Almost everything that ruins a meeting is decided before anyone walks into the room.
TL;DR: In January 2023, Shopify deleted every recurring meeting with more than two people, and the instructive part was watching which ones people fought to get back. A meeting is a symptom: the damage was done days earlier, by someone dragging a block onto a calendar. Overstuffed invite lists are what an organization that writes nothing down looks like. If your AI avatar could attend without loss, that is a document with catering. The undelivered action item costs more than the silent attendee; it books the next meeting. Judge your AI by the meetings it prevents, then cancel one and watch.

One email, one morning, and thousands of hours handed back. In January 2023, Shopify canceled all recurring meetings with more than two people, then told the company not to put anything back for a fortnight. Around the same time, Microsoft was measuring something funnier: across its own products, edits to PowerPoint files spike 122% in the final ten minutes before a meeting begins, an entire professional class cramming for an exam it scheduled itself. Everyone dunks on meetings. It is the safest joke in business, and the cheapest, because it never asks what the meeting was compensating for. So take this as a defense of meetings, which means it is going to cost you several.
We can describe a good meeting with unusual precision, which is a strange skill to have developed about something we claim to hate. Agenda, objective, the right people and only those, everyone participating, a decision taken. We can describe the other kind even better, because most of us were in one yesterday. You wondered why you were invited. Nothing to read beforehand. An hour in, nothing had moved. Your whole team was in there with you, which felt less like collaboration and more like a hostage situation. And half the faces were lit from below by a laptop screen, furiously answering email (or doomscrolling Instagram).
Almost none of those failures happened in the room. Each was settled days earlier by somebody dragging a block onto a calendar and adding names (each name costs the person adding it nothing and the person receiving it an hour). They keep getting added for a rational reason. When the only dependable way to learn what is happening is to sit in the room where it gets said, people sit in every room they can. The overstuffed invitation list is a symptom of an organization that never wrote anything down, and we keep debugging the symptom. The average executive now spends the better part of three working days a week in meetings. Nobody decided that. It accumulated, one reasonable invitation at a time.
The most useful thing I’ve heard recently about meetings is that they exist for argument and decision. Information transfer is what documents are for. Spend forty-five minutes bringing everyone up to speed, and you have not merely wasted the time; you have made the decision worse, because the gap in shared context the meeting was called to close is still sitting in the room while people vote.
Amazon’s answer is famous and still under-copied: PowerPoint dropped in 2004 for six-page written narratives, and the first twenty minutes of the hour spent in silence while everyone reads. They moved the context out of the room, then refused to assume anyone had read it.
Which is where AI earns its place, and not by attending. It helps assemble the brief from material scattered across a dozen systems. It states, in one sentence, the decision that needs making. It proposes the agenda, and more usefully the invitation list. The best AI meeting feature I can imagine is the one that emails everybody to say: “You do not need this meeting; here is the answer. Object by Thursday.”

