Apple’s WWDC Retreat, Liquid Glass and One Question, Meta Puts $14.8 Billion Toward an AI Reset
Summary
Apple’s most consequential WWDC signal was retreat: after promising AI beyond what OpenAI, Anthropic, or Google could deliver, it deferred unfinished features and demonstrated only what it could ship. Ben took the L for granting Apple too much benefit of the doubt, but called the reset “sensible” and “getting your house in order” — a necessary return to execution over aura.
Opening Apple’s on-device models gives developers cheap experimentation and gives Apple asymmetric platform upside, even though the models remain “vastly inferior” to cloud AI. A solo developer such as Overcast’s Marco Arment might add podcast transcription without cloud infrastructure: if it fails, the app absorbs the complaints; if it succeeds, high-end iPhones gain differentiated software. Ben’s caveat was blunt: “These models do still stink,” so this is the right move that may not move the needle.
Deeper ChatGPT integration could create a genuinely compelling AI reason to choose an iPhone, despite Apple not owning the underlying intelligence. iOS 26’s Visual Intelligence can send on-screen material directly to ChatGPT, while signed-in users can continue the conversation on a computer. Ben wants Apple to accept being “a ChatGPT vessel”: Apple supplies the tortilla chip, and OpenAI supplies “some tasty salsa.”
Apple’s distribution retreat may have arrived too late to preserve control, particularly after the appeals court declined to stay Judge Yvonne Gonzalez Rogers’ anti-steering injunction. Ben still favors locking down games, where consoles and Steam provide precedent and spending revolves around “whales,” gems, dances, and digital clothes, while freeing productivity apps and Kindle. The new Games app could support that bifurcation, but Andrew’s verdict was that Apple “lost so hard that you’ve now lost control.”
Liquid Glass is now a test of whether Apple still possesses its historical product advantage, not merely a cosmetic redesign. Ben expects roughness and says judgment may require one or two years, as iOS 7 did not settle until iOS 9 or 10; John Gruber’s favorable early view came with the possibility that Apple “shipped it a year early.” If the interface proves Settings-app bad rather than temporarily unfinished, the strategic question becomes whether Apple can still do what it traditionally does well.
Meta’s reported $14.8 billion investment for a 49% stake in Scale AI looks less like a capabilities acquisition than one of the world’s most expensive acqui-hires. Against Meta’s roughly $65 billion-$70 billion of annual CapEx and $42 billion of Q1 2025 revenue, the price is manageable; the stranger point is that Meta ownership may drive OpenAI and Google away from Scale’s data business. Installing Alexander Wang therefore signals “a real reset” after talent losses, a disappointing Llama 4, and an AI organization that appears messier than outsiders realized.
Meta is behaving from fear even though today’s LLMs could strengthen almost every layer of its existing business without achieving AGI. ChatGPT mobile usage reaching up to 20 minutes per user per day makes attention competition real. Andrew called losing time to OpenAI “classic losing,” and Ben agreed. Ben’s sharper organizational call was that Meta cannot progress until it removes Yann LeCun from the initiative—not because LeCun’s AGI skepticism is wrong, but because Ben sees his stance as undermining recruiting and being a poor fit for Meta’s immediate opportunities, such as advertising against “every single pixel.”
Deep dive
1. Apple traded its aura for credibility at a “sensible” WWDC
Ben’s self-audit began with the Apple aura: suppliers sometimes accept little profit because Apple funds them to learn how to do completely new things “that have never been done before,” knowledge they can later sell elsewhere. That history earned Apple extraordinary benefit of the doubt on both Vision Pro and Apple Intelligence.
Vision Pro remains “awesome,” but Ben joked that it costs $4,500, weighs “like, 4,500 tons,” and lacks applications. Its emerging destination is enterprise-specific work, while the consumer product resembles “the exercise equipment of computing devices”: satisfying once used, yet usually sitting untouched and inducing guilt.
Last year’s strategy looked structurally elegant: Apple controlled the device and private data, could handle personal tasks locally, pass broad work to foundation models, and perhaps auction model access like its Google Search deal. The missed qualification was execution; Apple was promising experiences that even OpenAI, Anthropic, and data-rich Google could not deliver.
Apple is culturally optimized for finished products—a phone must be boxed in September and its OS ready—not gazillions of edge cases, graceful failure, and continuous iteration. This keynote effectively validated John Gruber’s self-critique: unfinished AI moved into the coming year, demonstrated features were real, and the company began “getting your house in order.”
