Skip to content

How to build

How to build a meeting notes and follow-up agent

Most meeting tools solve the wrong half. A transcript is not the problem — nobody re-reads a transcript. The problem is that three decisions and five commitments were made and by Thursday nobody remembers who owns which. What you want extracted is not what was said, but what was agreed.

Start building free~10 min to first version

What you’re building

A meeting notes agent takes a recording or transcript and produces a short structured summary: decisions made, actions with owners and dates, open questions, and anything flagged as a risk. Those actions are then created in whatever tracker your team already uses.

Who this is a good fit for

  • Teams whose action items reliably evaporate between the call and Monday
  • Anyone running recurring calls — standups, client check-ins, one-to-ones
  • People who take notes and then never write them up

How to build it

  1. Step 1

    Describe the structure you want out

    Not "summarise the meeting" — say what the sections are. Decisions, actions with an owner and a date, open questions, risks. The structure is what makes the output usable on Thursday, and it is also what lets the actions be created automatically rather than read and retyped.

  2. Step 2

    Connect the calendar, the call source and the tracker

    Calendar gives it the attendee list, which is what lets "Priya will send the proposal" become an issue assigned to Priya rather than a sentence. Zoom or your recording source provides the transcript, and the tracker is where the actions land.

  3. Step 3

    It builds the extraction and the routing

    It sets up the transcript processing, the structured extraction, the owner matching against your attendee list, and issue creation. Owner matching is the fiddly part and worth checking early — first names are ambiguous in any team of reasonable size.

  4. Step 4

    Run it on one recurring meeting first

    Pick the meeting that generates the most dropped actions and run it there for a fortnight. The quality judgement is simple and worth being strict about: did it catch every commitment, and did it invent any? Over-extraction is more annoying than under-extraction, because a tracker full of phantom tasks gets ignored entirely.

What it does once it’s running

  • Decisions, actions, open questions and risks separated — not one block of summary
  • Owners matched against the real attendee list from your calendar
  • Issues created in Linear, Jira or Notion with owner and due date attached
  • A short summary posted where the team already talks
  • Recurring meetings tracked over time, so you can see what keeps slipping

Tools it connects to

ZoomGoogle CalendarMicrosoft TeamsLinearJiraNotionSlack

What it won’t do

  • Only as good as the audio — crosstalk and poor microphones cost accuracy
  • First-name ambiguity in larger teams needs correcting for the first few runs
  • It captures what was said, so a decision made in the chat or afterwards in a DM will be missed

Start from a template instead

Each of these opens with the prompt already written. Edit it before you build.

Common questions

Does it need a bot to join the call?
No — it works from the recording or transcript your conferencing tool already produces. Connecting Zoom or Teams is enough, which also avoids the awkwardness of a visible notetaker in client calls.
How does it know who owns an action?
It matches against the attendee list from the calendar invite. This is the part worth checking for the first few meetings, because first names get ambiguous in any team of reasonable size — corrections are a sentence.
Can the actions go into our tracker automatically?
Yes. Linear, Jira and Notion can be connected, and issues are created with the owner and due date already set rather than needing to be retyped.
What if it extracts things that were not really commitments?
Tell it to be stricter. Over-extraction is the more damaging failure — a tracker full of phantom tasks gets ignored entirely — so it is worth tuning down in the first fortnight.

Related use case

AI agent for operations & project automation

Create, sweep, and sync tasks across chat, issue trackers, and code automatically.

Build it in the next ten minutes

Start with a sentence. Leave with a working app on its own URL.