How to use AI to avoid meetings — not just summarise them.

01   The pitch is backwards

Every vendor is selling the same promise in slightly different shapes. Microsoft sells Copilot, Zoom sells AI Companion, Google sells Gemini in Meet, and a long tail — Otter, Fathom, Granola — sells the same thing with a different brand: an AI that listens to your meeting, writes the notes, and tags the action items.

That product is real. The claims about it mostly aren’t. And the way it’s pitched — “AI will save you from meetings by taking notes” — is backwards. A note-taker doesn’t save you from a meeting. It saves you from the ten minutes after the meeting. That’s worth something, but it’s roughly 2% of the cost of the meeting itself.

The leverage — the reason to care about AI in the context of your calendar — is not in assisting meetings. It’s in replacing them.

02   What the 2025 data actually shows

The single most useful piece of data I’ve read on this is Microsoft’s own 2025 Work Trend Index, “The Year the Frontier Firm Is Born.”The headline numbers are damning — for the “AI saves you from meetings” pitch.

All four of those numbers come from Microsoft’s own telemetry. Three years into the Copilot era. In the same report, Microsoft cites a customer — Husch Blackwell — whose team “saved an estimated 8,800 hours” using Copilot. That may be real. It’s also a customer anecdote from the vendor selling the product. It is not the answer to why Microsoft’s own platform data shows meetings growing, not shrinking, during the Copilot era.

The honest reading: AI meeting assistants are making meetings more processable, not less frequent. The note-taker is a coping mechanism, not a fix.

03   The replacement lever

What the peer-reviewed research actually supports is something more useful, and more specific: AI collapses the cost of producing written work. Noy and Zhang’s 2023 Science paper found that professionals given ChatGPT access completed occupation-specific writing tasks in 40% less time with 18% higher quality. Brynjolfsson, Li and Raymond’s NBER 2023 paper on customer-support agents showed a 14% productivity gain (34% for novice workers).

Neither of those studies measured meetings. That’s fine — because the leverage isn’t meetings. It’s written artefacts. A forty-percent reduction in the time to produce a decent written memo is the whole basis for the six-pagerworking at organisational scale. It’s the whole basis for cancelling a recurring review meeting and replacing it with a written proposal. The async move, which used to require one person sitting down for a long morning to write the doc, now takes forty minutes.

That’s the replacement lever. Not smarter transcripts. Faster proposals.

04   Three moves that work

The following are what I’ve seen land, in practice, with teams I coach. All three treat AI as a sync-to-async translator. None of them are note-takers.

Move 1: Kill the status meeting with an AI-drafted weekly update

Every week, paste your notes, commits, closed tickets, or customer transcripts into Claude/ChatGPT/Gemini and ask it to produce a two-paragraph update. Post it to the team channel. Offer to answer questions. The status meeting dissolves inside three weeks because nobody comes to it — the update already covered everything.

Move 2: Kill the review meeting with an AI-summarised thread

Before you schedule the meeting to “sync on this”, paste the relevant thread / doc / Slack channel into an LLM and ask for a one-page decision brief: the question, the options, the trade-offs, the recommendation. If the brief resolves the question, you don’t need the meeting. If it doesn’t, you now have a brief to read in the first five minutes of a much shorter meeting.

Move 3: Kill the brainstorm with AI-triaged proposals

Ask your team to send their ideas in writing, ahead of time. Feed them into an LLM, ask it to cluster duplicates, steel-man the strongest three, and pressure-test the weakest two. Walk into what would have been a brainstorm with a ranked short-list. Anthropic’s own engineering writeup (“How Anthropic teams use Claude Code”) describes the same pattern in a different domain — Claude as a thought partner, not a note-taker.

05   The meeting-theatre risk

There’s a specific failure mode I watch for when teams deploy meeting AI: the meetings don’t shrink, they just feel better. Notes are captured. Action items are tagged. A tidy summary lands in Slack. Everybody nods. The meeting keeps its slot next week.

This is accountability theatre with prettier props. The real test of whether AI is helping your calendar is not whether the notes are better. It’s whether the meeting count went down. If after three months of Copilot/Otter/Fathom deployment your calendar looks the same, you haven’t saved time — you’ve just added a new software cost on top of the unchanged meeting cost.

Microsoft’s own 2025 data is the sharpest evidence of this pattern in aggregate. Three years of Copilot, and ad-hoc meetings climbed to 60% of all meetings. The tool is not saving people from the calendar. It’s helping them cope with a calendar that’s still getting worse.

06   What to try this week

If you already pay for an LLM or have access to one at work, the experiment is 30 minutes long.

1. Pick one recurring meeting you suspect is a status meeting in disguise.
2. Before this week’s instance, paste the relevant artefacts (tickets, metrics, threads) into Claude/ChatGPT/Gemini. Ask it to draft a two-paragraph update covering what’s moved and what’s blocked.
3. Post the update to the team channel 24 hours before the meeting.
4. At the meeting, ask one question: “What did the update miss?” If the answer is “nothing,” cancel next week’s.

That’s the whole technique. AI isn’t saving you from meetings. You are — with AI as a lever that makes the async alternative twice as fast to produce. The meeting assistants are useful, at the margin, for the meetings you still need. The leverage is in the meetings you stop needing.

• Are AI meeting notes actually useful?−
  Yes — at the margin. If you're in a meeting that genuinely needed to happen, getting a clean transcript and action list saves 10–15 minutes of post-meeting note-writing. Zoom's own ROI study found net ~1.1 hours/week saved for heavy users. That's real but small. It's not what the marketing pages claim.
• What's the 'meeting theatre' risk?+
• Where does the Brynjolfsson 14% productivity stat fit?+
• So what should I actually do with AI?+
• What about Claude / ChatGPT / Gemini specifically for decision-making?+

Sources

1. 012025 WTI — The Year the Frontier Firm Is Born · Microsoft Work Trend Index2025
2. 022024 WTI — AI at Work Is Here. Now Comes the Hard Part · Microsoft Work Trend Index2024
3. 03Experimental evidence on the productivity effects of generative AI · Noy & Zhang · Science2023
4. 04Generative AI at Work (NBER w31161) · Brynjolfsson, Li, Raymond · NBER2023
5. 05Navigating the Jagged Technological Frontier · Dell'Acqua et al. · Harvard/BCG2023
6. 06Zoom AI Companion ROI Dashboard (n=688 knowledge-worker study) · Zoom2024
7. 07How Anthropic teams use Claude Code · Anthropic2025
8. 08Asana Meeting Doomsday · Asana2024