Iro AIWhat you'll be able to do
- Write a tight agenda with a stated outcome and realistic time boxes
- Turn a raw transcript or messy notes into a clean, skimmable summary
- Extract decisions and action items with a clear owner and due date for each
- Draft a follow-up message in seconds that people actually read
- Decide when a meeting could have been an async note instead
Inside the path
A focused set of five-minute lessons — each one ends with a hands-on exercise, not a quiz you can guess.
Do we even need this meeting? 4 min
Use AI to test whether a topic needs a live meeting or a two-line async message.
Agendas with an outcome 5 min
Prompt for a tight agenda that names the decision to make and time-boxes each item.
Notes to a clear summary 5 min
Turn a rambling transcript into a short, skimmable summary anyone can catch up on.
Decisions and owned actions 6 min
Extract decisions and action items with owners and due dates — and flag what's unclear.
The follow-up in 60 seconds 5 min
Draft a follow-up message from the summary that people actually open and act on.
Try a sample exercise
This is the kind of card you'd practice inside Iro — you do the thinking, then get feedback.
◆ Sample exercise · Prompt practice
You just finished a 45-minute project sync and captured a page of messy, half-typed notes. You want AI to turn them into action items — but the notes don't clearly say who owns everything, and inventing an owner would send the wrong person chasing the wrong task.
Your task: Choose the prompt that turns notes into trustworthy action items without making things up.
- "Summarize these meeting notes."
- "Here are raw notes from a project sync. Turn them into: (1) key decisions made, (2) action items as owner → task → due date, and (3) open questions. Only assign an owner if the notes clearly name one — if it's unclear who owns something, list it as 'unassigned' and flag it rather than guessing. Use only what's actually in the notes."
- "Read these notes and tell everyone what they need to do next."
- "Write a polished follow-up email based on these notes."
See why the second prompt wins
The winning prompt gives the model a structured extraction format — decisions, action items, and open questions — and demands the one thing action items live or die on: an owner, task, and due date for each. Crucially, it tells the model to only assign an owner the notes actually name and to flag anything unclear as unassigned rather than inventing one, and it grounds the output in the source ("use only what's in the notes"). That's what stops AI from confidently sending the wrong person after the wrong task. The others fail: "summarize these notes" gives you prose with no owners, "tell everyone what to do" invites the model to invent tasks and assignees, and jumping straight to a "polished email" skips the extraction and buries any guesses in a confident tone. In Iro you practice these notes-to-actions prompts and get feedback on whether they'd survive contact with a real, messy transcript.
The two places meetings break — and where AI fits
Meetings fail at the edges. Going in, there's no agenda, or an agenda that's a list of topics with no decision attached, so the conversation wanders. Coming out, someone says "we'll follow up," nobody owns anything specific, and the same discussion happens again next week.
AI slots into both gaps. Before the meeting, hand it your rough topic and have it draft an agenda that names the outcome you want, orders the items, and time-boxes each one — then you cut it down. After the meeting, hand it the transcript or your notes and have it produce a summary plus decisions and action items. The before-and-after are where the leverage is; the meeting itself is still a human conversation.
Turning notes into action items you can trust
The highest-value AI move for meetings is notes-to-actions — but only if the output is grounded in what was actually said. A summary that invents an owner or a deadline is worse than no summary, because people act on it.
- Hand off: summarizing the transcript, pulling out decisions, structuring action items as owner-task-due, and drafting the follow-up message.
- Keep: confirming the owners and dates are right, resolving anything the model flagged as unclear, and the decisions themselves.
The safeguard is a single instruction: tell the model to use only what's in the notes and to mark anything ambiguous as unassigned instead of guessing. That one line is the difference between a follow-up people trust and one they have to re-check. And when you do this consistently, you start noticing which meetings produced no real decisions at all — a strong signal they could have been an async message next time.