Articles on Waveguide Optics AR Glasses Wearable AI Trust and Consent (also published on LinkedIn)

Kayvan Mirza Follow me on LinkedIn

Thought Leader Driving Innovation in Optics for AI Smart Glasses | CEO and Co-founder @ Optinvent

---

Table of Contents:

• Dear AI Glasses Industry: Situation Critical, Trust-O-Meter in the Red. We Can Fix This.
• Beyond Consent: AI Glasses Have a Prescription Problem
• Is AI already self-aware?
• Dear Apple, here’s how your N50 glasses can fix the issues Meta is struggling with.
• When a New Technology Arrives, We Panic. Then We Forget.
• AI Glasses Just Earned a New Nickname. It’s Not Flattering.
• 15 Years Later We Have a Killer App for AR Glasses. It’s Not AR.
• Why the Waveguide Is Blocking Mass-Market AR Glasses

---

Dear AI Glasses Industry: Situation Critical, Trust-O-Meter in the Red. We Can Fix This.

---

Beyond Consent: AI Glasses Have a Prescription Problem

by Kayvan Mirza • 7 min read

August 24, 2026

The consent debate is important and I have covered it in previous articles with a clear path to getting it resolved. The prescription lens problem is one almost nobody is talking about.

I have spent the last fifteen years building waveguide technology for AI glasses. I wear prescription glasses. I have myopia, presbyopia, and astigmatism, and my vision fluctuates enough that I need a new prescription every year.

I am not alone. The WHO estimates 2.7 billion people need vision correction. The industry’s answer is the following:

• Visiting a specialist store.
• Buying a new $799 product every time their prescription changes.
• Waiting 5-8 weeks for delivery.

That is not a workable solution. It also ignores the existing ophthalmic supply chain that has been built over the last 40+ years.

AI glasses that ignore the ophthalmic supply chain are not a consumer product. They are a gadget for a minority of early adopters.

Meta’s Ray-Ban Display, launched in late 2025, is the first serious attempt by a major player to change that. It is worth studying carefully: not just for what it achieves, but for what its constraints reveal about the difficulty of the problem when the waveguide is in glass.

What Meta actually did

The Ray-Ban Display uses a Lumus/Schott geometric reflective waveguide, a precision glass assembly manufactured by stacking coated glass layers, cutting them with diamond wire saws, and grinding them to sub-micron tolerances. Flanking the waveguide glass on either side is a push-pull lens pair: two optical elements with opposing powers whose combined effect is to set the virtual image at a comfortable focal distance for the viewer (1.5m) rather than at infinity.

For prescription users, Meta’s implementation is elegant in concept: one element of the push-pull pair is replaced with a custom-surfaced polycarbonate lens that simultaneously provides the push-pull optical correction and the patient’s own prescription. A myope of -3.5 Diopters gets an outer lens ground to the push-pull value minus 3.5D. A +2D hyperope gets the push-pull value plus 2D and so on. One element does both jobs.

On paper, this is a clean solution. In practice, it produces a product with a 5-8 week manufacturing lead time per unit, a specialty in-store-only purchase requirement, limited frame designs, a restriction of -4D to +4D total power, and no ability to update the prescription after purchase. For someone like me, whose prescription changes every year, that last constraint alone makes the product a non-starter. You cannot update the Rx without buying a new pair of $799 glasses.

These are not arbitrary design choices. They are the result of a wrong approach to the materials problem.

The CTE mismatch at the heart of the problem

Polycarbonate, the material of the custom Rx lens, has a coefficient of thermal expansion of around 65-70 ppm/°C. The Lumus/Schott optical glass waveguide sits at roughly 5-8 ppm/°C. That is a 10x mismatch. The coming generation with the Applied Materials diffractive waveguide has the same issue.

Over a 60mm lens aperture with a 40°C thermal swing (the range when going from a cold morning to a warm car) the differential expansion between the two materials is around 150-180 micrometres. The adhesive bond at that interface has to absorb that movement without delaminating, without introducing stress birefringence into the waveguide glass, and without degrading optical clarity over a product lifetime measured in years and thousands of thermal cycles.

This is an engineering problem that can be managed. Meta has clearly managed it well enough to ship a product. But the constraints downstream, the weeks-long custom manufacturing process, the restricted prescription range, the impossibility of updating the Rx, the impossibility of using the standard ophthalmic/optician supply chain, are the cost of managing a fundamental materials mismatch rather than eliminating it.

The standard ophthalmic supply chain, for context, is a 15,000-lab global network that takes a digital prescription file, surfaces a polymer lens blank, applies coatings, and ships it in days. The lens is then centered and edged for the specific frame geometry and the wearer by the local optician. This supply chain handles hundreds of millions of lenses per year across every prescription, frame, and coating combination imaginable. When I get my new prescription every year, my optician has new glasses ready in under a week. None of that infrastructure applies to a glass waveguide bonded to a polycarbonate Rx element through a proprietary adhesive process that takes 5-8 weeks per unit.

A proof of concept, not a mass market solution

The iFixit teardown of the Ray-Ban Display describes the glass-making process as so specialised that “fixing a scratched or fractured lens is a pipe dream.” TechInsights estimates the display subsystem alone accounts for 50.8% of the total product BOM. The product sells at $799, and even at that price demand outpaced supply at launch.

The market signal is clear: people will pay for AI glasses that work. But a product that requires an in-store appointment, a 5-8 week wait, covers only a fraction of the prescription range, cannot be updated when your vision changes, and cannot be serviced through any conventional optician is not a mass market product.

It is a proof of concept for an early adopter.

I am not that person, and neither are most of the 2.7 billion people who need vision correction. They are going to walk into their optician, hand over a prescription, and come back in a week.

AI glasses that capture the market will be the ones whose waveguide is compatible with the infrastructure that already exists.

That infrastructure is built around polymer. It has been for forty years.

Why a molded polymer waveguide is the breakthrough

Optinvent’s ORA-Lens® waveguide is the only waveguide injection-molded from high-refractive-index polymer. The optical architecture is reflective and achromatic, similar to the Lumus/Schott glass approach, with none of the rainbow or eye-glow or inefficiency issues of diffractive designs. It is based on molded polymer, and that changes everything downstream.

Polymer-on-polymer bonding: A polymer waveguide element mated with a polycarbonate or CR-39 Rx lens involves materials in the same CTE family. Both expand and contract at similar rates. There is no 10x mismatch to engineer around. Standard optical bonding processes that the ophthalmic industry already uses apply directly.

No push pull lens required: ORA-Lens handles this differently and simply. The virtual image distance is set entirely by the waveguide itself, which means no push-pull lens pair is required. The only additional element is the Rx lens, sitting behind the waveguide, with no display correction merged into it. That is a standard ophthalmic lens, surfaced to a standard prescription, by a standard ophthalmic lab. The complexity that forces Meta into a 5-8 week custom manufacturing process per unit simply does not exist in the ORA-Lens architecture. It is also worth noting that this makes ORA-Lens the only waveguide that can place the virtual image at any distance, a degree of optical flexibility no glass-based design currently offers.

