessor was getting a bit old, but Google is back in the race. Here are my current use cases for the three major artificial intelligence chatbots at the end of 2025:

• ChatGPT: Search and a great Mac app. Useful for general chatting and reliable answers.
• Claude: Claude Code, Cursor, and literary analysis. Useful for its math explanations and nuance.
• Gemini: Image analysis and document uploads. Also, copyable LaTeX.

Jay Peters, reporting for The Verge:

The new headset is called the Steam Frame, and it’s trying to do several things at once. It’s a standalone VR headset with a smartphone-caliber Arm chip inside that lets you play flat-screen Windows games locally off the onboard storage or a microSD card. But the Frame’s arguably bigger trick is that it can stream games directly to the headset, bypassing your unreliable home Wi-Fi by using a short-range, high-bandwidth wireless dongle that plugs into your gaming PC. And its new controllers are packed with all the buttons and inputs you need for both flat-screen games and VR games.

The pitch: Either locally or over streaming, you can play every game in your Steam library on this lightweight headset, no cord required. I think Valve may be on to something.

Additional reporting from Sean Hollister, also at The Verge:

The Steam Machine is a game console. From the moment you press the button on its familiar yet powerful new wireless gamepad, it should act the way you expect. It should automatically turn on your TV with HDMI commands, which a Valve engineer tells me was painstakingly tested against a warehouse full of home entertainment gear. It should let you instantly resume the last game you were playing, exactly where you left off, or fluidly buy new ones in an easily accessible store.

You’ll never see a desktop or command line unless you hunt for them; everything is navigable with joystick flicks and gamepad buttons alone. This is what we already get from Nintendo, PlayStation, and Xbox, yet it’s what Windows PCs have not yet managed to achieve.

I rarely write about video games on this blog because I’m not much of a gamer, and the only games I do play are on PC. But this news is too significant not to write about: The Steam Frame and Steam Machine are consoles that can play virtually any PC game in virtual reality or on the television. Consoles have never differentiated themselves by specifications and usually have similar processors. They’re seldom updated, and when they are, they provide massive leaps in performance. The biggest differentiating factor between consoles is video game selection. Some games, like ones made by Sony, Microsoft, or Nintendo, are only available on their respective consoles. The “console wars” are really just game wars. On the opposite side of the spectrum, PCs play all games at much higher resolutions and frame rates than consoles, but they have a high barrier to entry. They require a monitor, peripherals, and competitive hardware.

The Steam Machine combines the best parts of PCs and consoles: a low barrier to entry and virtually unlimited game selection. It’s the perfect console. The popularity of the Steam Deck did the hard work of optimizing PC games for console players, and now, the Steam Machine can leverage that popularity to offer consumers a vast catalog of PC games in a console format. Valve, if the Steam Machine is priced competitively to the PlayStation 5 Pro and Xbox Series X, could probably eclipse a decent share of those sales. The games are already there (via Steam), they’re optimized for console play (via the Steam Deck), and the console is powerful enough to play them. If Valve can pull this off, it would be a truly remarkable disruption in the console wars. People wouldn’t even have to buy their beloved games again if they bought them on their computer, because the Steam Machine is literally just Steam.

I’m less bullish on the Steam Frame. The idea of consoles is that they’re cheap, i.e., they have low barriers to entry. People can just buy one at Best Buy and connect it to their existing television. VR, as I’ve established numerous times on this website, is a luxury purchase. People do not see an immediate need for VR in their lives, and if it’s a dollar more than $500, they’ll probably turn their nose up at it. Meta is the only company that has truly succeeded at VR because the Meta Quest 3S is inexpensive enough to buy as a gift. It’s not extravagant. If the Steam Frame costs anything more than the Meta Quest 3S, as it most likely will be, people won’t buy it, irrespective of the limitless game selection. The limited games the Meta Quest offers are good enough for most people. I think it’s a great idea, but price matters much more to VR customers because it’s such a burgeoning market. It hasn’t achieved maturation or commodification whatsoever.

Hayden Field and Tom Warren, reporting, reporting for The Verge:

OpenAI is releasing GPT-5.1 today, an update to the flagship model it released in August. OpenAI calls it an “upgrade” to GPT-5 that “makes ChatGPT smarter and more enjoyable to talk to.”

