The meeting that doesn't eat your day: a complete system with AI

A one-hour meeting with seven people doesn't cost an hour — it costs seven hours of human time. Most companies know this and call meetings anyway that have no goal, no agenda, and no output — because it's easier to get together and see than to figure out beforehand what actually needs deciding. Everyone knows the result: a day chopped into pieces too small to finish anything in, and on Friday the feeling that the whole week was spent talking.

Meeting notes can be automated, and that's a huge relief — but it's the last step. Good notes from a bad meeting are just a precise record of wasted time. This guide sits on top of the notes-taking tips: it covers the whole cycle, from the question “should this meeting even happen?” through the agenda, the materials, and the meeting itself, all the way to checking action items before the next one. AI doesn't do anything magical here. It does the mechanical half nobody has time for: breaks down the calendar, turns materials into a summary, budgets the time, pulls tasks out of the transcript, and checks before the next meeting what didn't get done from the last one.

Read it start to finish, or jump to the phase burning you most right now. Each one has prompts you can copy — just fill in the brackets. And one rule holds over the whole system: AI proposes, a human decides. Which meeting gets cancelled, who gets which task, and what goes out the door is always a human call. The model just supplies the material that makes the decision faster.

A typical scenario

Martin manages a team of eight. His calendar has eleven recurring meetings a week: an hour of Monday planning, a daily stand-up, a ninety-minute Wednesday project meeting, a Thursday status meeting with the whole team, two one-on-ones, and a handful of irregular ones. He's left with two uninterrupted blocks longer than fifty minutes in a whole week. Notes get written sporadically, meeting action items live in people's heads, and half the Wednesday meeting repeats what was already said on Monday.

The worst part isn't the number of meetings, it's their content. At the status meeting, eight people each spend three minutes talking about things that don't concern the other six. At the project meeting, forty minutes go into discussion and the last minute produces “okay, let's think about it some more.” Nobody knows what was decided last time, so it gets decided again — and when someone asks a month later why they went with option B, nobody has an answer.

After putting the system in place, Martin's week looks different. The status meeting is gone, replaced by a written status. The stand-up runs three times a week instead of five. The project meeting has an agenda with three decision points, a time limit on each, and materials participants got a day ahead — so it doesn't open with explaining context, it opens with opinions. Notes and action items fall out of the transcript within ten minutes of the meeting ending. Thursday morning, a routine writes up what's still undone from the last meeting. Altogether, his team frees up roughly five hours a week across everyone involved — and, more importantly, they stop deciding the same things twice.

Phase 1: A meeting audit — what to cancel, what to shorten

The cheapest way to improve meetings is not to hold them. Before you start tuning agendas, go through what's already on your calendar — almost nobody checks whether recurring meetings still make sense. They got created for a project that's since ended, or a problem that's since been solved, and they keep running on momentum.

The decision test: should this meeting happen at all?

Before you reach for AI, have a criterion. A meeting earns its place in three cases: a decision is being made (you need several minds in one place and shared accountability for the choice), a conflict or a sensitive matter is being resolved (where a written message would make things worse), and something is being built together (brainstorming, planning, where ideas need to build on each other in real time). In every other case — sharing information, status reports, “so everyone's in the loop” — a meeting is the most expensive tool available, and a written format works better, as the tip on a written status instead of a status meeting lays out.

Analyzing your calendar

Your calendar is the one honest record of where your time actually goes. Export the last month from it (most calendars support export; otherwise just list the recurring events by hand) and have it analyzed. If your calendar is connected via a connector, the model can read the events directly.

Here's a list of my meetings for the last [4] weeks: name, day,
time, length, number of participants, and my role (organizer /
invited).

[paste list]

Run an audit:
1. Calculate total meeting time and convert it into person-hours
   (length × number of participants). Rank meetings by person-hours,
   not by length.
2. Find recurring meetings that, over the tracked period, happened
   every time with the same group of people — for each, note how
   many person-hours a year it consumes.
3. Flag meetings that overlap in topic or participants and could
   probably be merged.
4. List the days where I don't have an uninterrupted block longer
   than 90 minutes between meetings, and count how many such days
   I have.
5. For each recurring meeting, suggest one of: CANCEL /
   SHORTEN / THIN OUT (less often) / REPLACE WITH A WRITTEN FORMAT /
   KEEP — with a one-sentence reason.