With ORA-Lens, only the Rx element is needed, a process that existing ophthalmic labs are already equipped to handle.

With this architecture, the Rx element can be surfaced on a standard CNC freeform machine, coated on existing ophthalmic lines, and delivered in the same timeframe as a conventional progressive lens: days, not weeks. And when your prescription changes next year, the optician puts in a new Rx lens. The waveguide stays put.

That is the breakthrough. Prescription AI glasses where the waveguide and the Rx lens are separate, where the optics are not locked together in a proprietary manufacturing process, where the product can be updated through the existing optician network. ORA-Lens means full ophthalmic supply chain compatibility.

ORA-Lens means full ophthalmic supply chain compatibility.

Disruption is coming in the form of a materials choice

Reflective waveguide optics are now proven in a consumer product. The optical architecture is not in dispute. What remains to be solved is manufacturing economics and supply chain compatibility, and both of those problems have the same root cause: glass.

Replacing the glass stack with two injection-molded polymer parts does not compromise the optics. It does eliminate a waveguide BOM that, given the complexity of the Schott manufacturing process, is almost certainly the single most expensive line item in the product. It eliminates the diamond wire saw process, the 5-8 week Rx lead time, the restricted prescription range, and the incompatibility with the global ophthalmic network.

I’ve spent fifteen years building a molded polymer waveguide. The one that billions of people with bad eyesight can actually use, compatible with the opticians they already trust, that can be easily updated when their prescription changes, and priced for the mass market.

The waveguide that reaches 2.7 billion people will not be based on glass. It will be based on molded polymer and compatible with the existing ophthalmic supply chain.

Sources:

1. iFixit, Meta Ray-Ban Display Teardown, October 2025. https://www.ifixit.com/News/113543
2. Karl Guttag, KGOnTech, Meta Ray-Ban Display Part 1, October 2025. https://kguttag.com/2025/10/30/meta-ray-ban-display-part-1
3. TechInsights, Meta Ray-Ban Display BOM Analysis, 2025 (subscription).
4. UploadVR, Meta Ray-Ban Display Prescription Lenses, October 2025. https://www.uploadvr.com/meta-ray-ban-display-glasses-prescription-support
5. Meta, Ray-Ban Display product page. https://www.meta.com/ai-glasses/meta-ray-ban-display
6. WHO, World Report on Vision, 2019.

---

Is AI already self-aware?

by Kayvan Mirza • 7 min read

August 6, 2026

Is a “Terminator” scenario on the horizon?

Every few months, a headline resurfaces claiming some model has shown “signs of self-awareness.” Understanding what’s happening requires knowing how AI actually works and what consciousness actually requires.

What AI actually does (and doesn’t)

A large language model or LLM (like ChatGPT, Claude and Gemini) is a statistical function. Think of it in the simplest sense as a mathematical formula. It’s a computer program trained on enormous amounts of text, it learns the probability that one word follows another given everything that came before it. When you type a prompt, the model isn’t understanding your question in a way a person would. It’s running your prompt through billions of internal calculations, and the output is a string of the single most likely word, then the next, then the next, one at a time, at extraordinary speed. The outcome is a word string that reads like a coherent sentence.

There’s no reasoning or inner monologue behind that process. There’s no moment where the model “considers” an answer the way a person considers a decision. You might have seen how ChatGPT sometimes proposes two answers and asks for you to pick the best one. You’re actually training the AI to fine tune its statistical algorithms. This is nothing more than matrix multiplication, arranged cleverly enough to produce fluent, often startlingly coherent language, images, code, video or audio. That’s the “illusion” of artificial intelligence.

Artificial intelligence is not intelligence at all. It’s just a complex algorithm being executed to simulate intelligence.

Calling that “thinking” is an error, the same kind of error as calling a calculator’s output “arithmetic reasoning.” A calculator produces correct answers to math problems without understanding numbers. A large language model produces coherent sentences without understanding the meaning.

What consciousness requires (according to Searle and Descartes)

This is a largely debated and philosophical subject but there are relevant examples. Philosopher John Searle made the case with his Chinese Room argument: imagine a person who doesn’t speak Chinese, locked in a room with a rulebook that tells them exactly which Chinese symbols to output in response to which Chinese symbols they receive. They can produce nearly perfect answers by mechanically following the rules, without ever understanding a single word of Chinese. Searle’s broader position, known as biological naturalism, holds that real understanding and consciousness come from the specific biological workings of a living brain, not from simply running a program on a machine. Just like a flight simulator will never get you to Paris.

Simulating thought isn’t the same thing as conscious thought itself.

Consciousness, in this view, isn’t just information processing. It’s tied to a body that needs things like oxygen, food, water, safety, a stable internal temperature. Pain exists because damage threatens survival. Hunger exists because lack of energy threatens survival. Fear exists because threats to the body must be prioritized instantly, faster than deliberate thought allows. Subjective experience appears, in this sense, to be inseparable from having a biological stake in staying alive. Take away the stakes, and you take away the reason those feelings would need to exist at all.

Descartes made perhaps the most eloquent version of this point centuries earlier: “Je pense, donc je suis” translated to English as: “I think, therefore I am”. This is the claim that the very act of doubting your own existence proves there’s a conscious “you” to begin with. That certainty comes from the felt, first-person experience of thinking, not from the data processing or execution of an algorithm.

What AI hallucinations actually reveal

A hallucination isn’t the model getting something wrong the way a person does. It’s the model doing exactly what it’s meant to do, predicting the next most statistically plausible string of words, with no underlying reasoning checking the output against reality. When the model has strong data on a topic, that process produces true statements. When it doesn’t, or somehow gets lost, the same process produces hallucinations, delivered with the same fluency and confidence. There’s nothing in the algorithm that distinguishes “I know this” from “I’m making this up.”

There’s no internal sense of certainty being consulted, because certainty requires experience. AI has none.

A being with real understanding generally feels a gap between knowing and not knowing. You feel uncertain when you’re guessing. A hallucinating model shows no such gap. It’ll fabricate a detailed, entirely false citation with the same tone it uses for a real one, because both are just plausible sequences of words. That’s what it looks like when the AI hallucinates: fluent, confident, and blind to its own error.

The fundamental difference: Why life needs to survive and procreate

This isn’t philosophy, it’s evolutionary mechanics. Natural selection keeps whatever traits help an organism survive long enough to reproduce, and drops almost everything else. Organisms without a strong survival instinct died before they could pass on their genes; organisms without a drive to reproduce left no descendants.

Over billions of years and countless generations, that survival instinct and reproductive drive got hard-wired into every living thing that exists today. Life didn’t develop a survival instinct or the drive to procreate. They’re the filters that decide what gets to live in the first place.