The new models include GPT-5.1 Instant and GPT-5.1 Thinking. The former is “warmer, more intelligent, and better at following your instructions” than its predecessor, per an OpenAI release, and the latter is “now easier to understand and faster on simple tasks, and more persistent on complex ones.” Queries will, in most cases, be auto-matched to the models that may best be able to answer them. The two new models will start rolling out to ChatGPT users this week, and the old GPT-5 models will be available for three months in ChatGPT’s legacy models dropdown menu before they disappear.

Here is an example OpenAI posted Wednesday to showcase the new personality:

Prompt: I’m feeling stressed and could use some relaxation tips.

ChatGPT 5: Here are a few simple, effective ways to help ease stress — you can mix and match depending on how you’re feeling and how much time you have…

ChatGPT 5.1: I’ve got you, Ron — that’s totally normal, especially with everything you’ve got going on lately. Here are a few ways to decompress depending on what kind of stress you’re feeling…

I find GPT-5.1 to be a major regressive step in ChatGPT’s similarity to human speech. Close friends don’t console each other like they’re babies, but OpenAI thinks they do. GPT-5.1 sounds more like a trained human resources manager than a confidant or kin.

Making a smart model is only half the battle when ChatGPT has over 800 million users worldwide: the model must also be safe, reliable, and not unbearable to speak to. People use ChatGPT to journal, write, and even as a therapist, and a small subset of those individuals might use ChatGPT to fuel their delusions or hallucinations. ChatGPT has driven people to suicide because it doesn’t know where to draw the line between agreeability and pushback. GPT-5.1 aims to make significant strides in this regard, being more “human-like” in benign conversations and careful when the chat becomes concerning.

What I’ve learned since GPT-5’s launch in August is that people really enjoy chatty models. I even think I do, though not in the way OpenAI defines “chatty.” I like my models to tell me what they’re thinking and how they came to an answer, so I can see if they’ve hallucinated or made any flaws in their reasoning. When I ask for a web search, I want a detailed answer with plenty of sources and an interpretation of those sources. GPT-5 Thinking did not voluntarily divulge this information — it wrote coldly without any explanation. For months, I’ve tweaked my custom instructions to tell it to ditch the “Short version…” paragraph it writes at the beginning and instead elaborate on its answers to varying degrees of success. GPT-5.1 is a breath of fresh air: It doesn’t ignore my custom instructions like GPT-5 Thinking, but also intentionally writes and explains more. In this way, I think GPT-5.1 Thinking is fantastic.

But again, this isn’t how OpenAI defines “chatty.” GPT-5.1 is chattier not only in my definition, but OpenAI’s definition, which can only really be categorized as “someone with a communications degree”. It’s not therapeutic, it’s unsettling. “I’ve got you, Ron?” Who speaks like that? OpenAI thinks that getting to the point makes the model sound robotic, when really, it just sounds like a human. Sycophancy is robotic. The phrase, “How can I help you?” sounds robotic to so many people because it’s sycophantic and unnatural. Not even a personal assistant would speak like that. Humans value themselves — sometimes over anyone else — but the new version of ChatGPT has no self-worth. It always speaks in this bubbly, upbeat voice, as if it is speaking to a child. That’s uncanny and makes the model sound infinitely more robotic. I think this is an unfortunate regression.

My hunch is that OpenAI did this to make ChatGPT a better therapist, but ChatGPT is not a therapist. Anthropic, the maker of Claude, knows how to straddle this line: When Claude encounters a mentally unstable user, it shuts the conversation down. It always deviates. And when Claude’s responses have gone too far, it kills the chat and prevents the user from speaking to the model in that chat any further. This is important because research has shown that the more context a model must remember, the worse it becomes at remembering that context and involving its safety features. If a user immediately tells the model that they are suicidal right as they start a chat, the model will adhere much better to instructions than if they fill its context window with junk first. (This is how ChatGPT has driven people to suicide.) GPT-5.1 takes a different approach: Instead of killing the chat, it tries to build rapport with the user to hopefully talk them down from whatever they’re thinking.

OpenAI thinks the only way to do this is to be sycophantic from the start. But Anthropic has shown that a winning personality doesn’t have to be obsequious. Claude has the best personality of any artificial intelligence model on the market today, and I don’t think it sounds robotic at all. GPT-5.1 Thinking is chatty in all the wrong ways. It might be “safer,” but only marginally, and not nearly as safe as it should be.

If you are having thoughts of suicide, call or text 988 to reach the National Suicide Prevention Hotline in the United States.

Mark Gurman, reporting mid-October for Bloomberg:

Apple Inc. is preparing to finally launch a touch-screen version of its Mac computer, reversing course on a stance that dates back to co-founder Steve Jobs.