Calculate points 1-4 from the data I pasted. Point 5 is a proposal
for a decision, not a conclusion — for each one, also note the risk
that cancelling it would bring.

You'll get back an overview where point 1 is usually the surprise: a meeting that takes half an hour but has ten people in it costs more than an hour-long one with three. Spot-check the numbers — models make mistakes summing across twenty rows, and if the output is going to decide which meetings get cancelled, it needs to be right. And above all: the last column is a proposal, not a decision. Cancelling a meeting is a managerial and social move with consequences the model knows nothing about.

The three-month test

For recurring meetings you're on the fence about, a simple test works: cancel them for three months and watch what breaks. Usually nothing does. When something does, you'll know exactly what the meeting was for — and you can bring it back in a form that matches, usually shorter and with fewer people.

A replacement, so cancelling doesn't leave a vacuum

A cancelled meeting with no replacement comes back — people set up their own meetings because they still need the information they used to get there. So prepare the replacement before you send the cancellation.

We're cancelling the recurring meeting [name], [frequency],
[length], [number] of participants.

What I think it was for:
[description — what information circulated there, what got
decided there, what people would miss]

Propose a replacement:
1. Which written or asynchronous format covers each of the
   functions I listed — for each one, the concrete form, who
   handles it, how often, and where it gets written down.
2. A template for that written format: what questions people
   should fill in, at most 5, phrased so the answer can't be
   dodged with one sentence saying work is “in progress.”
3. What stays uncovered, and when that will still make an
   actual meeting necessary — a one-sentence trigger, not a
   recurring slot.
4. A 6-sentence announcement for the team: what we're
   cancelling, why, where the information will come from
   instead and starting when, and what people need to do.

For point 4, write it plainly and without apologizing. Don't
promise anything on my behalf — where my decision is needed,
leave [FILL IN].

You'll get back a replacement format and the announcement text. Don't skip point 3: honestly admitting what the written format won't cover is what separates a real cancellation from simply crossing something off the calendar — and a trigger like “if two departments can't agree on priority, we'll call a half-hour meeting” works better than a recurring slot kept around just in case.

Phase 2: An agenda that decides things

An agenda isn't a list of topics. An agenda is a list of questions that need answers and decisions that need to get made. The difference between “item 2: budget” and “item 2: approve moving CZK 200,000 from the campaign budget to development — Martin decides, materials attached” is the entire difference between an hour of talking and eight minutes of work.

The rule: no agenda, no meeting

Adopt one rule and hold to it even when it's inconvenient: whoever calls the meeting sends the agenda no later than the day before; if the agenda doesn't show up, the meeting is cancelled. The effect is immediate and twofold. First, meetings nobody could actually justify disappear — writing an agenda means admitting to yourself what you actually want. Second, the meetings that remain get shorter, because participants show up prepared. The rule itself is spelled out in the tip no agenda, no meeting; this guide builds the rest of the system on top of it.

The rule only works if it applies to you too. The first meeting you cancel because of your own missing agenda is the best investment you can make in its credibility.

An agenda from scattered materials

Writing a proper agenda takes twenty minutes, which is exactly why it ends up getting thrown together five minutes before the meeting as three bullet points. The input, though, is stuff you already have: email threads, notes, leftovers from last time, Slack messages.

Prepare a meeting agenda. My input is scattered material —
notes, fragments from emails, and things carried over from
last time.

Meeting: [name], length [60] minutes, [recurring / one-off].
Participants and roles: [name — role, …]
What I want to be true after the meeting: [goal in one sentence]

Material:
[paste notes, excerpts from threads, unfinished items from
last time]

Build an agenda where every item is phrased as a question to
be answered or a decision to be made, not a topic.
For each item, give:
- the question or decision in one sentence
- type: DECISION / INPUT (collecting opinions, deciding later)
  / INFORMATION
- who introduces the item and who the decision-maker is
  (one name)
- a time limit in minutes
- what participants need to know beforehand to be able to
  decide

Put decision items in the first half. Keep the total time
10 minutes shorter than the meeting's length.

