Think of the last time a piece of software did something on your behalf that you didn't sign off on. The card that got charged on a free trial you forgot to cancel. The reply-all that went to two hundred people. The calendar that quietly invited your whole team to a thing you were only mulling over. There's a specific feeling in that half-second — a small, cold freeze — and it isn't about the money or the embarrassment. It's about control. Something acted as you, and you weren't there.
The freeze
That freeze is the most honest research we have. Every AI product on the market is racing to remove the moment where you're in the loop, because being in the loop is friction, and friction is the enemy. We looked at that same moment and decided it wasn't friction. It was the point. The freeze is your judgement arriving a beat too late — and the whole design of boop is built to make sure your judgement always arrives first.
So boop does the chore, and then it stops. It holds the finished thing up and waits for you. Not because it can't finish — it can — but because the last inch of any consequential act belongs to a person. You say yes with your hand. That pause, the one everyone else is trying to delete, is the product.
The freeze is your judgement arriving a beat too late. boop makes it arrive first.
The two tiers
Asking about everything, all the time, would be its own kind of tyranny — a robot that pesters you to approve reading its own to-do list is a robot you'll unplug by Thursday. So boop runs on two tiers of consent, and which tier you're in is never a surprise.
Tier one · the everyday boop
Reversible, low-stakes work — a draft written, a list sorted, a reminder set. The mote glows, you press the head, it's done. One boop. Booped.
Tier two · the full card
Money, first-times, anything that leaves the building and can't be walked back. A head-boop isn't enough here. boop sends a full approval card to your phone, names the exact thing and the exact consequence, and waits.
The card is the sacred object of the whole product, so we play it dead straight. It never jokes, never nudges, never pre-selects the yes. It reads like this:
Boop wants to send this reply to Maya. Send it?
Not “confirm.” Not “authorize.” The object and the consequence, named in plain words, every time. What you read is what will happen. Nobody boops blind.
The cost we chose
Here's the honest maths, because a build note that hides its trade-offs isn't worth reading. Asking is slower than not asking. Every approval is a few seconds you spend that a fully-automatic assistant would save you. Over a week that adds up to real minutes. A competitor with no pause will always be able to claim a faster number than we can. We know. We chose the slower number on purpose.
- NOT ASKING · faster on paper · you find out what it did afterward
- ASKING · a few seconds each time · nothing consequential happens you didn't choose
- THE TRADE · we spend your seconds to keep your say. Every time.
Because the thing those saved seconds cost you is the one thing we won't spend: the certainty that nothing went out in your name that you didn't stand behind. The people the AI wave skipped — the ones who've been burned by software that acted first and explained later — don't want a faster black box. They want help they can trust that keeps them in charge. Slower, and theirs. That's the trade, and we'd make it again.
So that's why our robot asks permission. Not caution for its own sake, not a settings toggle you can switch off. It's the shape we started from. boop does the work. You keep the final say.