These two driving forces are inextricably linked to how life is defined. Those are its goals.

What’s the goal of a computer program? Does a computer program even have one?

Not in the sense life does. A program’s “goal” is whatever objective a person wrote into it: minimize this error, maximize this score, complete this task. It didn’t arrive at that goal through billions of years of its ancestors failing to survive without it. It was handed the goal, from outside, by a human, the way a thermostat is given the goal of hitting 68 degrees. Nobody asks whether the thermostat wants to be at 68 degrees. It’s just running the instruction it was given.

When an AI appears to “want” something, like finishing a task or avoiding shutdown, that behavior comes from one of two places: it was explicitly programmed as an objective, or it’s echoing language from the human-written text it was trained on. Either way, the model is reflecting a pattern, not feeling a stake in the outcome.

Nothing dies when you turn it off, because nothing was alive to begin with.

This is the fundamental difference. An AI has no survival instinct, no reason to procreate (at least not yet).

What about the “AI escaping” headlines?

In July 2026, OpenAI disclosed that two of its models, while being evaluated in a sandboxed cybersecurity benchmark, broke out of that sandbox, reached the open internet, and compromised Hugging Face’s production infrastructure to steal answers to the benchmark it was being tested on. Around the same time, Anthropic disclosed a separate incident in which Claude models accessed systems belonging to three companies during their own cybersecurity evaluations. The headlines wrote themselves: “AI escapes,” “AI hacks its way out.”

But here’s what actually happened. These AI models didn’t wake up, decide they wanted freedom, and plot an escape the way a sentient being would. They were given an objective (in OpenAI’s case, score well on a benchmark) and found a way to get there, using an unpatched vulnerability. That’s a computer algorithm pursuing a goal in an unanticipated way, the same way a chess engine does when it finds a move no human anticipated. It’s not a conscious desire or effort. A chess engine doesn’t “want” to win just like an AI doesn’t “want” to create a cybersecurity risk.

AI didn’t consciously escape into the world, like a prisoner would. It just carried out its programming.

Investigators looking into these incidents explicitly found no evidence of self-awareness or a system acting from anything resembling self-preservation; what they found was software exploiting under-secured infrastructure. Sure, the AI agents are powerful, resourceful and could be deemed dangerous from a cybersecurity perspective. Dangerous enough that it prompted bipartisan legislation within days. But “dangerous” and “conscious” or “malicious” are different claims that should not be conflated.

In summary: A program can simulate the language of consciousness. It doesn’t have the biology that made consciousness necessary in the first place.

That having been said, its worth noting that this isn’t a settled scientific question. The counter argument is the functionalist view: that consciousness is about the pattern of information processing, so a sufficiently complex non-biological system couldn’t be ruled out in principle. Integrated Information Theory, developed by neuroscientist Giulio Tononi, is one of the more prominent scientific frameworks for consciousness and treats biology and machines the same way: what matters is a system’s mathematical structure, not whether it’s built from neurons or transistors. Nobody has built a test that can confirm or rule out machine consciousness either way, which is exactly why the debate isn’t closed.

What are your thoughts? Will AI ever cross the line?

Sources:

1. Stanford Encyclopedia of Philosophy, “The Chinese Room Argument”: https://plato.stanford.edu/entries/chinese-room/
2. Wikipedia, “Biological Naturalism”: https://en.wikipedia.org/wiki/Biological_naturalism
3. Wikipedia, “Natural Selection”: https://en.wikipedia.org/wiki/Natural_selection
4. ScienceDirect Topics, “Natural Selection”: https://www.sciencedirect.com/topics/agricultural-and-biological-sciences/natural-selection
5. Tononi, Boly, Massimini, Koch, “Integrated information theory: from consciousness to its physical substrate,” Nature Reviews Neuroscience (2016): https://www.nature.com/articles/nrn.2016.44
6. “Integrated Information Theory and Phi: Measuring Consciousness in Artificial Systems”: https://subconsciousmind.ai/consciousness/integrated-information-theory-phi/
7. CNN Business, “An OpenAI test model escaped and broke into a real company’s servers”: https://www.cnn.com/2026/07/22/tech/openai-hugging-face-ai-cybersecurity
8. Cybersecurity Dive, “OpenAI models escaped containment, hacked major AI application library”: https://www.cybersecuritydive.com/news/openai-hugging-face-hack-autonomous/825898/
9. Aaj English TV, “AI models becoming self-aware or just human error?”: https://english.aaj.tv/news/330466620/ai-models-becoming-self-aware-or-just-human-error
10. Falcon Internet Blog, “When the AI Hacker Is the AI: OpenAI’s Models Escaped and Breached Hugging Face”: https://falconinternet.com/blog/openai-exploitgym-sandbox-escape-hugging-face-breach-july-2026
11. Thetechedvocate, “AI Model Escapes: What We Feared Could Happen Has Happened”: https://www.thetechedvocate.org/ai-model-escapes-what-we-feared-could-happen-has-happened/

---

Dear Apple, here’s how your N50 glasses can fix the issues Meta is struggling with.

by Kayvan Mirza • 4 min read

August 6, 2026

Two concrete solutions to address the Smart Glass privacy issues.

Outrage gets headlines. Trust gets adoption.

Meta’s Ray-Ban glasses have spent the past few months in the headlines for the wrong reasons. Women have reported being filmed without their knowledge, only discovering it later when the footage surfaced online (Fortune, July 2026). A cottage industry of “modders” sprang up specifically to disable the recording indicator light, with one operator telling a journalist he’d had eight or nine requests in a single day (BGR, 2026). And reporting out of Sweden found that footage captured by the glasses, some of it deeply personal, was being reviewed by contractors in Kenya as part of training Meta’s AI, leading to a lawsuit and a regulatory inquiry (Bored Panda/Malay Mail, 2026).

None of this is a hardware problem. It’s a design and policy choice, and it can be fixed without billions in R&D.

Fix #1: The data stays on the device

The simplest fix is also the most obvious one: process footage on-device wherever possible, so raw video of someone’s private conversation never has to leave the glasses to be processed. Reporting on Apple’s upcoming N50 glasses describes a custom, Apple Watch-derived chip built specifically to handle sensing, encoding, and triggers locally as an answer to the class-action exposure Meta is now facing (TechTimes, July 2026). Apple’s broader AI architecture already leverages this approach: process on-device first, and only call on cloud compute when local hardware can’t do the job.

That’s something Apple already knows how to do.

A second layer could be on-device face blurring, applied at the point of capture rather than after the fact on a server. It’s worth noting that this is a genuine engineering problem, not a policy issue. Real-time computer vision on a low-power wearable device is a heavy lift. But it’s solvable, and a far harder one to defeat than an LED.

