Tomasz Wojcik leads an eleven-person operations team at a regional payments company, and one ordinary Wednesday he added up his calendar for the first time. Twenty-three hours that week were meetings. He sat in a recurring Monday status sync where six people read updates aloud that everyone could have read in two minutes; a "quick" project check-in that always ran thirty minutes long and produced no written decision; and a cross-team coordination call where the same three blockers resurfaced because nobody owned them. He was not short on meetings. He was short on outcomes. That afternoon he started using AI to do the mechanical parts of meeting work, the agenda drafting, the note-taking, the action-item extraction, so that he could spend his own attention on the two things a machine cannot do: deciding what a meeting is actually for, and reading the room while it happens. This lesson follows Tomasz as he rebuilds his meeting practice from prep to follow-up.

What This Lesson Covers

Advanced meeting management means running a meeting well from end to end: deciding whether it should happen at all, preparing it so people arrive ready, capturing what is said, pulling out who committed to what, and closing the loop afterward. AI can carry a large share of that load. It drafts agendas, transcribes and summarizes the conversation, extracts action items, and writes the follow-up. Your job is the judgment around it: the purpose, the participant list, the facilitation, and the call to cut a meeting that has stopped earning its place.

You will learn how to name a meeting's purpose so its format follows, how to use AI to build a tight agenda and pre-read, how live notetakers and transcription change what you can do in the room, how to extract action items and assign clear ownership with the RACI framework, how to automate follow-up, and how to audit a recurring meeting's real cost and decide to keep, shorten, or kill it. We will follow Tomasz through each step with the actual prompts and decisions he used.

Name the Purpose Before You Book the Room

Most bad meetings are bad because nobody decided what kind of meeting they were. A meeting has one of a few purposes, and the purpose dictates the format. Tomasz learned to label every meeting before scheduling it.

• Decision meeting: the group needs to choose something. This needs a pre-read, a clear decision method, and someone with the authority to call it.
• Alignment meeting: people need a shared understanding of a direction already set. This needs clear framing and time for questions, not debate.
• Problem-solving meeting: the group works through something messy together. This needs the right people in the room and live back-and-forth, so it genuinely has to be synchronous.
• Status meeting: people share updates. This is the one that almost always wastes time, because reading is faster than listening. It usually belongs in writing.
• Relationship meeting: a one-on-one or team connection where the point is the human contact, not the agenda.

When Tomasz labeled his Monday sync, he realized it was a status meeting wearing a problem-solving costume. People showed up to recite progress, and the rare real blocker got buried under updates nobody needed to hear out loud. Naming the purpose told him the format was wrong, which is the first decision AI cannot make for you.

Building the Agenda and Pre-Read With AI

Once Tomasz knew the purpose, AI made preparation fast. A good agenda has topics in a sensible order, a time box on each, a named owner for each item, and a stated outcome. A good pre-read gives people what they must know so the meeting time is spent deciding, not catching up. Here is the prompt Tomasz used to prep a genuine decision meeting about whether to migrate his team's reconciliation process to a new vendor tool:

"I am running a 60-minute decision meeting on Thursday about whether to move our reconciliation process to Vendor B. Attendees: me, two senior analysts, our finance partner, and an engineer. We need a decision by Friday. Draft an agenda with time boxes and an owner for each item, a decision method I should state up front, and a one-page pre-read outline listing what each attendee should review before the meeting. Keep status updates out of the live agenda; assume people read the pre-read."

The draft came back with a clean structure: five minutes to restate the decision and the method, fifteen for the case to migrate, ten for the case to stay, fifteen for open concerns, ten to decide, five for next steps. It proposed a consultative method, meaning attendees weigh in and Tomasz makes the final call, which fit his actual authority. The pre-read outline listed the cost comparison, the migration risk summary, and the two analysts' current pain points. Tomasz cut one item, fixed the order so concerns came before the decision rather than after, and added a line naming who would own the rollback plan if they migrated. Ten minutes of editing replaced the forty-five it used to take him to build a decision meeting from scratch.