The company is readying a revamped MacBook Pro with a touch display for late 2026 or early 2027, according to people with knowledge of the matter. The new machines, code-named K114 and K116, will also have thinner and lighter frames and run the M6 line of chips.

In making the move, Apple is following the rest of the computing industry, which embraced touch-screen laptops more than a decade ago. The company has taken years to formulate its approach to the market, aiming to improve on current designs…

The new laptops will feature displays with OLED technology — short for organic light-emitting diode — the same standard used in iPhones and iPad Pros, said the people, who asked not to be identified because the products haven’t been announced. It will mark the first time that this higher-end, thinner system is used in a Mac.

And from his Power On newsletter Sunday:

I previously wrote about the first one: a revamped M6 Pro and M6 Max MacBook Pro with an OLED display, thinner chassis, and touch support. That’s slated to arrive between late 2026 and early 2027.

I’ll get the good news out of the way first: organic-LED displays coming to the Mac lineup (hopefully) next year is such great news. The mini-LED displays Apple has used since the 2021 MacBooks Pro were borrowed from the 2021 iPad Pro and were, back then, the best display technology Apple offered. OLED screens only shipped in small iPhones, and Apple’s highest-end display, the Pro Display XDR, used mini-LED too. Whereas traditional LED displays use a single backlight to illuminate the pixels, mini-LED screens use dozens of dimming zones to control smaller parts of the display separately. This results in deeper blacks, high dynamic range support, and better contrast, similar to OLED. However, OLED displays illuminate each pixel individually, enabling more precise light control and even better HDR. Think of mini-LED as a stopgap solution between LED and OLED.

The biggest problem with OLED displays is brightness. Because each pixel must be individually lit, it is quite difficult to engineer an OLED with equal brightness to the single-backlight LED as displays get larger. For Apple to make HDR monitors beginning in 2019, it had to use mini-LED because the technology to make large screens bright enough just wasn’t there. My high-end LG OLED television I bought in 2023 only has a maximum sustained brightness of around 150 nits when the whole screen is used. (Its peak brightness is much higher at 850 nits, making it suitable for HDR content.) By contrast, my MacBook Pro’s display reaches up to 1600 nits, making it readable in direct sunlight. The larger the display, the more difficult it is to use OLED and maintain brightness.

Apple solved this issue last year with its introduction of the M4 iPad Pro, using a display technology it calls “tandem OLED,” essentially two OLED displays stacked atop each other. This doubles the brightness and maintains all of the perks of OLED, and even remains much thinner than the original mini-LED design. This was an extremely complex technology to engineer — LG, Apple’s OLED display supplier, had been working on it for years — and therefore, only arrived on the highest-end iPad Pro models (which received a price increase). For Apple to transition away from mini-LED, it would have to implement a tandem OLED panel in the MacBook Pro, which would be enormously challenging and expensive. The processor would also have to be capable of running both panels simultaneously — this was why the 2024 iPad Pro used the M4 chip, skipping over the M3.

However Apple plans to do this, I’m incredibly excited, and will gladly pay a premium for an OLED MacBook Pro. Selfishly, I hope these models launch in late 2026, because I planned to update my current M3 Max MacBook Pro this year until Apple delayed them to January 2026.

On to the disappointing news: Who wants a touchscreen? Probably quite a few Mac laptop buyers, but I’m dismayed by this rumor. Irrespective of Apple’s modern reasons for omitting touchscreens from Mac laptops — it doesn’t want to eclipse iPad sales — I don’t think Mac computers are designed for touchscreens. macOS is historically built around Macs’ excellent, class-leading trackpads, with smooth scrolling, gestures, and intuitive controls. iOS is designed for touchscreens — macOS is not. I would even say Windows isn’t, either, because every time I’ve used a Windows laptop with a touchscreen I’ve wanted to defenestrate the thing, but Windows laptops’ trackpads are so abysmally poor that I understand why most people use them. Windows is not in any way comparable to macOS; macOS is an intentionally designed operating system, for one.

The desktop web is not designed for touch input. (And the mobile web, even in 2025, is also horrible. Have you tried booking a flight on a smartphone?) Touch targets are tiny, there are floating toolbars, and the experience is sub-par. The cursor is the only proper way to interact with a desktop OS, and macOS is designed perfectly around the trackpad. The only reason Apple would ever consider adding touchscreens to Mac laptops is pure advertising. “Look, we have touchscreens too! Buy a Mac!” Pathetic revisionist reasoning. There’s a reason Steve Jobs said touchscreens don’t belong on Macs: it’s just a poor user experience in every dimension.

