The AI proposes. You decide. Nothing happens in between.
UTTER IN uses AI to understand what you said, and never to act on it. Here is the rule that keeps those two jobs apart, and why.
UTTER IN2 min read
There is one rule at the top of how UTTER IN is built, above every other decision, and it has four parts:
The AI interprets and proposes. Our own code checks. You approve. Then, and only then, the app acts.
Each part exists because the one before it is not enough on its own.
The AI interprets
Language models are very good at the hard part: turning "lunch with Khaled at four, and it's dad's birthday on the ninth" into two separate things with a time, a person and a date. That is what we use them for.
They are not reliable at being certain. A model can read "at four" and not know whether you meant morning or afternoon, and it can be confidently wrong in a way that looks exactly like being right.
Our code checks
So nothing the model returns is trusted as it arrives. It is checked by ordinary, predictable code with a fixed set of rules: is this a real date, is it in the future, is anything missing, was the year actually said. The model is never the one who decides a date is valid.
When something is missing or uncertain, the card says so plainly. When there is a small set of sensible answers, such as 4 AM or 4 PM, you get them as taps. The options only ever come from things you have said yourself. The app will not invent a place, a person or a time and slip it in as a suggestion.
You approve
Nothing is saved as a plan until you approve it. Not a reminder, not an event, not a note. Every item in a capture gets its own card, and you can approve, change or dismiss each one on its own.
Then the app acts
Only after approval does anything happen, and what happens is limited on purpose. UTTER IN does not make purchases, book tables, send messages on your behalf or take any action outside the app by itself. It saves what you approved and reminds you when you asked to be reminded.
Why draw the line so firmly
Because the alternative fails quietly. An assistant that acts on its own best guess is right most of the time, and the times it is wrong are the ones you do not see until later: the reminder set for the wrong day, the event that never existed.
The product is meant to take things off your mind. It cannot do that if you have to keep an eye on it. So the deal is simple: it does the understanding, you keep the say.
Something to add, or something we got wrong? Tell us.