List separately:
a) items that, based on the material, don't need a meeting
   and can be resolved in writing — suggest a form for each
b) items where the material doesn't make clear who decides

You'll get back an agenda you can send as-is, plus two extra lists. List a) is the main payoff of the whole prompt: it typically knocks out two or three items and the meeting gets shorter before it even starts. Keep an eye on list b) yourself — models like to default to whoever called the meeting as the decision-maker, even when that's not right. And watch the item types: almost everything tends to get labeled DECISION because it sounds more important; relabel the items that are really just information by hand.

Checking the agenda before you send it

A short second pass that costs a minute and saves fifteen.

Review this agenda as someone who's going to be sitting in
the meeting, and list what you find. Don't rewrite anything.

1. Items phrased as a topic, not as a question or a decision.
2. Items where it isn't clear how we'll know they're closed.
3. Participants who aren't touched by any agenda item — for
   each one, note which items they do relate to, and whether
   it would be enough to invite them just for that part.
4. Items that require material that doesn't exist yet.
5. An estimate of whether the time limits are realistic given
   the number of participants and the item's complexity; where
   they aren't, say why.

Agenda:
[paste]

Point 3 is the most uncomfortable and the most useful. Inviting someone to a full hour-long meeting for a single five-minute item is the quietest way to waste someone else's time — and the fix is simply inviting them for the first ten minutes.

Phase 3: Materials a day ahead

The most common reason an hour-long meeting isn't enough is simple: the first twenty minutes get spent explaining context everyone could have read beforehand. Materials sent a day ahead erase that block. The condition is that it has to be something readable in five minutes — nobody opens a forty-page document attached to an invite.

A summary of materials for participants

Prepare materials for the meeting [name], happening [date].
Participants: [names and roles]. Decisions that need to be
made: [list from the agenda].

Materials:
[paste documents, numbers, email threads, analysis results]

Write materials that read in under 5 minutes:
1. Current state in at most 8 sentences — what's changed
   since last time.
2. For each decision item from the agenda: what the problem
   is, the realistic options (2 sentences each), what argues
   for and against each, and what needs to be known to decide.
3. The numbers the decision rests on — in a table, with a
   source for each figure.
4. Three questions participants should answer for themselves
   before the meeting.

Rules:
- work exclusively from the material I pasted
- don't add numbers or conclusions that aren't in the material;
  where a figure is missing, write [MISSING: what]
- no recommendation on which option to pick — that's what the
  meeting is for
- no introduction and no closing summary, just the content

You'll get back materials to send along with the agenda. The ban on recommendations in the last rule is deliberate: materials that already contain the answer either steer the meeting unnecessarily or turn it into a rubber-stamping ritual. And don't delete the [MISSING: …] markers — it's better to walk into the meeting knowing one figure is missing than to have it there made up.

Sensitive meetings and data

When the material includes personal data, salaries, health information, legal matters, or non-public financial figures, one rule applies with no exceptions: sensitive data belongs only in a paid account with a contractual data-processing agreement, ideally a tool your company has approved, and even there with names stripped out wherever possible. Anonymizing isn't enough for a meeting about one specific person, though — the context gives them away even without a name. In that case, write the materials by hand; the rest of the system still works.

Making sure it actually gets read

Materials sent a day ahead get read by about half the people. Two things improve that. Put in the invite what happens if they don't read it (“I'll open the first item straight with the question of which option you're picking”), and send it at the same time every week so it becomes a habit. A more radical option that works: spend the first five minutes of the meeting reading the materials in silence. It only feels awkward exactly once.

Phase 4: Timeboxing and running the meeting

An agenda with time limits nobody enforces is just a wish list. Timeboxing isn't about discipline, it's about budgeting: every item gets a slice of time, and when it runs out, the item gets closed or moved. Time is a resource you can only allocate once — and a meeting is the one place where it's spent at a multiple of however many people are in the room.