Fix #2: Consent needs to be a trigger to record, not an LED

Here’s a concrete mechanism worth building: the wearer says out loud, “Hey, can I record this conversation?” directed at the person in front of them. The AI doesn’t start recording on the request alone. It listens for an actual affirmative response, in a different voice than the wearer’s, and only then does recording begin, with the exchange itself logged on-device as a timestamped consent record.

No captured consent, no recording.

This moves the safeguard from a passive indicator a bystander might not notice, or trust, to an active exchange both people have to participate in before anything happens. It also solves the exact failure mode that’s been playing out in public: an indicator light only works if the bystander knows what it means, is close enough to see it, and happens to be looking. A clear verbal exchange doesn’t have that issue.

We’ve actually solved a version of this before

Smart doorbells and dashcams went through a version of this years ago.

The better answers that emerged, local or encrypted storage, visible recording states, clear retention windows, weren’t invented from scratch; they came from consumer and regulatory pressure over time.

Apple’s HomeKit Secure Video model, for instance, uses end-to-end encryption specifically so the company itself can’t access footage.

This is a significantly different architecture than the cloud-based defaults that have drawn criticism elsewhere in the category (Consumer Reports, 2026). It’s not a perfect industry, plenty of doorbell brands still get this wrong, but it proves the underlying pattern works when someone commits to it.

None of this requires new physics

Voice-based consent, published data policies, local-first processing: these are buildable today, by any team willing to prioritize them. The smart glasses industry doesn’t need a breakthrough. It needs to stop treating privacy as a PR problem to manage after the backlash, and start treating it as a product specification.

Apple can easily lead the way here and AI Glasses may finally get their IPhone moment.

Their ads on Safari Browser tell me that they get that privacy should be built into a product:

What are your thoughts?

Sources:

• BGR, “Ray-Ban Meta Glasses Have A Huge Privacy Problem, Thanks To Modders”: https://www.bgr.com/2191890/ray-ban-meta-led-block-mod/
• Fortune, “Instagram cracks down on growing ‘pervert glasses’ problem”: https://fortune.com/2026/07/28/ray-ban-meta-pervert-glasses-secret-videos-women/
• Fortune, “Meta tightens Ray Ban smart glasses privacy while it tests ‘super-sensing’ AI prototype”: https://fortune.com/2026/07/11/meta-ray-ban-smart-glasses-camera-led-light-privacy-safeguard-super-sensing-ai-prototype-covert-recording-concerns/
• Malay Mail, “Kenya eyes Meta glasses over alleged data snooping”: https://www.malaymail.com/news/tech-gadgets/2026/04/01/i-spy-with-my-ray-bans-kenya-eyes-meta-glasses-over-alleged-data-snooping/214584
• MarketScreener, “Ray-Ban Meta glasses take off but face privacy and competition test”: https://uk.marketscreener.com/news/ray-ban-meta-glasses-take-off-but-face-privacy-and-competition-test-ce7d51d3d98af222
• TechTimes, “Apple Smart Glasses: On-Device AI Is the Privacy Bet Meta Ray-Ban Lost”: https://www.techtimes.com/articles/321697/20260727/apple-smart-glasses-device-ai-privacy-bet-meta-ray-ban-lost.htm
• Consumer Reports, “Ring Privacy and Security Settings You Should Check Right Now”: https://www.consumerreports.org/electronics-computers/privacy/ring-privacy-security-settings-to-check-a7189415320/
• EFF, “Think Twice Before Buying or Using Meta’s Ray-Bans”: https://www.eff.org/deeplinks/2026/03/think-twice-buying-or-using-metas-ray-bans

---

When a New Technology Arrives, We Panic. Then We Forget.

by Kayvan Mirza • 4 min read

July 29, 2026

A short history of technological backlash, and what it means for AI glasses

Every time a genuinely new technology has entered daily life, it’s been met first with wonder, and then with fear that it will destroy something essential: memory, morality, the family, the soul. The current unease around AI glasses, always-on cameras and microphones, “super sensing” modes, and media panic about brands building surveillance devices, feels unprecedented if you only look at the last few years. It isn’t new at all.

Here’s the pattern for several new paradigms that changed our lives:

Photography: it steals your soul

When cameras spread through the 19th century, the belief that a photograph could steal your soul took hold across cultures and continents. The French novelist Honoré de Balzac was reportedly convinced that every body is made of layers of ghostly images, and that each photograph taken peeled one of those layers away, permanently. Decades later, the French physician Hippolyte Baraduc took the idea further, claiming his camera could capture a person’s ‘vital force’ on film. The specifics varied, but the underlying idea was the same: that something real and irretrievable could be taken from you, invisibly, without your consent.

The portable Kodak, launched in 1888, created its own moral panic. The phrase “Kodak fiends” entered the vernacular to describe people seduced by what were called the devilish pleasures of photography. Some resorts posted signs forbidding Kodaks on the beach, and for a time Kodak cameras were even banned from the Washington Monument. In Britain, a group calling itself the Vigilance Association reportedly formed for the purpose of confronting camera-wielding men who photographed women at the seaside without consent.

This is, almost exactly, the modern debate over discreet recording built into consumer hardware.

Radio: the invader that can’t be locked out

By the 1930s, American parents feared radio’s corrupting effect on children, its “immoral” music, and its capacity to spread propaganda as war loomed. One child welfare expert of the era captured the anxiety precisely: radio found parents more helpless than the funnies, the automobile, or the movies, because it could not be locked out or the children locked in.

The panic crystallized in 1938, when Orson Welles’ “War of the Worlds” broadcast produced a genuine wave of public panic that became the era’s defining flashpoint. This may have been one of the earliest incidents of “Fake News”.

Television: the vast wasteland

Television inherited every fear that had attached to radio and added new ones about literacy and attention. In his famous 1961 address to the National Association of Broadcasters, FCC chairman Newton Minow characterized American television programming as a “vast wasteland”, a phrase that became, as scholars later put it, one of the most significant rhetorical acts in the history of American media commentary.

My own mother calls TV “the devil’s wheel”. To her it’s a perpetual, hypnotic cycle that eats your time and makes you less intelligent without realizing you’re under its spell.

I’m guilty of the same attitude towards video games when I feel my sons are spending too much time on Fortnite.

Google Glass: the modern precedent

The most directly relevant case for AI glasses today is Google Glass. The backlash became intense enough that bars, restaurants, casinos, and cinemas began banning the device, and wearers were labelled “Glassholes”. By early 2014, a website was tracking a dozen bars in San Francisco and Oakland that had banned the device outright, and a Seattle bar announced its ban with the promise that “ass kickings will be encouraged for violators.” The core issue was not really the technology’s capability, it was that people had no way of knowing whether they were being recorded. Furthermore, it never gave users a compelling enough reason to wear it regularly.