2. Developer access converts weak local models into platform optionality
Ben’s essential technical caveat: an on-device model running from a phone battery will be “vastly inferior” to ChatGPT running across massive NVIDIA infrastructure, including GPUs he characterized as costing $50,000. Privacy may be real, but “I’m sorry, Apple, you don’t get to save money and sell it as a feature.”
Opening the model changes the economics for small developers. Overcast creator Marco Arment cannot make recurring cloud inference fit a one-person podcast-app business, but free access to the user’s chip might let him offer transcription without acquiring cloud expertise or absorbing usage-based infrastructure costs.
The risk-reward favors Apple: if Overcast’s transcription stinks, Marco handles bad reviews; if it works, iPhone podcast apps gain a capability that may be inconsistent across Android’s mix of high- and low-end hardware. Developers can also discover “weird things that you never thought to use AI for” because experimentation is free.
Ben allowed that Apple may always have planned this cadence: build against its own API for a year, revise it through dogfooding, then accept the permanent support burden of public release. He nevertheless took the W while preserving the limit: local models “do still stink,” and the right strategic step “may or may not move the needle.”
3. Apple is becoming the vessel for ChatGPT rather than its substitute
iOS 26’s Visual Intelligence can analyze what is visible on an iPhone screen and offer ChatGPT immediately, avoiding the previous sequence of trying to answer on-device and handing off only after failure. Ben read that as “the appropriate level of humility about your capabilities.”
For signed-in ChatGPT users, the integration preserves conversation continuity across the iPhone and computer. Android may receive deeper and faster Gemini integration because Google controls the entire stack, but an iPhone becomes comparatively attractive if ChatGPT itself influences the customer’s phone choice.
Ben also viewed OpenAI as a better partner than Google because ChatGPT is winning in the consumer space. Apple can benefit from that demand without pretending its own model is sufficient.
Andrew’s demand was simple: users do not want ChatGPT treated as the backup after Apple’s model fails—“just give me ChatGPT.” Ben’s metaphor accepted Apple’s humbler but potentially profitable position: “I want a ChatGPT vessel.” Apple is the tortilla chip; OpenAI provides “some tasty salsa.”
4. Apple’s App Store retreat followed a preventable legal defeat
The new Games app is not yet a separate commercial regime, but games can appear there and in the App Store, laying groundwork for differentiated purchasing rules. Ben classified it as another retreat, while noting that some retreats occur because “you’re getting shot in the back of the head.”
After the appeals court denied Apple’s request to stay Judge Yvonne Gonzalez Rogers’ injunction, Ben considered app steering effectively open. His preferred compromise had always been permissive commerce for most apps but console-like restrictions for games, where both consoles and Steam demonstrate the legitimacy of a tightly controlled model.
His economic distinction is deliberately unsentimental: game publishers target “digital whales” for gems, dances, and character clothes, while Kindle and productivity developers help sell powerful devices and need sustainable commerce. Apple could have protected gaming revenue while loosening everything else; in the US, its delay means games must now steer too.
5. Making the iPad more Mac-like revives an unresolved platform bargain
Ben has largely abandoned the iPad beyond drawing in Procreate, partly because a longstanding sleep problem leaves it uncharged. Andrew defended the opposite experience: a single-window, notification-free device that sits between phone and Mac and preserves focus rather than maximizing the processor.
The lost ideal, for Ben, was Steve Jobs demonstrating GarageBand at the iPad 2 keynote with the satisfaction that “now anyone can make music.” Apple then undermined ambitious full-screen software by offering developers weak monetization—initially one-time purchases—while concentrating on revenue from gems in games.
New overlapping windows feel like Apple conceding, “Let’s make the iPad more like a Mac, I guess. People like Macs.” Ben no longer objects on behalf of the iPad, but asks for reciprocity: if a locked-down iOS device can resemble a computer, “let the Mac be like the Mac and not like an iPhone.”
His concrete grievance came while remotely transferring Andrew’s French Open file through a Mac Mini and NAS. The transfer did not run until he opened screen sharing and found a permission pop-up asking whether Terminal could access external devices: “I’m in the Terminal. I don’t need to be asked.”
6. Liquid Glass will test Apple’s remaining core competence
One listener, a materials engineer, protested that glass is an amorphous solid and “there is no such thing as liquid glass”; another said the interface resembles what a Hollywood production designer would create for an expensive science-fiction movie. Ben dismissed the literal objection but withheld any product verdict.
