Mastermind · May 31, 2026
Talking to Computers in Public
I was at a cafe today trying to dictate to Claude.
I got two sentences in and stopped. The guy at the next table had headphones off. The barista was three feet away. I could feel everyone tracking what I was saying, the way you track a stranger's phone call on a train. So I lowered my voice, then I whispered, then I just opened my laptop and started typing instead. The fastest input I have, the one I actually prefer, the one Anthropic and OpenAI are betting the next five years on, was unusable because of the people sitting near me.
This is going to be a problem for everyone, very soon.
Voice is becoming the dominant way we talk to AI. Claude has voice. ChatGPT has voice. Whisper turned dictation from a novelty into the fastest text input most people have ever used. The next two years of consumer AI are going to be people walking around talking to their computers the way they currently talk to their AirPods, except the conversations will be longer, more confidential, and more constant.
And most of us work in cafes. Or coworking spaces. Or open offices. Or trains. Or hotel lobbies. The phone booths are always full. The library is silent. There is nowhere to go.
The fix I keep coming back to is a mask.
Not a medical one. A small, soft, matte wearable that sits over your mouth and chin, with a button on the side. Press it once and the outer shell muffles your voice. The internal mic still picks you up clean, sends it over Bluetooth to your laptop, and Claude hears you perfectly. But the person at the next table just hears a soft mumble, the same way someone wearing AirPods sounds when they take a call. Press the button again and the vents open, the seal breaks, your voice projects normally, you can order another latte.
That is the whole product.
The secondary thing is hygiene. It is a mask. In flu season, on a long flight, in a dusty city, you get the second use for free. But that is not the wedge. The wedge is the cafe.
This is not a new idea, which is part of why I trust it. There was a thing called Hushme around 2018, a mask that did roughly this. It looked clunky, it never took off, the voice-AI moment had not arrived. The problem they were solving did not actually exist yet, because nobody was talking to their computer for twenty minutes at a stretch. Now we are. The product was right, the timing was wrong by about seven years.
The parts I do not know how to solve yet are the social and industrial design ones. Can you make it small enough that it does not look weird. Can you make passive foam absorption do enough work, or do you need active noise cancellation pointed outward at the room. How does it feel against your face after an hour. Does it fog your glasses. Can you make it look like something Nothing or Teenage Engineering would ship, instead of something from a hospital.
Those are real questions. But they are bounded engineering questions, not category questions. The category is clearly going to exist. Someone is going to build this, because the alternative is that the most important interface shift of the next decade gets bottlenecked by other people's ears.
I keep adding ideas like this to a list. Most of them I will never build. But this one I would actually wear tomorrow.