The Escalator Problem: Why Meta's Smart Glasses Might Finally Work

Most attempts to build AI wearables fail because they ask users to adopt a new device and a new behavior at the same time. Alex Himel, Meta's VP of Wearables, argues the opposite: start with something people already wear, make it a little better, and let the AI become the reason they never take it off. The real insight is that glasses don't have to be smart to win. They just have to be worth wearing. For product leaders, hardware investors, and anyone betting on ambient AI, this conversation shows why the conventional wisdom about killer apps and form factors gets cause and effect backwards. The advantage goes to companies willing to invest through the boring years, not the ones chasing flashy demos.

The Escalator That Never Breaks: Why Glasses Don't Need to Be Smart to Win

Himel's most revealing moment comes early, when he reframes the whole category question with an analogy:

"An escalator never breaks, it becomes stairs. And so similarly the glasses, even if they run out of battery and or they're just turned off, there's still something you would wear if you were optical glasses or you were sunglasses."

That's the opposite of how most tech companies think. They build devices that are useless when the battery dies, disconnected from the network, or worse, when the AI is not good enough yet. A Humane Pin or a Rabbit R1 with a dead battery is a plastic brick. A pair of Ray-Ban Metas with a dead battery is still a pair of Ray-Bans. That baseline utility creates a feedback loop most startups cannot access: the more people wear them for reasons unrelated to AI, the more chances the AI gets to prove itself useful. Patience pays off.

But there is a deeper effect on user psychology over time. Himel's team noticed that engagement and retention are highest when glasses have clear or transition lenses, meaning people wear them indoors where they are more likely to encounter situations that benefit from AI. The hardware creates the conditions for the software to become indispensable. Six months later, users forget the glasses were ever "dumb."

Conventional wisdom says you need a killer app to drive adoption. Himel flips that: make adoption frictionless, then let the killer app emerge organically. The payoff is delayed but durable.

The Saturday Pivot: When a WhatsApp Message Changed a Product's Trajectory

This part might be uncomfortable for anyone who believes in clean product roadmaps. The original Ray-Ban Meta glasses were not designed to be an AI device. They were a modest hardware refresh of a first-generation product that had not set the world on fire. Himel admits the team debated whether to even release them. Then LLMs took off.

"I was on a Saturday. I had my kid in the child's seat and the backseat. And I start getting a wall of text from Mark on WhatsApp, which led with, hey, I think these glasses might be a great AI device. And then just like thought after thought... by Monday we pivoted 200 people on the team to be building AI for these glasses and their rest is history."

The implication: the hardware upgrade that seemed incremental, better audio, better cameras, slimmer frames, turned out to be exactly what the AI needed to function in real-world conditions. The audio quality crossed a threshold where phone calls were viable. The image quality reached a point where sharing videos felt natural. The hardware was not a breakthrough. But it was good enough to become a platform when the software caught up.

This points to a counterintuitive strategy: build the hardware as if you do not know what the software will be. Keep investing through the iterative cycles, even when the device does not seem revolutionary. Because when an external breakthrough happens, in this case the LLM explosion, you want to be the one already in market, not the one starting from scratch. The delayed payoff comes from having placed the bet before anyone knew it was a bet.

The risk, of course, is that you invest for years and the breakthrough never comes. Meta's willingness to do exactly that, Himel notes they "doubled and tripled down" when others pulled back, is what separates them from competitors who launched, failed, and retreated. Intermittent commitment does not work.

The Trust Paradox: Why Facial Recognition Is Everyone's Line

One moment that required careful navigation is when Himel addresses facial recognition. The feature is not launched, but it is clearly under consideration. He acknowledges it is "one of the most frequently requested features," especially from the blind and low-vision community, while also admitting that a version where strangers can be identified on the street would be "creepy."

"We do not have a feature that's launched that helps you right. We don't have a feature called that or does that feature that's launched in the glasses... It is one of the most frequently requested features from people... But to launch a feature like that requires being really thoughtful with how it works, making sure its privacy first, making sure that it's not creepy."

This is where systems thinking collides with social norms. The same technology that helps a blind person know who's in the room can also be used to scan a crowd and identify strangers without consent. The consequences are not technical, they are cultural. Himel argues that if users feel uncomfortable with people wearing the glasses, they will stop wearing them themselves. The feedback loop works in reverse: social backlash kills adoption.

So the real question is not "can we build it?" but "can we build it in a way that changes the social contract slowly enough to avoid rejection?" Himel's answer is to start with contexts where identification is expected, name tags at conferences, family gatherings, rather than random encounters on the street. The system adapts, but only if you give it time. (Try rushing this, and the trust evaporates.)

Key Action Items

• Over the next quarter: Audit your existing hardware products for "escalator" value. If your device is not useful when disconnected or offline, you are fighting adoption with one hand tied behind your back. Find the baseline utility that exists independent of AI.
• In the next 6 months: Build agentic capabilities that operate in the background, not just proactive queries. Himel's example of auto-capture, glasses that automatically detect and record meaningful moments, shows how removing user friction creates a qualitatively different experience.
• In the next 12 months: Invest in on-device AI for latency-sensitive and privacy-critical use cases. The balance between cloud and local inference will shift, and the companies that control the edge will own the user relationship. Start testing where small models can replace larger ones without sacrificing quality.
• This pays off in 12 to 18 months: Identify one "Saturday pivot" scenario in your product roadmap, an external breakthrough that could transform your device if you are already in market. Prepare hardware that is good enough to be a platform, even if you do not know what the platform will be.
• Over the long term: Engage with the trust paradox proactively. Facial recognition will come, but it must be introduced in contexts where identification is socially acceptable. Do not wait for regulation to define the boundaries; define them yourself, in public, with user comfort as the primary metric.
• Immediate discomfort for lasting advantage: Spend the next quarter deliberately deprioritizing flashy features in favor of hardware durability and comfort. Himel's team made glasses lighter, added adjustable nose pads, and focused on fit. This feels unsexy. It is precisely why it works, while competitors chase AI demos, you are building the device people will actually keep on their faces.