If iOS 7 is the guide, the first year will produce broken screenshots and complaints, with the design perhaps not settling until the equivalent of iOS 9 or 10. Ben will not install the beta: “I am a tech analyst. I am not a product reviewer,” and short battery life offers him no analytical advantage.
Andrew saw the possibility of Apple’s characteristic tactile joy but found that last year’s unfulfilled videos had destroyed automatic trust: “Will any of this actually work?” His frustration with Photos, Settings, and declining reliability means even Apple’s traditional strengths now require proof.
John Gruber reportedly likes Liquid Glass despite rough edges and suggested Apple might have “shipped it a year early,” perhaps because it lacked another headline. Its processing demands still showcase genuine integration—Apple controls processors, graphics, and efficient subroutines—but Ben’s closing hedge was pointed: “Hopefully they’re still good at it.”
7. Scale AI’s economics reveal a Meta leadership reset
Ben compared the headline to Apple’s $3 billion Beats purchase when Apple was worth roughly $600 billion: the nominal figure overwhelmed its strategic significance. Meta is spending an estimated $65 billion-$70 billion on CapEx this year, while Andrew cited $42 billion of Q1 2025 revenue, making $14.8 billion large but absorbable.
The harder question is why Meta wants 49% of Scale AI. Scale began in data labeling, including documenting reasoning steps and low-cost overseas labeling operations; Meta, Google, and OpenAI have all needed such data, although OpenAI increasingly performs the work internally.
Partial Meta ownership potentially destroys Scale’s independent value: why would OpenAI or Google fund a supplier that might pass data to a 49% owner and direct competitor? “The nature of the acquisition kills the value of what you’re acquiring,” Ben argued, so Scale’s standalone capabilities cannot be the complete rationale.
The circumstantial explanation is organizational crisis: Llama personnel departed, Llama 4 disappointed, benchmark presentation was allegedly massaged, and Zuckerberg sounded unfocused—“spaghetti against the wall”—when discussing AI. Wang is not a frontier researcher, but neither is Sam Altman; this looks like “one of the world’s most expensive acqui-hires” for a leader.
8. Meta’s defensive crouch obscures an unusually additive technology
A listener cited ChatGPT mobile usage of up to 20 minutes per user per day and reports that Meta product managers view its consumer AI placement as forced from the top. The warning was that relative weakness versus ChatGPT, Gemini, and Claude could damage a crucial top-line growth vector in Meta’s core attention business.
Ben accepted the “fear” framing: fear creates a defensive crouch that hides opportunities. He was more optimistic about AI’s fit with Meta than Meta appeared to be. During his interview with Mark Zuckerberg, he pushed him on benefits spanning existing products and the metaverse; Ben said Zuckerberg’s articulation was poor and omitted the metaverse point, which had appeared in the earnings call published before the interview.
He also corrected the terminology: OpenAI taking attention is not disruption but Meta getting beaten. In the sustaining-versus-disruptive framework, adding AI to Meta’s existing products is sustaining innovation; the concern is that another company may execute it better.
9. Meta needs today’s LLMs more than tomorrow’s “God machine”
Ben’s categorical organizational call was: “Facebook’s not going anywhere until they fire Yann LeCun.” He stressed that LeCun is a Turing Award-winning pioneer and that he shares LeCun’s skepticism about LLMs reaching human-like AGI; the problem is fit, not intellect or personal respect.
Joseph’s recruiting argument carried the disagreement: why join an organization whose most famous researcher dismisses frontier LLM work when rivals pitch, “Get in, loser. We’re building the God machine”? Alexander Wang may coordinate the reset, but Scale AI supplies data rather than the hard research Meta is trying to recruit.
Meta does not require AGI for Ben’s most ambitious commercial vision. Existing AI could recognize Andrew’s green shirt, let its maker bid whenever the item appears, make it clickable, and ultimately advertise against “every single pixel on a Facebook property.” That is sustaining innovation available with current technology.
Zuckerberg is therefore right to spend aggressively, including possible nine-figure researcher packages; hardware worth tens of billions needs exceptional human software. Ben’s correction to his own Meta thesis was that structural opportunity obscured execution risk: he underweighted talent losses, including that Mistral is largely made up of ex-Llama engineers, and now concludes “stuff is a much bigger mess than it seems.”