Note-taking crossed over quietly. 75% of professionals now use an AI note-taker, adoption runs highest among small teams (78–81%) and lowest in big enterprises, and 62% of users report saving about four hours a week (Laxis, The State of Meeting Note-Taking 2026). The category is worth around $740M this year and climbing.
Transcription is solved. Follow-through isn't.
Here is the honest sequence in most agencies: the bot joins, the transcript lands in a doc, someone skims the summary, and the six things the client asked for stay in that doc. The board doesn't move. Nobody is assigned. Two weeks later the client asks about the third one and everyone re-reads the notes.
A summary nobody acts on is a better-written version of the problem. You didn't buy minutes — you bought the outcome of the meeting, and a transcript is only evidence that it was discussed.
The three jobs after the call
- Decide what was actually agreed — and record it as a decision, dated and attributed, so "we never agreed to that" has an answer three months from now.
- Turn commitments into tasks on the project the meeting was about, with an owner. Not a bullet list; a row on the board your team already works from.
- File the notes against the project, not in a folder of transcripts organised by date, which is where context goes to be un-findable.
Do those and the meeting genuinely compresses. Skip them and you've automated the stenography and kept the work.
What we built instead of a transcript app
Gavril joins the call, answers when someone addresses it, creates the tasks in the project as they're agreed, and files the notes where the project lives. In-person meetings go through the same path from a phone. The raw transcript is kept as evidence and cleaned up only when notes are written — so a second pass over the same recording reads the same room.
The point was never the recording. It's that the meeting stops being a status readout and starts being the moment work changes — with the routine stuff moved async where it belonged.
← All posts