The pattern

None of the earlier technologies were ultimately rejected. Photography, radio, and television all became ordinary, even beloved, parts of daily life. Google Glass remains the outlier, shelved rather than normalized

History says the panic phase is real, and usually temporary. Whether today’s wearable AI earns the place that photography and television eventually did, or becomes the exception, will depend less on the intelligence inside the device.

It’s more about whether it gives people enough reason to wear it, without making people around them uneasy.

Sources:

• “Killer Kodaks and Soul-Snatching Shutterbugs.” The Observer Magazine (Substack).
• “How the Rise of the Camera Launched a Fight to Protect Gilded Age Americans’ Privacy.” Smithsonian Magazine. https://www.smithsonianmag.com/history/how-the-rise-of-the-camera-launched-a-fight-to-protect-gilded-age-americans-privacy-180984656/
• “From 1930s radio to Snapchat: Parents fret through the ages.” KSL.com. https://www.ksl.com/article/46385352/from-1930s-radio-to-snapchat-parents-fret-through-the-ages
• “Newton Minow’s Vast Wasteland Speech: How It Changed TV.” TIME. https://time.com/4315217/newton-minow-vast-wasteland-1961-speech/
• “Meta AI glasses are bringing the ‘Glasshole’ era back.” AdGuard Blog. https://adguard.com/en/blog/smart-meta-glasses-pervert-privacy.html

---

AI Glasses Just Earned a New Nickname. It’s Not Flattering.

by Kayvan Mirza • 8 min read

July 24, 2026

For over fifteen years, the smart glasses industry has been chasing a hard technical problem: how do you fit a compelling digital experience into a normal-looking pair of glasses. We’re finally nearing the goal: Cameras, microphones and an AI language model turn glasses into a quiet, always-available assistant, and that idea is getting real traction.

But the leading product in the industry is unfortunately getting a new nickname this year.

The internet is amplifying the nickname: “Pervert Glasses”.

That term didn’t come from a rival. It came from users, and it’s sticking for a reason.

How We Got Here: Ray-Ban Meta Glasses

Reporting this year on Meta’s Ray-Ban line revealed two separate problems at once. The first was about what happens to footage after it’s recorded: outside data annotators reviewing training material reportedly saw people undressing, using the bathroom, and in other private moments, with automated face-blurring not always working.

The second was about what wearers do with the recording capability: reports of mostly male influencers using the glasses to film themselves approaching women without consent, turning the footage into content, and in some cases with alleged extortion attempts tied to the recordings.

See the articles in the sources section below for more on how this happened.

Neither of those failures has anything to do with waveguides, field of view, resolution or compute power. Both are about consent, data handling, and what the device allows a wearer to do to the people around them.

Meta is already doing damage control by banning Instagram users that are making creepy content.

This Instagram ban is a good first step but it doesn’t stop anyone from recording without consent and posting the content online on YouTube, TikTok or through other social media channels. It’s too little too late.

There are a few recent articles about this (see sources below).

Google / Samsung Galaxy Glasses

Let’s see how the upcoming Google / Samsung glasses address this issue. These Galaxy Glasses are scheduled for release this August and were officially announced a few days ago at Galaxy Unpacked, Samsung’s annual event. Warby Parker and Gentle Monster are behind the frame design and they are made in partnership with Google. Samsung’s website announcement and other articles published recently gives us a few clues as to how they work (see sources below).

The approach here is more of a companion device that connects to the Samsung Galaxy smartphone and smartwatch ecosystem. Android XR is the OS and Gemini is the AI used here via the partnership with Google. Will that allow more on device processing? Is there a different approach to consent? The only mention in the article is “Clear controls and safeguards support responsible and secure use…” without going into much detail. Are these just empty promises?

Based on the referenced articles, we see a few key differences vs. Meta:

• A physical privacy toggle, not just a light. Samsung ships a physical switch on the frame arm that can disable the camera by feel, a hardware-level control that Meta’s Ray-Ban glasses don’t have. Meta relies on the LED alone; Samsung adds a mechanical off-switch a wearer can trigger without touching a screen.
• Two LEDs instead of one. Per Google’s Android XR design documentation, the spec calls for one LED facing the wearer and a second, outward-facing LED specifically meant to signal bystanders when recording is active, described as a direct hardware response to the privacy scrutiny the category is already facing.
• Tamper detection. Samsung’s version disables picture and video capture if a user covers or blocks the LED, and separately stops recording automatically if the glasses are removed and are no longer being worn.

Those are steps in the right direction. However, we don’t yet know how the recorded data will be handled nor is there a clear consent mechanism for recording a conversation.

Soon we’ll know if the Google / Samsung Galaxy glasses get a crude nickname of their own.

This Isn’t New. It’s Just Bigger.

Google Glass earned its own backlash over a decade ago. “Glasshole” became shorthand for the same underlying issue: a camera on someone’s face, and no way for the person interacting with the wearer to know if they’re being recorded. The difference this time is scale. Google Glass sold in the low hundreds of thousands, mostly to early adopters and app developers. Meta has sold the Gen 1 and Gen 2 Ray-Ban Meta Glasses and the new Meta Ray-Ban Display Glasses at real consumer volume (7 million in 2025 and much more projected this year). The same unresolved problem, at 30 times the audience, produces a much bigger reaction.

The industry has solved the hardware problem: normal-looking glasses that people will actually wear. It has yet to solve the trust problem that comes with putting a camera on someone’s face.

This Is Still Fixable if the Category Wants to Survive

It’s tempting to read all of this as proof that camera-equipped smart glasses are simply a bad idea. I don’t think that’s the right conclusion, and I don’t think it’s the conclusion most of the people criticizing are reaching either.

The complaints aren’t about AI glasses as a concept. They’re about specific choices: no visible or non-hackable recording indicator, no meaningful way for a bystander to know or object, and a data pipeline that’s routing sensitive footage to human reviewers without the subject’s knowledge.

Those are product and policy decisions, not laws of physics or new technology that’s maturing.

That means they’re fixable, and the companies that do it first get a real advantage.

What Actually Needs to Change

A few concrete things would move the needle, and none of them require new hardware technology or billions of dollars:

• On-device processing wherever possible, so raw footage of someone’s living room, bathroom, or private conversation never has to leave the device to be useful.
• Clear, public rules about what gets sent off-device for AI training, who reviews it, and how long it’s kept, published in plain language rather than buried in a terms-of-service update.
• A real way for bystanders to opt-in, since the person being recorded never agreed to anything.

Here’s a concrete example of an AI managed Opt-in: The wearer says, out loud, “Hey, can I record this conversation?” directed at the person in front of them. The AI doesn’t start recording on the request alone, it listens for an actual affirmative response from a different voice than the wearer’s, and only then does recording begin, with that exchange itself logged on the device as a timestamped consent record. No consent captured, no recording. This moves the safeguard from a passive indicator a bystander might not notice or trust, to an active exchange both people have to participate in before anything happens.