Budgeting time for each item

Here's the agenda for a [60]-minute meeting with [7]
participants.

Split the time across items and return a table:
item | type | minutes | why that much | what to do if time runs out

Rules for splitting:
- decision items get more time than information items
- for an item where every participant needs to speak, budget
  realistically for [1] minute per person plus time for a
  wrap-up
- put 3 minutes at the start for recapping last time's action
  items
- leave 5 minutes at the end for going through tasks and
  deadlines
- leave 10 percent of the time as an unallocated buffer

In the “what to do if time runs out” column, write a concrete
procedure: close by vote / push to next time / delegate to two
people outside the meeting / [name] decides alone by [deadline].

The minutes must add up to the meeting's length.

You'll get back a breakdown to keep next to you during the meeting. The last column is the most valuable part: a pre-agreed procedure for when time runs out is the one thing that stops the last item from getting hashed out five minutes past the scheduled end with half the people already gone. Check the minutes add up by hand.

Closing questions for every item

Items most often don't get decided because nobody asks a question that can actually be decided. “What do you all think?” opens another round; “are we going with A or B, and who's against?” closes the item. Preparing that phrasing ahead of time takes a minute.

For each decision item on this agenda, prepare two things:

1. A CLOSING QUESTION — one sentence that ends the item so
   the answer is a decision, not another round of discussion.
   It must name the concrete options being chosen between.
2. A FALLBACK — what I do if consensus still doesn't happen
   even after the closing question: who decides alone, by
   when, and what they need for it.

For each item, also add one sentence I should NOT say, because
it reopens the discussion (typically a general question asking
for opinions).

Agenda:
[paste]

You'll get back a set of phrasings to keep alongside your time budget. The last part, about sentences not to say, is quietly the most useful — a general opinion question asked in minute eight of a ten-minute item will reliably add twenty more.

Three habits that keep a meeting on time

Start exactly on time, even if someone's missing. Waiting for a latecomer punishes the people who showed up on time, and guarantees they'll be late next time too.

Announce the move to the next item out loud. The sentence “this item has eight minutes, let's go” sounds bureaucratic and it works: it gives the whole group shared awareness of how much time is left.

Park tangents on paper, not in the discussion. Keep a visible “off-agenda” list (a corner of the whiteboard, a shared doc, a channel). A tangent gets written down there and the meeting moves on. After a few meetings, that list is the best input for building the next agenda differently — and it's often where topics that deserve their own meeting come from.

Phase 5: Notes and action items

This phase is the best-covered part of the system already, and there's no point repeating it in full here. The manually-driven version — from recording consent through cleaning up the transcript to notes that separate decisions from discussion — is in the tip from a meeting recording to notes with tasks. The fully automated chain, where tasks get created directly in your task list via a connector and a searchable archive of decisions builds up on its own, is described in meetings that write themselves up. What belongs here, from the perspective of the whole system, are three things.

Notes follow the agenda, not the timeline

When you send the agenda ahead of time, feed it into the notes prompt as the skeleton. The notes then aren't a transcript of how the meeting went, they're the answer to the questions you set out to answer — and you can immediately see which item never got discussed.

From the attached transcript, produce notes organized by the
agenda I'm attaching, not chronologically.

For each agenda item, give:
- what was decided (one sentence, stated as fact), who decided
  it, and who explicitly agreed
- if no decision was made, write NOT CLOSED and what's needed
  to close it, who owns getting it done, and by when
- tasks that came out of the item: action with a verb | one
  owner | deadline

At the end, add three sections:
ITEMS NOT DISCUSSED — what from the agenda never got opened
OFF AGENDA — what was discussed that wasn't on the agenda
NO OWNER — what needs doing but nobody took it

Work exclusively from the transcript. Where a deadline or
owner wasn't stated, write NO DEADLINE STATED / OWNER UNCLEAR
— don't guess.

Agenda:
[paste]

Transcript:
[paste]

You'll get back notes that show not just what was decided, but also what you planned to cover and didn't get to. Watch the boundary between a decision and a discussion: models like to upgrade “we could” into “it was decided,” because it sounds more finished.

