Quoting Nilay Patel

The reason: A comment that sparks deep reflection

Simon Willison shared a comment from The Verge's editor-in-chief Nilay Patel on his blog, directly addressing the core dilemma of augmented reality (AR) glasses. Patel put it bluntly: to build the AR glasses everyone imagines, you must place a camera in front of your eyes continuously recording everything you see and send that data to the cloud for processing. There is no other way. And that means you inevitably invade people's privacy. He even suggested that perhaps we shouldn't build such a product at all.

This comment deserves serious attention because it doesn't indulge in techno-optimistic fantasies of "the future will solve it," but calmly points out the deadlock between physics and product form.

Breakdown: Why AR glasses are "destined" to violate privacy?

First, the ideal experience of AR glasses is to overlay information in your field of view in real time. This requires the glasses to "understand" your surroundings: recognize objects, faces, text, track gestures, and comprehend spatial relationships. Today's most advanced AI models can indeed do this, but they require massive computing power.

The problem is that cramming such power into a device the size of ordinary glasses seems impossible for now. Patel notes that no chip exists that is both "powerful enough and power-miserly enough" to fit into a glasses temple without turning it into a branding iron. You could make the device larger, like an Apple Vision Pro with a separate battery pack, but that's not the everyday glasses people want.

Therefore, the only viable path is to stream camera data to the cloud in real time, let colossal models process it, and send back the results. During this process, everything around you—passersby on the street, documents on your desk, passwords on your screen—is recorded in its raw form and passes through the manufacturer's servers. Even with so-called "privacy protection" promises, the very act of transmitting and processing data in the cloud opens backdoors. Privacy invasion becomes the product's "original sin," not a bug that can be patched later.

Trend insights: The ethical ceiling for AI hardware

This reveals a larger trend: AI hardware is hitting an ethical ceiling. We used to say "software defines everything," but when hardware interacts with the physical world, it inevitably touches privacy boundaries. Smart speakers have been listening to you, but at least they don't video you; your phone can photograph you, but it's not constantly on. AR glasses, however, would need to record everything in your line of sight every waking second, fundamentally changing the depth and breadth of privacy invasion.

We're already seeing regulatory signals. The EU's AI Act imposes strict limits on biometric recognition and real-time surveillance, and such products might hit a wall in Europe. Consumer sensitivity to privacy is also growing, so even if built, they could face fierce backlash. Nilay Patel's question — "maybe you shouldn't" — may become reality sooner than we think.

Practical implications: What should practitioners consider?

If you're a product manager or developer, this comment serves as a reminder: some product experiences may come at a cost too high to accept. When planning AI hardware, you can't just focus on "coolness" but must also consider "can we protect users' most basic privacy?" Perhaps the answer lies in abandoning the ambition of "all-day AR" and returning to scenario-based use (e.g., turning on the camera only for specific tasks); or waiting for a genuine edge AI breakthrough—but that might require decades of materials science advancement.

For consumers, the next time a company pitches you the "next-gen smart glasses," you should at least ask: does it need a constant cloud connection? How long are the images stored? Is there a physical camera cut-off switch? As Patel implies, we might need to decide before these products flood the market whether to allow them at all.

Counterintuitive: Tech progress doesn't always solve ethical problems

Many assume that technical problems will eventually be solved by technology. For instance, future ultra-low-power neuromorphic chips might process visual information locally without uploading. But even if such chips emerge, privacy issues persist: because local processing means your glasses contain your entire life record, and if the device is hacked or lost, the consequences are equally catastrophic. This might be a "contradiction that technology can't perfectly resolve," forcing us back to the level of social contract: are we willing to accept a certain erosion of privacy for convenience? If society as a whole says no, then such products should never see the light of day.

Nilay Patel's comment essentially reminds the tech world: not everything that can be built should be built.