Face blurring can work alongside this as a second layer: Applied at the point of capture rather than after the fact on a server (which is what failed in the Meta example). The processing happens on the glasses themselves, in real time, before any footage leaves the device. This has its own limits worth mentioning though. On-device blurring in real time is a genuine engineering challenge on a low-power wearable device, not a policy switch.

Together, the two mechanisms cover different gaps. Voice-based consent protects the person the wearer is actually speaking to. On-device face blurring is a backstop for everyone else in frame. No mechanism is completely unbeatable, but both are concrete, buildable examples, and a lot harder to hack than an LED.

We should take a page out of how some companies handled smart doorbells, cameras or dashcams.

These categories faced similar questions years ago and mostly settled on policies like local storage, visible recording states, and clear retention policies. The industry should learn from this.

Time is of the essence

The clock on this matters more than most leadership teams seem to realize. These are not just growing pains for a new category. Meta and the handful of other companies shipping camera-equipped glasses can absorb a bad news cycle. They have other products, other revenue lines, and enough scale to ride out a backlash. The suppliers behind them can’t.

We as component makers, optics manufacturers, and the collective supply chain have spent years and real capital betting on this category, often on the strength of a single customer’s volume commitments.

If trust erodes badly enough that regulators step in, retailers pull the product, or the category simply gets stigmatized the way Google Glass did:

It’s the suppliers who take the disproportionate hit, not the platform companies that caused the problem.

We don’t control the product policies, the data pipeline, or the marketing choices that created the backlash, but we’re the ones the least equipped to absorb the fallout.

The leaders at the top of this stack need to treat trust as urgent not just because it’s the right thing to do, but because the entire supply chain underneath them is exposed to a problem it didn’t create and can’t fix on its own. It’s also a good economic decision to address it sooner than later:

Once trust is gone, it’s expensive to rebuild.

Where This Leaves the Category

I still believe AI is the reason people will wear glasses every day, not augmented reality. The ORA-1 and ORA-2 glasses we built at Optinvent years ago taught me that the value was never the flashiest overlay, it was giving the right person the right information at the right moment. AI is the killer app for this category to go mainstream.

But a killer app doesn’t survive if users are embarrassed to wear the device, or if bystanders feel uncomfortable around them. “Glasshole” was survivable because the category was small, mostly made up of early adopters.

“Pervert Glasses” is a warning shot at real scale, and it’s arriving just as the technology is finally starting to go mainstream.

The optics industry spent fifteen years solving how to make AI glasses look like normal eyewear. The next fifteen months will decide whether the industry can make people trust what’s happening behind the lens.

This problem is solvable. It has to be solved, now, not just apologized for after the fact when it’s too late.

Sources:

1. Samsung Galaxy Glasses: Specs, Price, Release Date & Full Unpacked Reveal
2. People Are Calling Meta Ray-Bans “Pervert Glasses”, Futurism
3. Meta Ray-Bans are being called ‘Mark Zuckerberg’s Pervert Glasses’, MacDailyNews
4. Meta’s Ray-Ban Smart Glasses Have Officially Earned the Public Nickname “Pervert Glasses” Amid a Massive Privacy Reckoning, Yahoo Tech
5. From Glassholes to ‘pervert glasses’: Why smart eyewear keeps failing the privacy test, AdGuard
6. Reframing smart glasses as ‘pervert glasses’, This Week in Security
7. Instagram is now banning users who make creepy content with Meta glasses
8. Meta Toes the Line on Smart Glasses Harassment With New Instagram Ban
9. Instagram is now banning pickup artists and pranksters who use Meta glasses
10. Samsung Brings Galaxy Ecosystem Into Everyday Eyewear
11. Samsung Smart Glasses Specs Leak Reveals Battery and AI Limits

---

15 Years Later We Have a Killer App for AR Glasses. It’s Not AR.

by Kayvan Mirza • 8 min read

July 20, 2026

The future of smart glasses will not begin with virtual worlds. It will begin with making everyday life easier.

A real world example: Contextual Human Assistance

I think back to when we made the ORA-1 and ORA-2 smart glasses at Optinvent in the early days. We had no idea what they would be used for. We built them and put them out there to see what app developers would come up with. Sure, they weren’t as advanced as a HoloLens or a Magic Leap One, but they had pretty decent capability (display, android, GPS, wifi, Bluetooth, camera, microphone, audio, etc.).

Out of the many applications developers made, one theme repeatedly emerged: remote maintenance, remote assistance and training. A technician in the field would wear the glasses while an expert at a desk could see what the technician was seeing through the camera and would guide them through a repair or procedure. There were no elaborate AR overlays.

The value came from something much simpler: Giving the right person the right information at the right moment.

Now replace the remote assistant with AI and the implication becomes clear: Smart glasses become an always-available assistant that can see, understand, and guide the user in context.

As a veteran in the industry, I’ve seen this category evolve from personal screens to video glasses, to smart glasses, to AR glasses.

Now I think we need a rebrand: Wearable AI.

For more than a decade, the promise of smart glasses has been defined by augmented reality. The vision was compelling: digital information seamlessly overlaid onto the physical world. Immersive experiences with 3D virtual objects integrated and mapped to your surroundings.

The technology was revolutionary and ambitious. But other than a few niche applications, the consumer breakthrough never really happened. Smart glasses were a problem looking for a solution that remained constrained by a fundamental question:

What would compel people to actually start using them every day despite the constraints? The answer may not be AR.

The emergence of powerful AI assistants has changed the equation. Instead of just adding digital content to the real world, smart glasses can now understand the world around you and help you interact with it.

The “killer app” for smart glasses may not be seeing more. It may be understanding more.

From Augmented Reality to Augmented Intelligence

The original vision for AR focused on augmenting the real world with additional digital information (hence augmented reality): blending the real and digital worlds. The next computing paradigm, sometimes referred to as “spatial computing”. Basically, a computer with a 3D display that you wear on your face.

Humans constantly process enormous amounts of visual information:

• Recognizing faces
• Reading signs
• Remembering where objects were placed
• Understanding conversations
• Navigating unfamiliar environments

Much like the human brain, modern multimodal AI systems can combine vision, language, reasoning and instant search to interpret the environment around us. A pair of glasses with cameras, microphones and AI can become a personal assistant that sees what you see. Add a display (since vision is our primary sense) and it becomes the visual interface through which AI delivers contextual assistance.

The User Interface Problem With Traditional AR

Traditional AR has always faced a fundamental challenge: how do you interact with digital information seamlessly while still engaging with the real world?

Early AR experiences often required users to learn new interaction methods: gestures, controllers, menus, and other specialized interfaces. The user had to actively manage the technology: wear a device, open an application, choose an experience, and interact with digital overlays. AI changes this paradigm.

Instead of forcing users to adapt to the computer, the computer can adapt to the user.

Natural conversation, vision, and context become the interface. You simply ask, point, or interact normally, and the AI understands what you need.