Live Capture: Notetakers and Transcription

The second place AI earns its keep is during the meeting. An AI notetaker joins the call, transcribes it, and produces a summary with decisions and action items. This frees you from the impossible task of facilitating well and writing everything down at the same time. Tomasz used to scribble notes and miss half the discussion; now he stays present, watches faces, and lets the tool capture the words.

Two cautions matter here. First, transcription is not perfect: it mishears names, technical terms, and crosstalk, so the raw transcript is a draft, not a record. Second, an always-on recorder can feel like surveillance and quietly chill the conversation. Tomasz handles both with one habit. At the top of every recorded meeting he says, plainly: "I am recording this so the notes and action items are accurate, and I will share the summary with everyone afterward. If anyone wants something off the record, just say so." Naming it removes the unease, and the offer to go off the record preserves the psychological safety that makes people raise real concerns.

Extracting Action Items and Assigning Them With RACI

A meeting only delivers value if its commitments survive contact with the next day. AI is excellent at scanning a transcript and pulling out every "I'll handle that" and "we should follow up on" into a clean list of tasks. But a raw list of tasks is not enough, because the most common failure is ambiguous ownership: three people each thought someone else had it. This is where Tomasz layers in RACI.

RACI is a simple way to assign roles on any task so nothing falls through the cracks. For each action item you name four roles:

• Responsible: the one person who actually does the work. Exactly one name, never two.
• Accountable: the one person who owns the outcome and answers if it slips. Often the manager. Also exactly one name.
• Consulted: people whose input is needed before it is done. Two-way conversation.
• Informed: people who just need to know it happened. One-way notice.

The discipline that fixes most broken follow-through is the rule that Responsible and Accountable are each a single person. "The team will fix the reconciliation bug" is how things die. "Responsible: Lena; Accountable: Tomasz; Consulted: the finance partner; Informed: the wider ops channel, by Friday" is how things get done. Tomasz asks his AI to take the extracted action items and format them into a RACI table, then he corrects it, because the tool can guess who spoke but cannot know who should truly own a thing. After his Thursday decision meeting, the AI extracted nine commitments; Tomasz merged two duplicates, reassigned one Responsible from a person who was overloaded, and confirmed each item had a due date. The whole cleanup took six minutes.

Automating the Follow-Up

The follow-up email is where AI gives you back the most time for the least judgment. Within an hour of a meeting, while it is fresh, Tomasz feeds the cleaned summary and RACI table to AI and asks for a short recap: the decision made and why, the action items as a who-owns-what-by-when list, the open disagreements still unresolved, and the date of any next meeting. He reviews it, fixes any tone or sensitive wording, and sends it within twenty-four hours. He also has AI draft the next meeting's agenda directly from the open items, so the loop closes into the following session instead of starting cold. The rule of thumb he follows: the more the follow-up is plain documentation, the more you can lean on AI; the more it carries sensitive or political weight, the more you write it yourself.

Deciding What Should Be a Meeting at All

The highest-leverage move is not running meetings better; it is running fewer of them. Some meetings should be a written message. Some messages should be a meeting. The dividing line is the kind of work involved.

• Move to async (written): status updates, information sharing, and decisions that have a clear framework and no real disagreement. These are faster to read than to attend.
• Keep synchronous: genuine problem-solving, sensitive or emotional conversations, complex negotiation, and relationship-building. These need real-time human presence.

Tomasz converted his Monday status sync into an async ritual: every Monday by end of day, each person posts a three-line update, what they finished, what they are doing, and any blocker, into a shared channel, and people comment on each other's posts. He kept a thirty-minute synchronous slot on Wednesdays but reserved it strictly for blockers and decisions that the async thread could not resolve. Most weeks, that slot now ends in fifteen minutes.