A task with no owner and no deadline isn't a task

What comes out of the notes has to be a who-what-by-when list, where the owner is one specific person. “The team will look into it” means nobody looks into it. If an owner wasn't stated in the meeting, fill it in right when you review the notes — not next time.

A human sends the notes

A draft of the notes is ready in a few minutes, but the name on the task, the number in the deadline, and the decision about what even makes it into the notes are your responsibility. Check the names, numbers, and deadlines, cut any irony and sharper exchanges — notes often reach people who weren't even in the meeting — and send them within thirty minutes of the meeting ending, while everyone still remembers the context and will correct a mistake right in their reply.

Phase 6: Follow-up and checking before the next meeting

This is where it's decided whether the whole system was worth it. A meeting whose action items never get done is no better than a meeting with no notes at all — it's just better documented. The key moment isn't the day after the meeting, it's the day before the next one.

Checking on last time's commitments

Always run this prompt before the meeting, never after it. Two minutes of work replaces a fifteen-minute opening round of “so where did we leave off.”

Here are the notes from the last meeting and what I know
about the status of the tasks.

Notes from [date]:
[paste]

What's happened since then (emails, messages, status in the
task list):
[paste]

Prepare an opening block for today's meeting:
1. A table of last time's tasks: task | owner | deadline |
   status (DONE / IN PROGRESS / NOT STARTED / STATUS UNKNOWN)
2. Overdue tasks, sorted by how long they've been overdue.
3. Decisions from last time that have since turned out to be
   contested or that something has cast doubt on — quoting
   what raised the doubt.
4. Open items from last time that need closing today, sorted
   by urgency.
5. Three questions to open the meeting with.

Where I don't have status information, write STATUS UNKNOWN
and add it to the list of things to ask about — don't assume
it's done.

You'll get back material that lets the meeting start with substance instead of reminiscing. The STATUS UNKNOWN category matters more than it looks: when most items land there, the problem isn't the people, it's that tasks don't live anywhere and nobody's tracking them.

A routine that asks on its own

Once the check works manually, you can take it off your plate entirely. Scheduled tasks (routines) can run a prompt on a schedule — say, every Thursday at eight in the morning before an afternoon meeting — and with a calendar, mail, and drive connector wired up, they'll find the material themselves.

Every [Thursday at 8:00 AM], prepare material for the
afternoon meeting [name].

Steps:
1. Find the latest notes from this meeting in [storage /
   folder].
2. Go through my mail since the date of those notes and find
   messages relevant to the tasks listed in them.
3. For each task, determine a status from the messages found:
   DONE / IN PROGRESS / NO MENTION.
4. List overdue tasks, with the most recent mention, its
   date, and who sent it, for each.
5. Attach three questions to open the meeting with.

Don't send anything and don't message anyone — give the
output to me only.
Where you're not sure of the status, write NO MENTION, don't
guess.

The last paragraph is a safety catch, and don't skip it: a routine with access to mail must never send anything on its own. A human sends the reminders — if only because the right tone for a nudge to a colleague who's behind because they're sick can only be chosen with context the model doesn't have. Setting up tracking of open loops across your mail is covered in the tip the follow-up watchdog.

Dealing with tasks that didn't get done

A task pushed back for the third time isn't late — it's badly defined, badly assigned, or shouldn't be done at all. Adopt a rule that a third push means a decision: either the task gets cancelled, or it gets a new owner, or it gets split into smaller pieces. Without that rule, the list of meeting action items turns into an archive of reproach that nobody reads.

Phase 7: Measuring the system and a quarterly clean-up

A single meeting shows you one meeting. Ten sets of notes show you how you actually operate. Once a quarter it's worth looking at the whole series as one thing — it's uncomfortable, and it's the most useful half hour in the entire system.

I'm attaching notes from the [name] meeting for the last
[3] months.