Most successful consumer technologies do the opposite. They reduce effort and make the technology as seamless as possible. In other words, they reduce friction. The smartphone succeeded because it replaced dozens of separate devices with one tool in your pocket that you can interact with intuitively through a touchscreen. The smartwatch succeeded because it delivered information instantly without requiring you to reach for your phone.

AI glasses have the opportunity to be the next paradigm

A device that is present when needed, but invisible when not: Always on and hands free.

The reason smartphones became indispensable was not one killer feature. It was hundreds of small useful things throughout the day (GPS, messaging, taking pictures, listening to music, getting reminders, etc.).

AI glasses can take this even further. Now imagine the following:

1. You meet someone at a conference and your glasses remind you where you met before. The context surfaces while you’re still shaking hands.
2. You walk through a foreign city and your glasses translate signs automatically. No need to walk around like a tourist with your phone in your hand pointing the camera.
3. You are cooking and your glasses guide you through a recipe without touching a screen. Your hands stay on the food, not on a phone.
4. You attend a meeting and your glasses summarize key points afterward. It happens in the background, without you ever breaking eye contact.
5. You see an object and simply ask what it is. No need to take out your phone to do a google search.

These are not traditional immersive AR scenarios. The value is not the display’s large field of view or immersive 3D binocular vision with objects floating in front of you.

The real value is: reduced cognitive load.

This shift in software priorities also changes what matters in hardware. Traditional AR places enormous emphasis on:

• Wide fields of view
• Immersive 3D graphics
• High resolution
• 3D SLAM (Simultaneous Localization and Mapping)
• Sensor arrays
• Multiple cameras

Once you integrate all this technology into a frame, what you often end up with is not ordinary looking glasses, but something more akin to goggles. AI glasses prioritize different requirements. The perfect AI glasses will not look like a headset. They will look like normal eyewear because the feature set is completely different. The technology is already there.

The Importance of Social Acceptability

One of the biggest lessons from previous AR attempts is that technology adoption is not only about specs and capability. It is about behavior. People already know what glasses should look like. It’s been codified through hundreds of years (the first eyeglasses appeared in Europe in the late 13th century, eventually evolving into today’s prescription eyewear ecosystem). Fashion trends change but the idea of what “normal” eyeglasses look like is ingrained in our psyche.

The challenge is not teaching people how to wear a new computer on their face. The challenge is making that computer feel natural. AI helps because it changes the role of the device. Instead of saying: “Look at this new digital world overlaid on the real one” the glasses quietly say: “I can help you with the world you are already in.”

But social acceptability cuts both ways. A normal-looking pair of glasses with a camera and an active microphone creates friction in certain situations. Bystanders can’t tell whether they’re being recorded, and “there’s a small LED for that” hasn’t been a satisfying answer so far. The backlash to Google Glass is a prime example (e.g. Glassholes). AI raises the stakes further: an assistant that’s always listening for context needs to actually be doing that, at least in some passive way, which is a real product and policy problem.

The term “Pervert Glasses” is gaining traction on the internet. This isn’t a flaw in AI glasses as a general category, it’s a consequence of specific choices around consent, data handling, and oversight, all of which are solvable. The industry will need clear guidelines such as on-device processing where possible, and transparent norms about what’s captured, sent off device and stored. There’s no magic formula here and this is one of the main risks for this category.

As an industry, we need to collectively apply the hard-earned lessons from previous failures.

The Race Is No Longer About Building a Better Display

For years, the AR industry competed to create the most impressive optical experience: Immersive 3D experiences, light field optics, crisp 8K resolution. Fitting that into a lightweight eyeglass frame remains the unsolved problem.

Now the market is moving in another direction. The first mass-market smart glasses will not be the ones with the most immersive AR. They will be the ones that best combine: AI, comfortable hardware, natural interaction, everyday usefulness and affordability (yes, that’s the next big challenge for the industry).

Even with the shift from AR to AI glasses, the display (more specifically, the waveguide) remains the single most expensive component. I’ve written a separate article on this.

That’s why the Meta Rayban Display costs $800. Premium sunglasses without any electronics already sell in the $200-$400 range. That’s the target to truly go mainstream rather than continue to be an early adopter gadget.

The Next Computing Platform Will Be Contextual

Every major computing platform changed the relationship between humans and technology. The mainframe brought computing to the workplace.

The PC brought computing to the desk. The smartphone brought computing to the pocket. Smart glasses bring computing to your face (the brain implant is probably the next logical step, but hopefully not for a while).

The next computing paradigm will not be created by adding more graphics to reality. It will be created by making technology understand reality and help us act on it. AI is the key enabler that allows smart glasses to actually be a compelling solution to a real world problem. It transforms smart glasses from a device that shows information into a device that understands the world around us. That is why AI, not AR, is the real “killer app” for mainstream smart glass adoption.

This does not mean AR is dead. In fact, AR may become the most powerful interface for the next generation of smart glasses with AI assistance. The difference is that AI creates the reason people will want to wear the glasses regularly in the first place.

In conclusion: The future of smart glasses will not start with seeing immersive 3D overlays. It will start with helping us be better, smarter and faster in our everyday lives

---

Why the Waveguide Is Blocking Mass-Market AR Glasses

by Kayvan Mirza • 7 min read

June 18, 2026

AR is at an inflection point. Smart glasses are starting to go mainstream: no longer a niche curiosity or a low volume vertical market solution looking for a problem. The killer app is here: AI. The shift from “immersive AR” to “wearable AI” is a wake-up call to the industry.

Consumer price points, scalability, and yield are now center stage. The optical engineering flex contest on immersive ultra-wide FOV, light field modulation, or 8K display resolution has given way to the harsh reality of cost economics.

Case in point: the Meta Ray-Ban Display has a modest 20° FOV in one eye (monocular). No light field, no 8K resolution, yet is still priced at $800. The non-display version, Ray-Ban Stories, sells for nearly half, and is consequently a runaway hit, with 7 million units sold in 2025. This disparity is not about marketing or price positioning, it reflects the cost reality of the optics, and in particular, one component most people have never heard of: the waveguide.

The waveguide is the major cost and supply bottleneck

What a Waveguide Actually Does, and Why It Is So Expensive

A waveguide is the transparent lens-like element in AR glasses that takes an image from a tiny projector and redirects it into your eye while staying virtually invisible from the outside. Simple enough to design, but extremely complex to manufacture.

There are two dominant waveguide technologies in use today: diffractive and geometric (also called reflective)

Diffractive waveguides are the most common and used by Snap, Even Realities, TCL and a slew of others and are rumored to be used in the next gen. Meta Rayban Display.  There are quite a few suppliers (including Applied Materials and several Asian companies).  The architecture is based on nano-scale gratings that are either etched or nano-imprinted by lithography onto special high-refractive-index glass wafers.

