Meetings That Write Their Own Notes

The meeting after the meeting

"I leave meetings with no record, or spend 30 minutes writing one."

Both halves of that sentence cost you. The no-record meeting quietly bills you later: 'wait, what did we decide?', the commitment nobody wrote down, the same discussion re-held in two weeks because no one can prove it already happened. The 30-minute write-up costs you immediately — and if you have four meetings a week, that's two hours of after-meeting admin. Maria, from our worked example in Module 2, was spending 25 minutes per meeting, four times a week.

Here's the good news: meeting notes are the single most natural AI automation in office life. The input is abundant (everything said), the transformation is pure Module 2 material (summarize, extract, flag), and the output format barely changes from meeting to meeting. You already built the starter version of this workflow in 'Build Your First Automation.' Today we turn it into a permanent system: capture, process, review, distribute, archive.

By the end of this lesson, 'writing up the meeting' stops being a task on your list. It becomes six minutes of review.

Capture first — you can't automate what wasn't recorded

The four-stage pattern starts with input, and here's the meeting-notes trap: the input has to exist. AI cannot summarize a meeting that lives only in five people's fading memories. So the first fix isn't AI at all — it's a capture habit.

Two capture modes, in order of simplicity. Messy live notes — you (or a rotating notetaker) type fragments into a doc during the meeting: half-sentences, names, decisions, 'Dana → budget by Fri.' The notes can be genuinely terrible; that's fine, because cleaning up messy notes is exactly the 'almost the same' transformation AI excels at. Messy-but-complete beats tidy-but-thin every time. Transcripts — many meeting platforms can produce a transcript or AI-generated notes. Richer input, better extraction — and one hard rule attached:

Recording and transcription require consent. Laws differ by region, workplaces have policies, and — beyond the legal question — surprising people with a recording damages exactly the trust meetings run on. The professional habit is simple: announce it, get agreement, and make it normal ('I'm turning on notes so nobody has to scribble — everyone okay?'). If consent isn't practical, messy live notes work almost as well.

Whichever mode you use, capture ends with one move: the raw record lands in the same place every time. That consistency is what makes the next stage automatic.

The standing instruction block

In Module 2, Maria pasted instructions into her AI assistant each time. The upgrade that makes this a *system* is the standing instruction block: your processing instructions written once, saved next to your Reply Library, and reused for every meeting forever.

A complete standing block has four parts. The job: summarize the discussion; extract every commitment as task – owner – deadline; flag ambiguity instead of guessing. The ingredients that never change: your team roster with roles (so names attach correctly), your output format (Summary / Action Items / Flagged), your rules (never invent owners or deadlines — write ASK; separate real commitments from ideas people floated). The blank that changes: today's raw notes. The standard: under 250 words, plain language, posted within an hour.

Why this matters beyond convenience: a standing block gets *better with use*. Every time your review catches something — a guessed deadline, a floated idea promoted to a task — you add one line to the block ('never invent deadlines' was Maria's first). Ten meetings in, your instruction block has absorbed ten reviews' worth of judgment. That's the improve-loop from Module 2 running on rails — your recorded fixes literally become the system's instincts.

One workflow, four meeting types

Not all meetings produce the same output — and this is where beginners' meeting automations quietly fail: one generic summary format flattens every meeting into the same mush. The fix is small: your standing block stays the same, and one line at the top declares the meeting type, which changes what the AI hunts for.

Four types cover nearly everything. Declare the type, and the output snaps to fit:

Review, distribute, archive — and the compounding payoff

Review ★ for meeting notes has three specific traps to catch, in order. *Wrong owners* — the AI heard 'Dana mentioned the budget' and assigned Dana a budget task she never took. *Promoted ideas* — someone mused 'we could try X' and it became an action item; brainstorm chatter is not commitment. *Lost nuance* — the summary says 'team agreed' when the room was actually split; check anything politically sensitive before it becomes the written record. Six minutes of this review beats twenty-five minutes of writing — and produces better notes than tired-after-the-meeting you ever wrote.

Distribute is a one-liner in your done state: same channel, same format, within the hour. Speed matters more than polish here — notes that arrive while the meeting is fresh get read; notes that arrive tomorrow get skimmed.

Archive is the payoff nobody expects. Every processed meeting lands in one folder, one doc, one searchable place — and three months in, you own something rare: a searchable record of every decision, who made it, and when. 'Didn't we already decide this?' becomes a thirty-second lookup instead of a re-litigated meeting. Maria's archive ended a two-month-old dispute in one search. The notes save you minutes; the archive saves you meetings.