Analyze the series:
1. Topics that came back three times or more without ever
   getting closed — for each, note when it first appeared and
   what's blocking closure.
2. Decisions that were later changed or reversed — both
   versions and the gap between them.
3. Task distribution by owner: who gets how many, and whose
   deadlines keep getting pushed.
4. The ratio of decisions to total meeting time — how many
   decisions per hour of meeting.
5. Items that show up on every agenda and never lead to a
   decision — candidates for dropping or moving to another
   format.

For each finding, cite which notes it's based on.
Where there isn't enough material for a conclusion, say so
instead of guessing.

You'll get back an uncomfortably accurate mirror. Spot-check two or three numbers against the notes — models make mistakes summing across ten documents. Think twice before sharing point 3: a breakdown of tasks by name is sensitive material, and belongs in a one-on-one conversation more than a group email. Point 5, on the other hand, share freely — it's the fastest route to a shorter agenda.

Common mistakes

• Automating the notes and leaving the meetings unchanged. A precise record of a bad meeting is still a bad meeting. The audit and the agenda have to come before automation.
• An agenda that's a list of topics. “Budget” isn't an item, it's an area. An agenda item is a question that needs an answer, with a named decision-maker — otherwise you spend forty minutes sharing feelings.
• Inviting everyone just in case. Every extra participant multiplies the cost of the meeting and lowers the odds anything gets decided. Someone who's only relevant to one item comes for that one item.
• Sending material that can't actually be read. Nobody opens a forty-page attachment, and the meeting starts with explaining. Materials should read in five minutes and end with questions.
• Letting a routine send reminders. An automation that writes “you're behind” to a colleague on its own can damage relationships faster than the system saves time. AI prepares, a human sends.
• Letting AI decide which meetings to cancel. The model doesn't know the politics, the history, or that a particular meeting is the only place two departments actually talk to each other. A proposal is input to a decision, not the decision.
• Checking tasks after the meeting instead of before the next one. Checking the day after a meeting just confirms nothing's happened yet. Checking the day before the next meeting is the only check that actually changes anything.

The best tools

• A calendar with export — the input for the audit; without person-hour numbers, the debate over cancelling meetings is just about feelings.
• Projects (standing context) — save your agenda format, a list of participants with roles, a glossary of company jargon, and a sample of finished notes; then all you have to do is paste in the material.
• Transcription in Teams or Google Meet — a built-in feature under a company account, nothing to install, and the recording never leaves the company's environment.
• Connectors to your calendar, mail, and drive — the model finds materials and task statuses on its own instead of you copy-pasting into the chat; scope them to only what it actually needs.
• Scheduled tasks (routines) — run the commitments check on a schedule, the day before the meeting, without you having to remember to do it.
• A shared space for notes — one permanent location and consistent naming with the file starting with the date; what can't be found doesn't exist.

What you get out of it

• Time: for an eight-person team, cancelling one status meeting and shortening two others frees up on the order of five hours a week across all participants. Preparing an agenda and materials with AI takes roughly ten minutes instead of fifty.
• Money: indirectly, but noticeably. A decision delayed by two weeks usually costs more than the whole meeting — and the most expensive thing of all is a topic decided three times because everyone forgot what was agreed each time.
• Peace of mind: a calendar with uninterrupted blocks, and confidence that nothing got forgotten, because the commitments check runs before every meeting. You don't have to remember anything yourself.
• Better decisions: separating decisions from discussion, plus written materials in advance, means decisions get made from information, not from who talks the longest. After six months you also have a history where you can trace why a given decision was made.

Pro tip

Build the system backwards. Most teams start with notes, because it's the most visible and the easiest part — and end up with pristine documentation of meetings that shouldn't have happened. Start with the calendar audit, then roll out the agenda rule, then materials, and only once that's running, tackle automating notes. Each step saves time on its own; but in the wrong order, the time you save gets immediately eaten up maintaining notes nobody reads.

And the rule that sits above the whole guide: the goal isn't better meetings, it's fewer meetings in which more gets decided. AI is great at the mechanics in this system — it breaks down the calendar, summarizes materials, budgets the time, pulls out tasks, tracks deadlines. The decisions about who to invite, what to cancel, and what to tell whom stay with a human. And that's how it should be, because in the end a meeting is about people, not documents.