All of which makes one rule enforceable that was merely aspirational a decade ago. No agenda, no background, no decision to be taken? Then no meeting. Not shorter. Cancelled. That used to be unfair, because writing an agenda, a brief, and a clear decision statement cost the organizer an hour they did not have. That excuse has expired.
Two questions before you accept anything. Could this have been an email (yes, I am old-school), a Slack message (I am adapting), or a document? If the payload travels one way, the room is the wrong channel. And: do I move this forward? Attendance is a contribution test, not a roll call. Tobi Lütke put the arithmetic underneath both plainly: say yes to a thing and you say no to everything else you could have done with that time.
Are you ready for the uncomfortable part? Running the second question properly means occasionally declining your own boss’s invitation. You do it anyway. Carefully.
O meeting, you were never the problem.
We booked you to carry what a page could hold.
We invited the room.
Then the room’s colleagues.
Then everyone who might otherwise wonder what they missed.You asked for an argument.
We brought a status update.
You asked for a decision.
We brought a transcript.
You asked for one person to say, I own this, by Thursday.
We thanked everyone for their time and booked you again.
So what survives, once the context has moved out and the follow-up is automated? The argument. The dissent that exposes a risk nobody wrote down. The colleague who changes their mind in public, which is close to the most trust-building thing that happens inside a company. And one person saying out loud that they own the decision.
If a meeting were only information transfer, sending a convincing copy of yourself would be fine, and people are testing precisely that. An American bank had its chief executive’s AI clone deliver the prepared remarks on an earnings call. Meta and Zoom both want their CEOs’ clones in the room. Zoom’s Eric Yuan put it with admirable clarity: “I can send a digital version of myself to join so I can go to the beach.”
For prepared remarks, the clone is arguably the honest choice; a human reading a script aloud was already a recording with extra steps. The tell is where it stops. Nobody should send an avatar into a conversation where a decision is made, or into one where somebody has to be told something difficult. The line where the clone becomes unacceptable is precisely the line where the meeting becomes worth holding.
Which makes the clone a useful test. Before you accept, ask whether an AI avatar of you could attend without anybody losing anything. If the answer is yes, that is not a meeting. It is a document with catering.
Then there is the team that auto-summarizes every meeting and still cannot tell you what was decided on Tuesday. A transcript is raw material. Accountability is the product. Somewhere in the last two years, the summary got promoted above the commitment, and it has been coasting ever since.
Capturing the commitment is the easy half, and machines now do it well: who owns what, by when, against a definition of done. Landing it on the agreed date is the other half, which raises a question I find uncomfortable. Who costs you more: the person who sits silent through the meeting, or the person who takes the action item and never delivers it? The silent one costs you an hour. The other costs you the next meeting, because the next meeting exists to establish what happened to the last one. That is how a single decision becomes a standing weekly. Most calendar bloat is undelivered commitments that compound.
Better tooling will make meetings rarer and better, and not by attending them. It will do so by addressing the reasons we hold on to the bad ones. A meeting is a symptom, the visible and expensive place where unclear ownership, missing context, and undefined decision rights surface at once. Fix what sits upstream and the meeting was never needed, which is the highest praise this ode can offer. The meetings worth keeping are the ones that had to happen.
The real measure of your AI is not how many meetings it records. It is how many it makes unnecessary.
Try this: pick one recurring meeting on next week’s calendar and ask what it is compensating for. Move its preparation into a brief that lands a day ahead, protect the hour for judgment, turn its output into dated commitments, then chase the dates. Then cancel it and see what happens. The complaints are your real agenda.

Sources
Leslie A. Perlow, Constance Noonan Hadley, and Eunice Eun, “Stop the Meeting Madness,” Harvard Business Review, July–August 2017. https://hbr.org/2017/07/stop-the-meeting-madness Accessed 2026-07-25
Microsoft WorkLab, “Breaking Down the Infinite Workday,” Work Trend Index Special Report, June 17, 2025. https://www.microsoft.com/en-us/worklab/work-trend-index/breaking-down-infinite-workday Accessed 2026-07-25
Tristan Bove, “Shopify is axing all meetings involving more than two people in a remote work ‘calendar purge’ that the company itself calls ‘fast and chaotic’,” Fortune, January 3, 2023. https://fortune.com/2023/01/03/shopify-cutting-meetings-worker-productivity/ Accessed 2026-07-25
Emma Burleigh, “CEO of a $25.9 billion bank had his AI clone lead the company’s earnings call, as Mark Zuckerberg builds his own digital twin,” Fortune, April 28, 2026. https://fortune.com/2026/04/28/ceo-of-customers-bank-sam-sidhu-ai-clone-lead-earnings-call-mark-zuckerberg-building-digital-twin/ Accessed 2026-07-25
Nilay Patel, “Zoom CEO Eric Yuan Wants AI Clones in Meetings,” Decoder, The Verge, June 3, 2024. https://www.theverge.com/2024/6/3/24168733/zoom-ceo-ai-clones-digital-twins-videoconferencing-decoder-interview Accessed 2026-07-25
Colin Bryar and Bill Carr, Working Backwards: Insights, Stories, and Secrets from Inside Amazon (New York: St. Martin’s Press, 2021).
Kevin J. Delaney, “Book Briefing: ‘Working Backwards’ by Colin Bryar and Bill Carr,” Charter, February 12, 2021. https://www.charterworks.com/book-briefing-working-backwards-by-colin-bryar-and-bill-carr/ Accessed 2026-07-25
Nilay Patel, “Can Patreon Fight Fire with Social Media Fire?” Decoder, The Verge, June 22, 2026. https://www.theverge.com/podcast/952607/patreon-ceo-jack-conte Accessed 2026-07-25
Originally published at www.orionplaybook.com.

