Pei's core argument is that the smartphone's app model is structurally inefficient. His working example is mundane and well-chosen: arranging to meet someone for coffee currently requires four separate apps a messaging app, maps, a ride service, a calendar. His proposed alternative is a system that reads user intent and executes through AI. "I know you very well, and if I know your intention, I just do it for you," he told TechCrunch at SXSW.

The broader point is that the smartphone's interaction model lock screen, home screen, app store, full-screen apps hasn't meaningfully changed in roughly 20 years. "The current way we use phones is very old-school. It's pre-iPhone," Pei said. "There used to be Palm Pilots and PDAs back in the day. And if you think about the user experience, it's still very similar."

Nothing is building toward this through what it calls "AI-native devices" and tools like Essential Apps homescreen widgets designed to reduce reliance on full applications as early signals of a broader platform shift, per 9to5Google. The timeline, though, is not short. Pei has previously said a no-app future could take 7–10 years because "people love using apps," and Nothing's current OS still lets users build their own mini-apps a hedge that runs somewhat counter to the agent-first vision he describes.

Smart glasses are the logical hardware endpoint of this thesis: a form factor with no home screen, no app store, and an interface built for AI rather than human thumbs. The gap between that vision and a shippable product is real. What matters for evaluating Nothing's prospects is the type of device Pei's thinking implies and what the market looks like for anyone trying to build it right now.

For most of the past decade, smart glasses failed to find an audience. Over the past year, Meta's Ray-Ban and Oakley partnerships shifted the category toward mainstream not through AR displays, but through normal-looking frames with open-ear audio, a small camera, and a voice-activated AI assistant, as the EFF documented three weeks ago. Fashion and fit moved units where technology alone could not. The lesson is that simple.

The privacy model underneath that success is a different story.

There is no way to use Meta's AI features without sending data to Meta's servers. Every AI interaction requires a live connection to Meta's infrastructure, regardless of what users believe they've opted into. The glasses were marketed with phrases like "designed for privacy, controlled by you" and "built for your privacy," per ALM Corp's reporting three weeks ago. The class action filed in the US on March 5 argues those claims did not reflect the product's actual data practices.

The investigation that prompted the lawsuit was damaging in concrete terms. Swedish newspapers Svenska Dagbladet and Göteborgs-Posten revealed that contractors at a subcontractor facility in Nairobi were reviewing footage from Meta's glasses including video of users undressing, using the toilet, and handling bank cards. "We see everything from living rooms to naked bodies," one worker told the journalists, according to ALM Corp. Another offered a line that has circulated widely since: "You think that if they knew about the extent of the data collection, no one would dare to use the glasses."

The UK's Information Commissioner's Office responded by formally writing to Meta, calling the findings "concerning." The Register reported the ICO's statement on March 5:

"Devices processing personal data, including smart glasses, should put users in control and provide appropriate transparency. This includes where user data is used to train or develop AI systems. Service providers must clearly explain what data is collected and how it is used."

That standard describes what current hardware does not deliver. Meta built the category's first mass-market success on terms-of-service disclosures rather than anything the wearer or bystander can actually observe. The ICO's intervention and the US class action signal that the bill for that approach is coming due.

Pei's four words collapse two distinct problems. The first is an optics engineering challenge. The second is a trust architecture challenge. They are related, but solving one does not solve the other.

On transparency: the "see-through" part requires a waveguide, an optical component embedded in the lens that channels projected light to the wearer's eye while allowing real-world light to pass through simultaneously. Waveguides must balance field of view, weight, transparency, image quality, optical efficiency, and cost, and IDTechEx's analysis from seven weeks ago makes clear that no current design wins on all of them. Thinner, more fashionable diffractive designs tend to lose on optical efficiency, which hits battery life directly. The more efficient reflective designs carry high manufacturing costs. This is precisely why the glasses that sell in volume today skip the display altogether.

On "light up": Meta's glasses already include a small recording indicator light, but the EFF notes it was designed to be minimally visible to people outside the frame and can be disabled through inexpensive hardware modifications. A visible, tamper-resistant indicator that communicates capture and processing state to bystanders is an architectural choice, not a default. It is also what regulators are now starting to demand explicitly.

The EFF has also flagged that Meta is exploring facial recognition for its glasses a capability that would, in the organization's words, "obliterate the privacy of everyone." Physical legibility becomes more important, not less, as AI capabilities expand. A recording light that bystanders can actually see isn't a cosmetic detail. It's the hardware equivalent of a consent mechanism, and right now the category's leading product treats it as an afterthought.

For a new entrant, the trust architecture problem may be more tractable than the optics problem and more commercially meaningful right now. Nobody is holding off on buying smart glasses because waveguide efficiency is suboptimal. A growing number of people are holding off because they don't know what the glasses are doing, who can see what they capture, or whether the marketing reflects reality.

The commercial evidence points clearly toward a viable near-term form factor: audio-first glasses with a camera and AI, where the display is optional or absent. Meta's seven million units came overwhelmingly from exactly that design. A first-generation Nothing glasses product that prioritized fit, audio quality, and visible AI-state indicators over waveguide optics would be competing on the terrain that actually moves product today.

The differentiator with the clearest brief is privacy that is physically legible rather than contractually stated. In practice, that means:

• Capture indicators that are prominent, visible to bystanders, and genuinely non-bypassable not a dim LED that can be disabled with a cheap hardware mod
• On-device processing for basic functions, with explicit, visible notification when a feature requires a cloud connection the specific failure point the ICO and the US class action have both flagged in Meta's model
• Default behaviors that limit data transmission without requiring users to hunt through buried settings menus

Nothing's existing brand already communicates transparency physically, through visible circuit boards and exposed components, at a price point established players typically underserve. That aesthetic vocabulary maps directly onto what regulators, privacy advocates, and a growing share of consumers are now asking the category to become. The design language isn't a stretch it's almost a direct translation. The harder question is whether Pei's team treats "light up" as a genuine architectural commitment baked into the hardware from the start or as a tagline applied after the fact.

The next company to win in this space may not be the one with the best AI or the most sophisticated optics. It may be the one that makes capture, data use, and processing state legible to the person wearing the glasses and everyone around them.

Pei's SXSW comments confirm a direction, not a product. AI-native hardware, a shifted OS model, new form factors he hopes are transparent and visibly active. By his own prior estimate, the no-app future that would make those devices most useful could still be 7–10 years away, per 9to5Google.

The market need is real and present regardless of when the software vision arrives. Seven million AI glasses sold in a single year, for a product without a display, shows the category has already crossed from novelty to consumer product.

The structural problem Meta has exposed mandatory data transfer to use core features, human review of intimate footage, marketing language that said otherwise has created a regulatory and reputational gap that better specs alone cannot close. The ICO's standard is now on record. The class action is in the courts. And 2026 is shaping up as the year regulators in both the UK and US start demanding that smart-glasses makers show their work, not just cite their terms of service.

The winning formula in the next phase of the category is straightforward to state and hard to execute: hardware that looks like glasses, AI that works without requiring users to hand over everything, and physical signals that make the device's state legible to everyone in the room. "See-through and light up" turns out to be a better product brief than it first sounds provided whoever builds it understands that the "light up" part is the point.