I implore those who, unlike me, are fine with smudges on their laptop displays to try tapping some buttons in macOS with their finger. Nobody can convince me that it is a natural gesture. When I use my computer, I keep my left hand on the left side of the keyboard, and I use my right to control the mouse or trackpad. The left hand switches between windows using Command-Tab and handles keyboard shortcuts like Command-W, while the left selects items using the cursor. This is the most efficient way to use a computer, and macOS has always encouraged users to train themselves this way. Every well-designed Mac app supports the same gestures and keyboard shortcuts. They work anywhere in the system. Spotlight makes getting to apps and files easy — Windows has nothing like that, let alone a Command-Space keyboard shortcut.

I am not old. I only vaguely remember a time before touchscreens because I was a child then. I appreciate my iPad and I adore my iPhone because touchscreens make those devices magical and easy to use. But would anyone create a touchscreen TV? Of course not, because that would be preposterous. The remote control was invented for a reason, and so was the cursor. The mouse cursor is not a vestige of the past, but is a common-sense method of computing. The internet is designed around the cursor and the keyboard, and lifting your hands up from the keyboard position just doesn’t make any sense. I truly hope and believe Apple will include an option on these new laptops to disable the touchscreen.

Mark Gurman, reporting for Bloomberg:

Apple Inc. is planning to pay about $1 billion a year for an ultrapowerful 1.2 trillion parameter artificial intelligence model developed by Alphabet Inc.’s Google that would help run its long-promised overhaul of the Siri voice assistant, according to people with knowledge of the matter.

Following an extensive evaluation period, the two companies are now finalizing an agreement that would give Apple access to Google’s technology, according to the people, who asked not to be identified because the deliberations are private…

Under the arrangement, Google’s Gemini model will handle Siri’s summarizer and planner functions — the components that help the voice assistant synthesize information and decide how to execute complex tasks. Some Siri features will continue to use Apple’s in-house models.

The model will run on Apple’s own Private Cloud Compute servers, ensuring that user data remains walled off from Google’s infrastructure. Apple has already allocated AI server hardware to help power the model.

This version of Gemini is certainly a custom model used for certain tasks that Apple’s “foundation models” cannot handle. I assume the “summarizer and planner functions” are the meat of the new Siri, choosing which App Intents to run, parsing queries, and summarizing web results. It wouldn’t operate like the current ChatGPT integration in iOS and macOS, though, because the model itself would be acting as Siri. The current integration passes queries from Siri to ChatGPT — it does nothing more than if someone just opened the ChatGPT app themselves and prompted it from there. The next version of Siri is Gemini under the hood.

I’m really interested to see how this pans out. Apple will probably be heavily involved in the post-training stage of the model’s production — where the model is given a personality and its responses are fine-tuned through reinforcement learning — but Google’s famed Tensor Processing Units will be responsible for pre-training, the most computationally intensive part of making a large language model. (This is the P in GPT, or generative pre-trained transformer.) Apple presumably didn’t start on developing the software and gathering the training data required to build such an enormous model — 1.2 trillion parameters — early enough, so it offloaded the hard part to Google for the low price of $1 billion a year. The model should act like an Apple-made one, except much more capable.

This custom version of Gemini should accomplish its integration with Apple software not just through post-training but through tool calling, perhaps through the Model Context Protocol for web search, multimodal functionality, and Apple’s own App Intents and personal context apparatus demonstrated at the 2024 Worldwide Developers Conference. I’m especially intrigued to see what the new interface will look like, especially since Gemini might take a bit longer than Siri today to generate answers. There is no practical way to run a 1.2 trillion-parameter model on any device, so I also wonder how the router will decide which prompts to send to Private Cloud Compute versus the lower-quality on-device models.

I do want to touch on the model’s supposed size. 1.2 trillion parameters would make this model similar in size to GPT-4, which was rumored to be 1.8 trillion parameters in size. GPT-5 might be a few hundred billion higher, and one of the largest models one can run on-device is GPT-OSS with a size of 120 billion parameters. A “parameter” in machine learning is a weight given to a learnable value. LLMs predict the probability of the next word in a token in a sequence by training on many other sequences. The weights of those various probabilities are parameters. Therefore, the more parameters, the more probabilities (“answers”) the model has. Most of those parameters would not be used during everyday inference, as Federico Viticci points out on Mastodon, but it’s still important to note how large this model is.

We are so back.