A Worked Example: Auditing a Recurring Meeting's Cost

To decide whether to keep, shorten, or cut a meeting, Tomasz put a dollar figure on it. The math is simple and changes the conversation completely. A meeting's cost is the number of attendees multiplied by the hours it runs multiplied by a loaded hourly rate, which is salary plus benefits and overhead, roughly 1.4 times base pay.

His Monday status sync had eight attendees, ran one hour, and recurred weekly. He used a loaded rate of $75 per hour. So one session cost 8 people times 1 hour times $75, which is $600. Over a year of about 48 working weeks, that is $28,800 for a meeting whose entire content was status that nobody discussed. He fed the numbers to AI to sanity-check the arithmetic and to model alternatives, but the decision was his.

He weighed three options:

1. Keep it as is: $28,800 a year for read-aloud updates. Hard to justify once the number is on the table.
2. Shorten to 30 minutes: halves the cost to about $14,400 a year, but the format is still wrong; people would still recite status.
3. Cut it and go async, with a 30-minute Wednesday slot for four people who actually need to coordinate: the async update costs each person about five minutes to write, and the trimmed sync runs four people times half an hour times $75, which is $150 a session, or roughly $7,200 a year.

Tomasz chose option three. He recovered about $21,600 of loaded time a year and, more importantly, gave eight people back roughly forty-five minutes every week to do the work the meeting was supposed to be about. He ran the same audit across his recurring calendar and killed one more meeting and shortened two others. The point of the exercise is not the exact dollar figure, which is always an estimate; it is that naming the cost out loud makes the trade-off honest and gives you the evidence to defend the change when someone asks why their meeting disappeared.

Four Ways This Goes Wrong

Over-automating until you check out. If the notetaker is so good that you stop listening, you lose the read of the room that is the entire reason to meet live. Let AI capture the words; you capture the meaning, the hesitation, the thing not being said.

Ambiguous ownership. Extracting action items without assigning a single Responsible and a single Accountable just produces a tidy list of things nobody owns. The RACI discipline is the whole point.

Recording without telling people. Silent transcription erodes trust the moment it is discovered. Always announce it and offer an off-the-record option.

Cutting a meeting people needed. The cost audit is a tool for judgment, not an axe. A relationship or problem-solving meeting can be "expensive" by the math and still be the most valuable hour of the week. Cut status, protect connection.

Facilitation, Authority, or Both

Naming the purpose tells the room what the meeting is for. Naming your own mode tells them what your role in it is, and this is the single clarification that saves the most friction. There are three modes, and you should know which one you are in before you speak.

• Facilitation mode: your job is to create space for other people's thinking. You hold your own opinions lightly, ask more than you assert, and let the group reach the answer. Useful when the expertise sits with them, or when buy-in matters more than speed.
• Authority mode: you are going to decide, and what people need is to understand your reasoning. This is faster and clearer, and it is not a failure of collaboration. Sometimes the call is genuinely yours.
• Hybrid mode: listen first, then decide. This is the most common and usually the best pattern for a manager, but only if you say out loud that it is what you are doing.

The damage comes from ambiguity. People argue hard for twenty minutes believing they are shaping a group decision, and then you announce your call and they feel their input was theatre. Or you sit back waiting for consensus while everyone waits for you, and the meeting ends with nothing decided. Tomasz now states the mode in the first thirty seconds: "I am making this call after hearing everyone, so give me your strongest objections now," or "This one is genuinely ours to decide together, and we are not leaving without a decision." When he prepped his reconciliation meeting, the consultative method the AI proposed was the right mode, but it only worked because he said it before the discussion rather than after.

Two More Ways Meetings Fail

The wrong people are in the room. Every extra attendee slows a meeting down, and every missing one costs you context you will not get back. A decision made without the person who knows the operational constraint gets unmade a week later. An eight-person invite for a two-person problem burns six people's afternoon. Tomasz now asks two questions of every invite list: does each person here have something to contribute or a decision to make, and is anyone whose context we actually need missing? If someone only needs to know the outcome, they belong on the follow-up email, not on the call.