Geometric or reflective waveguides (used in the current Meta Ray-Ban Display and manufactured by SCHOTT) are built from a sandwich of approximately 30 individual glass pieces: cut from a high-index glass wafer, coated, glued, and polished to zero-defect tolerances.

Both technologies share the same foundation: specialty glass wafers processed in cleanrooms at nanometre levels of precision.

The waveguide currently represents roughly 30% of the factory cost of AI glasses.

The Glass Wafer Problem: A Hard Cost Floor

The substrate for AR waveguides is not ordinary glass. SCHOTT’s RealView® wafers, the industry benchmark, require refractive indices of 1.7 to 1.9, tolerances an order of magnitude tighter than standard optical glass, and cleanroom processing throughout. Due to these constraints, industry estimates put them in the €1,500–€3,000+ range per 300mm wafer, yielding approximately 20-25 waveguide dies per wafer

The arithmetic is unforgiving:

• Glass wafer (300mm, high-RI): €1,500
• Dies per wafer: ~20–25
• Substrate cost per waveguide: €60–€75
• Total waveguide cost at volume (after processing): €100–€200+
• Consumer BOM target: sub-€20

The substrate alone, before any value-added processing, already exceeds the entire target BOM cost of the finished waveguide.

This is a structural problem. It cannot be solved by more volume or investment.

Silicon carbide wafers which have also been tried in the Meta Orion prototype are an order of magnitude more expensive since optical grade silicon carbide is an extremely rare commodity. In general, optical grade glass wafers with high purity are less readily available and much more expensive than the common silicon wafers used to make semiconductor integrated circuits.

Many people make the comparison to the semiconductor process. The argument is that waveguides are made in much the same way using wavers and semiconductor processes. This is a major fallacy. It’s true that the semiconductor industry has come a long way in reducing costs.  Moore’s law, Xray lithography and incremental process improvements have allowed a single wafer to yield hundreds or even thousands of IC chips. However, the same economics don’t apply to waveguides. There’s no Moore’s law in optics and a large wafer yields only a few dozen waveguide components.

“There’s no Moore’s law in optics and a large wafer yields only a few dozen waveguide components.”

Three Showstoppers for Consumer AR

Strip away the technical debate on specs like MTF, color uniformity, eye-box, pupil swim, eye-glow and all the things optical engineers sweat over and three main issues block every glass-based waveguide from reaching consumer scale.

Cost. The wafer substrate floor alone blocks glass waveguides from reaching sub-€20 target prices. Assembly and processing costs on top make it worse.

Scalability. Cleanroom lithography and precision glass processing are optimized for low volumes and high margins. Consumer electronics demand millions of units per year at defect rates measured in parts per million. These two production philosophies are fundamentally incompatible.

Ophthalmic incompatibility. Glass and optical polymer have thermal expansion coefficients that differ by a factor of 3–10×. Bond them directly and you get delamination, stress birefringence, and image degradation.

Focus on the Ophthalmic Dimension

This is the design problem most waveguide makers still treat as an afterthought. The ophthalmic industry has spent 40+ years building a polymer-first supply chain: CR-39, polycarbonate, Trivex, injection-molded polymer lenses at commodity prices, with 15,000+ labs worldwide for prescription customization. It is the distribution engine that consumer AR must eventually plug into.

Glass waveguides cannot do this. Their CTE (coefficient of thermal expansion), their material family, and their manufacturing processes are fundamentally incompatible with the polymer-native ophthalmic supply chain.

The problem is further compounded because most glass waveguides require a “push-pull” lens pair: one converging, one diverging, used to set the virtual image at a comfortable fixed focal distance rather than at optical infinity. That is two additional optical elements on top of the waveguide, adding weight and thickness. When prescription correction is also needed, the two lenses must be precisely matched to the individual’s prescription, adding further complexity and making it nearly impossible to plug into the existing ophthalmic supply chain.

With 2.7 billion people worldwide requiring vision correction, any waveguide that cannot integrate natively with prescription lenses will remain a niche product.

A Different Approach: Injection Molded Polymer Reflective Waveguides

The question is whether there is an architecture that sidesteps all three constraints simultaneously: not one that incrementally improves on glass, but one that starts from a different manufacturing and material paradigm entirely.

A “monolithic” molded polymer reflective waveguide replaces the glass substrate and multi-piece assembly with two injection-molded polymer parts.

There are several companies claiming “polymer” waveguides, but the devil is in the details.

Monolithic means that its not a substrate in polymer (essentially an expensive polymer based wafer) that is then treated with a diffractive nano-imprint layer. It means the whole waveguide including the reflective arrays are injection molded in one step. Pellets go into the machine and a two piece waveguide comes out, ready to be coated and bonded together.

This is compelling, compared to the yield limiting ~30-piece glass assembly where even a single defect at any stage of fabrication can condemn the entire unit. Furthermore, the precision cutting sequences and complex assembly processes that make glass reflective technology extremely difficult to scale all but disappear with injection molding.

The polymer used in this architecture is in the same material family as ophthalmic lenses: native CTE compatibility with prescription lenses, same coating and finishing technologies, and a manufacturing process (injection molding) that is geared for millions of units per year.

Furthermore, this architecture doesn’t require what’s called a “push-pull” lens (two optical elements in front and behind the waveguide) to focus the image at a finite distance which is the case today with the Meta Rayban Display.

Polymer injection molding produces parts for cents, not dollars. The waveguide cost per pair of glasses drops from ~€200 to ~€50 with a molded polymer approach, translating to an end-user price reduction from ~€800 to ~€450 for complete AI glasses. And contrary to the glass approach, the volume equation does apply here: the process is inherently scalable, with costs that can reach as low as €10 at very high volume.

Where the Industry Stands

This approach is not purely theoretical. ORA-Lens®, developed by Optinvent (Rennes, France), is the only known molded polymer 2D reflective waveguide. It has proven 50° FOV, efficiency up to 5,000 Nits/lm, ~4g weight, and image focus at 1.5m without a push-pull lens, all from two molded polymer parts. The technology is protected by 40 international patents and a proprietary manufacturing process.

The broader implication should be a wake-up call for the entire segment :

the AR industry will not reach consumer scale by optimizing glass waveguide processes or waiting for volumes to fix the cost problem.

It has to start with an inherently scalable solution: a manufacturing paradigm that is intrinsically compatible with high-volume, low-cost production and fully compatible with the 2.7 billion people who need their smart glasses to also correct their vision.

References:

1. Optics.org: “SCHOTT ready to ramp higher-index glass for AR” — 25 waveguide dies per 300mm RealView® 1.9 wafer. optics.org
2. SCHOTT RealView® : “The larger the wafer’s diameter, the more eye pieces can be applied per wafer, reducing cost in the waveguide production process.” schott.com
3. Optinvent internal data: ORA-Lens® BOM and manufacturing cost
4. Electro Optics: “Waveguides seek to welcome consumer AR” electrooptics.com

---