Here is a detail that should make you pause. The people who build AI for a living — the engineers at the companies selling you a lighter workload — have more meetings than they used to. They sit in rooms "botsitting" colleagues' AI agents. They field nudges from bots designed to make them more productive. The promised calm, automated workday arrived at the companies building the technology first. It did not go well.

The reporting from inside AI-native companies reads like a gentle satire of the pitch deck. Agents everywhere, output everywhere — and humans coordinating it all in the oldest format known to work: the meeting. Plus a brand-new chore nobody listed in the job ad: watching other people's bots.

The lesson is not that AI failed. It is that AI cannot fix what was never a technology problem. Meeting culture is an operating habit — and no tool, however clever, has ever fixed an operating habit.

Why the Agents Made More Meetings

Follow the logic and it is almost inevitable. Agents multiply output — drafts, options, plans, analyses, all produced at machine pace. But every piece of output still needs a decision behind it: is this the right direction, does this get approved, who owns it. The volume of things needing decisions went up. And where do decisions still happen in most companies? In meetings.

So the AI-native office invented a new bottleneck. Machines removed the production constraint and exposed the decision constraint — which was meetings all along. Then the bots added their own overhead: reminders to review their output, pings when a delegated task stalls, attendance at standups on behalf of absent humans. The tool didn't remove work. It relocated it to the calendar.

This is the part the efficiency case always skips. Automation math assumes the saved hours disappear as pure gain. In practice they get reabsorbed by whatever the next constraint is — and inside the AI companies, the next constraint is the same one it has always been: coordination among humans who haven't agreed on how to decide.

The Fix Predates the Technology

The companies drowning in meetings don't lack AI. They lack norms. And the norms that prevent meeting-sprawl were solved problems before the first chatbot shipped — they just require the discipline most teams never install.

Write it down. The single highest-leverage habit in modern work is defaulting to documents over discussions: a written proposal people comment on replaces a recurring meeting people attend. Default async. Decisions that don't need real-time back-and-forth shouldn't get calendar real estate — most meetings are information transfer wearing a discussion's clothes. And make the meeting earn its invite: an agenda, a decision to be made, a reason the attendees specifically are needed. The meetings that survive that filter are usually excellent. The meetings that don't were the problem.

Here is the uncomfortable part for the AI era: none of this can be delegated. An agent can draft your status update, but it cannot create the culture where a written update replaces a meeting. An agent can summarise the thread, but it cannot make your team a place where decisions have owners. Norms are set by humans — usually by whoever is willing to behave differently first and make it safe for others to follow.

The Scripts Nobody Teaches You

The norms are known; the missing piece is usually the language. Two scripts cover most of the territory.

Declining a meeting, offering a doc instead. "I'm going to skip this one — I'll write up my input on [topic] in the doc by end of day so it's there for the discussion. If a decision lands that needs me, pull me in async." This works because it concedes nothing: you still contribute, the meeting still happens, and you quietly demonstrate that your presence was the optional part. Do it twice, politely, and the third invite often arrives with an agenda — or doesn't arrive at all.

Proposing the async norm. "Can we try something for the next two weeks — status updates go in the shared doc by Friday noon, and we only meet for things that actually need a back-and-forth? If nothing needs live discussion, we give everyone the hour back." Frame it as an experiment with an end date and you skip the culture war. Experiments get tried; mandates get resisted. Most async norms that stick started as a two-week trial that nobody wanted to give up.

What a Good Async Update Actually Looks Like

The objection that kills async norms is legitimate: most written updates are terrible — walls of prose nobody reads, which is how the meeting got invited in the first place. A status doc that replaces a meeting has a shape — our Async-First Productivity course teaches the full written-norm system — and the shape is learnable. One line of context (what this is), three to five lines of substance (what changed, what's blocked, what's decided), one line of ask (what you need from the reader, if anything), and a deadline for comments. Two minutes to read, thirty seconds to respond to. When written updates take longer to read than the meeting they replaced, the doc was the problem — not the format.

It is also worth saying which meetings deserve to survive, because async purists get this wrong too. Real-time earns its place for genuine conflict (text escalates, conversation resolves), for decisions where the trade-offs need rapid iteration, and for the moments a team needs to feel like a team. The goal was never zero meetings. It was a calendar where every meeting could explain why it exists.

The Meeting Audit: Twenty Minutes, Once a Quarter

For individuals, the ordering applies at calendar scale. Before you automate your calendar, audit it. Block twenty minutes, open the last month, and run three questions on every recurring meeting.

Does it have a decision attached — a real one, made in the room, that couldn't happen in a comment? Does it exist because writing things down feels harder than booking thirty minutes? And the honest one: if you declined it politely with a written contribution, would anything actually break? The meetings that survive all three questions are usually excellent. The rest are calendar debt — and unlike most debt, you can cancel it.

One more rule from inside the AI companies, learned the hard way: the bots need norms too. Every agent that can ping you will eventually ping you — status nudges, review requests, completion notices — and a dozen ungoverned notifications is just a new kind of meeting that follows you everywhere. The fix is the same as the human version: batch the interruptions (agent updates get read at set times, not on arrival), route them to a queue instead of your attention, and make anything that interrupts justify the interruption. If a bot wouldn't be allowed to tap your shoulder in a meeting, it shouldn't be allowed to push a notification.

The honest promise of AI at work was never fewer hours. It was better-spent hours. Inside the AI companies, they are learning that the hours only get better if somebody defends them. That somebody is not the bot. It is the person willing to run their week on different rules.

What makes the inside-the-AI-companies reporting so valuable is that it removes the last excuse. You can't blame your meeting load on your employer being behind the curve — the furthest-ahead companies have the same problem, worse. The variable was never the technology stack. It was always the operating norms, and norms are set by whoever behaves differently first and makes it safe to copy. That person is hireable in every industry on earth.

The timing matters more than usual right now. Q4 compresses every workplace: budget season, year-end deadlines, and the annual illusion that December somehow has the same number of days as March. Meeting-sprawl always peaks in the crunch — which makes this the highest-leverage quarter to install the norms, because the meetings you decline in November are the hours you still own in December.

And the starting move fits in this week: run the meeting audit on next month's calendar, pick the single recurring meeting most likely to fail the three questions, and use the script — skip it, send the doc, see what happens. One experiment, two weeks, an hour back. The AI-company engineers drowning in meetings can't show you the way out. But the person who declines politely and leaves a document behind? They're about to become the norm everyone else copies — and they'll have the reclaimed calendar to prove it worked.

Those rules are a learnable system, and they are the core of our 7-day email course Async-First Productivity — replacing meeting reflexes with written norms, building deep-work blocks that survive the pings, and running a week where the calendar serves the work instead of the reverse. If the end-of-year crush already has your calendar in a death grip, pair it with our piece on the Q4 pressure cooker — the burnout nobody budgets for starts exactly here. For the inbox-and-calendar mechanics specifically, our Email and Calendar Productivity course is the tactical companion.