We shipped the first version of Afternotes assuming we knew what “a meeting assistant” should look like. A hundred early-access users later, three of our biggest assumptions turned out to be wrong.
We killed the meeting bot
Our first version worked the way most meeting tools do: a bot joins the call as a visible participant, sits in the grid, and records. It’s the industry-standard pattern, and it was the first thing people asked us to remove.
The complaints were consistent — a bot in the call changes how people behave in the call, especially in smaller or more sensitive meetings. It’s also just visual clutter in a grid that’s already crowded. We rebuilt capture to work at the OS level instead, with a clear on-screen indicator instead of a fake participant. Feedback since then has been the opposite: people forget it’s there, in a good way.
The brief needed way less in it, not more
Our first “meeting prep” feature tried to be comprehensive — full history, every related thread, every past action item. Users skimmed past it. The problem wasn’t coverage, it was signal. Nobody wants to read a report before a meeting; they want the two or three things that would actually change how they show up.
We rebuilt it around a much harder editorial question: if this person only reads one sentence, what should it say? That constraint made the feature worse at completeness and dramatically better at actually getting read.
The unexpected side effect
Once briefs got shorter, people started trusting them enough to stop doing their own prep as a backup — which was the real goal all along. A feature nobody reads isn’t saving anyone time, no matter how much it captures.
Local storage went from an option to the default
We originally offered cloud sync as the default, with local-only storage as a setting for privacy-conscious users. Early conversations made it clear that framing was backwards — the “privacy-conscious” users weren’t a niche, they were everyone, once they actually thought about what a meeting transcript contains.
We flipped the default. Notes stay on-device unless a team explicitly turns on sharing. It made onboarding one step longer. It was worth it — trust in a product like this has to be earned by the default behavior, not explained away in a settings page.
What didn’t change
The core bet — that a good meeting assistant should feel like a colleague who remembers everything, not a transcription tool — held up. Everything above was about getting the execution right, not questioning whether the idea was right. A hundred users in, that’s the part we’re most confident about.