What If You Could Talk to AI Without Speaking?

I saw a demo today that genuinely stopped me for a moment.

The Subvocal Company is developing glasses that it says can decode deliberate, almost-invisible silent mouthing into text at more than 200 words per minute. The company positions the glasses as an input method for things we already use—messages, notes and conversations with AI—rather than as another destination app.

The claim is still a claim: the hardware is in development, and that 200+ WPM performance has not yet been independently verified. But the idea itself is fascinating.

What if I could communicate with an AI at roughly the speed of speech, without anyone around me hearing the conversation?

Voice is natural. Voice in public is not.

Typing is private and socially invisible, but it creates friction. Voice is much more fluid, but try dictating a long prompt in a quiet office, on a train, in a library or next to someone on a flight. Suddenly the most natural interface becomes awkward.

That matters more as AI becomes conversational. I increasingly want to explain context, refine an idea, ask follow-ups and think through a problem with an assistant. A keyboard works, but it is not how humans naturally carry a rich conversation.

Subvocal is attacking that gap. According to the company, the system is designed to decode words you deliberately mouth—not arbitrary thoughts—and turn them into text without requiring audible speech.

It potentially combines the bandwidth of conversation with the social privacy of typing.

That is the part that awed me.

The missing layer in ambient AI

Over the last several months I have been thinking a lot about where the interface to AI is going.

Agents are beginning to move beyond answering questions and toward performing work. Connectors and APIs give those agents access to specialized capabilities. Cameras and smart glasses can give an assistant some understanding of what we are looking at. Local models can make parts of that intelligence more private and available even when the cloud is not.

But there is still a basic human problem: how do I communicate continuously with that intelligence without constantly pulling out a phone, typing, or talking to myself in public?

Silent speech suggests a surprisingly elegant answer.

None of these layers needs to come from the same company. That may be the most interesting part.

The interface could disappear

In my earlier note, The Next App Store Might Not Have Apps, I wondered whether some future products will primarily exist as capabilities that agents discover and invoke rather than interfaces people manually navigate.

Subvocal made me think about the other side of that argument.

If software interfaces become less visible, perhaps the human input interface becomes less visible too.

I could be looking at code and silently ask an agent to investigate a failure. I could be walking through an airport and privately ask it to rework my itinerary. I could be sitting among coworkers and formulate a detailed prompt without putting on headphones, opening a phone or speaking aloud.

The screen does not necessarily disappear. Visual interfaces are still better for many things. But the keyboard and touchscreen may stop being the mandatory doorway into computing.

There are plenty of reasons to remain skeptical

This is early hardware. Real-world accuracy matters much more than a polished demo. Different faces, languages, accents, technical vocabulary, long sessions, movement, battery life and calibration could all determine whether silent speech becomes delightful or exhausting.

There are also important privacy questions. Physiological signals used as computer input deserve strong guarantees around processing, storage, permissions and what exactly can—and cannot—be inferred.

And the product should not be confused with mind reading. Subvocal itself says the system is intended to decode deliberate silent mouthing. That distinction matters.

Still, new interfaces often look strange until one of them crosses a threshold where the old interaction suddenly feels cumbersome.

The thing I keep coming back to

AI is becoming capable enough that the bottleneck is increasingly not only intelligence.

It is interaction.

How quickly can I give the system context? How naturally can I express what I mean? Can I use it around other people? Can it understand what I am looking at? Can it act once it understands me?

Subvocal is interesting to me because it attacks one of those bottlenecks at the physical interface between a human and a computer.

Maybe this exact product becomes mainstream. Maybe another implementation wins. Maybe silent speech remains niche.

But after seeing the demo, I find the direction difficult to unsee.

We have spent years teaching computers to understand our words. The next interface shift may be letting them hear us without us making a sound.

Sources & further reading: The Subvocal Company · Digital Trends coverage, Oct. 30, 2026