The meeting ends without a stated outcome. A good discussion happens, everyone nods, and people walk out with four different beliefs about what was decided. The recap email helps, but it arrives too late to correct a misunderstanding that formed in the room. The fix costs ninety seconds: before anyone leaves, say what was decided, why, who is doing what, and what happens next, and let people correct you on the spot. The written follow-up should confirm what the room already heard, not reveal it.

Your Judgment Checkpoints

AI can draft the agenda, capture the words, and format the action items. These five checks are where you, and only you, decide whether the meeting is worth holding and whether it worked.

• Purpose. Is this genuinely a meeting, or is it a written update wearing a meeting's clothes? Name the type before you book it.
• Decision authority. Are you facilitating or deciding? Say which, out loud, before the discussion starts.
• Participant fit. Does everyone in the room need to be here, and is anyone critical missing?
• Time allocation. Is the agenda spending its minutes on what actually matters, or has the easy topic swallowed the hard one again?
• Follow-through. Does everyone leave clear on what was decided and who owns what? If you cannot state it in two sentences, neither can they.

Related Lessons

Meeting management touches most of the collaboration work a manager does. Three lessons sit closest to this one.

• Difficult Conversations Preparation covers the meetings this lesson deliberately leaves alone. When a conversation is sensitive, the cost audit and the notetaker matter far less than how you prepare for the human dynamics, and recording is often exactly the wrong call.
• Cross Team Collaboration Support extends the same practices past your own team. The cross-team coordination call where Tomasz kept seeing the same three unowned blockers is a collaboration problem as much as a meeting problem, and RACI is how the two connect.
• Workshop and Brainstorming Facilitation handles a meeting type with different rules. Generative sessions need divergence before convergence, which is a facilitation dynamic that a tight time-boxed decision agenda would actively damage.

Key Takeaways

• Name the purpose before you book the room. Decision, alignment, problem-solving, status, or relationship: the purpose dictates the format, and getting it wrong is what wastes everyone's time.
• Let AI prep, capture, and follow up; you decide and facilitate. Agendas, transcription, action-item extraction, and recap emails are mechanical work AI does fast. Purpose, participants, and reading the room are yours.
• Assign every action item with RACI. Exactly one Responsible and one Accountable per task, with due dates, is what turns a meeting's commitments into completed work.
• Announce recording every time. Transparency about notetakers, plus an off-the-record option, preserves the psychological safety that makes people raise real concerns.
• Default status meetings to async. If it can be read, write it. Reserve precious synchronous time for problem-solving, sensitive talks, and connection.
• Put a dollar figure on recurring meetings. Attendees times hours times a loaded rate exposes the real cost and gives you honest evidence to keep, shorten, or cut.

Frequently Asked Questions

Is it rude to use an AI notetaker in meetings? Not if you announce it. The problem is never the tool; it is the surprise. Say at the start that you are recording for accurate notes, promise to share the summary, and invite anyone to go off the record. Done openly, people relax because they know the action items will be captured fairly and they will not have to take notes themselves.

How do I cut a recurring meeting without upsetting the people who run it? Lead with the cost audit and the alternative, not with the cancellation. "This sync costs us about $29,000 a year of our time and it is mostly status; I want to move updates to a written thread and keep a short slot for the real blockers" is a proposal grounded in numbers, not a complaint. Offer the async replacement so people see their need is still met, just more cheaply.

Can AI just run my meeting for me? No, and you should not want it to. AI handles the words and the paperwork: agenda, transcript, action items, recap. It cannot judge whether a decision is right given context only you hold, manage the group dynamics, or sense when to slow down because someone is uncomfortable. Treat it as the fastest meeting assistant you have ever had, not as the facilitator.