When Robin gets it wrong: the misheard medication name
The NestBoard team · 2026-06-19
Last week someone on the team said "Add Zyrtec to the shopping list" to Robin. What showed up on screen was "Xirtec."
Not quite right. But also not catastrophic, because Robin didn't just silently add the wrong thing and move on.
Instead, the confirmation screen showed the item name in large type, with a simple "Is this correct?" and two buttons. The person caught it immediately, tapped "Edit," fixed the spelling, and the right item made it to the list.
Why we show you what we heard
Voice transcription—even the good stuff—makes mistakes. Medication names are especially tricky. They're often invented words, sometimes similar to other drug names, and people don't always enunciate them clearly when they're rushing through the kitchen between dinner and bedtime.
We could have built Robin to listen and save without showing its work. That would feel faster in the moment. But it would also mean wrong dosages, wrong medication names, or phantom chores showing up days later with no one sure how they got there.
The confirmation step adds one tap, but it closes the loop. You know what Robin heard. You can catch the error before it propagates into your shared medication bottles or ends up on someone's phone as a reminder.
What happens when you don't catch it
The interesting part is what happens if someone does miss the typo and taps "Confirm" anyway.
If it's a shopping list item, you might buy the wrong thing or spend a minute at the pharmacy trying to figure out what "Xirtec" was supposed to be. Annoying, but fixable.
If it's a medication reminder, the consequences are a little higher. You might not recognize the name when the notification fires. Or another family member might see it and wonder if someone started a new prescription without mentioning it.
That's why the confirmation screen isn't just a nice-to-have. It's the guardrail that keeps voice input useful instead of chaotic.
We're still learning
Robin gets most things right. But "most" isn't the same as "always," and we're not interested in pretending otherwise.
When transcription fails, we want to fail in a way that's visible and fixable. Show the mistake, make it easy to correct, and don't let bad data sneak into the system just because someone was in a hurry.
Voice input should feel like talking to a person who's trying to help—someone who checks in when they're not sure they heard you right. That's the version we're building.