The Death of Manual Meeting Minutes

There is a small, oddly resilient category of professional work that hasn't changed since the 1980s: writing meeting minutes by hand. Executive assistants do it. Project managers do it. Boards of directors require it for governance reasons that have not been reconsidered in 40 years. And almost everyone who does it agrees the work is tedious, low-value, and could be done another way.

The "another way" is finally here. It is not a meeting bot — those introduce their own problems (we wrote about them for the healthcare case, but the issues apply more broadly). It is a much simpler primitive: record the meeting normally, upload the file afterward, get a structured brief.

This post is for the EA, PM, or chief-of-staff who still types meeting minutes during the call. Here is what changes when you stop.

Why manual minutes survived this long

Three reasons, and they are worth naming:

1. Governance documents — board minutes, ethics-committee minutes, regulator-facing minutes — have specific format requirements and an evidentiary role. Until very recently, AI output wasn't structured enough to be defensible in front of an auditor.
2. Confidentiality posture — uploading audio to a third-party transcription service crossed a security line that most legal teams weren't willing to approve.
3. The note-taker as a power role — in many organizations, the person taking notes is also the person who shapes the canonical record of what was decided. That's not a coincidence, and it's not something every organization is willing to give up.

Three things changed simultaneously:

1. The output got good. Modern transcription pipelines produce structured Markdown that holds up next to a hand-typed minute book. They identify decisions, owners, and dates. They surface objections. They differentiate between "the team agreed" and "X said they would think about it".
2. The data posture got better. Transcription services with explicit no-training contracts, regional data residency, and SOC 2 / ISO 27001 attestations are now the norm, not the exception.
3. The note-taker role evolved. EAs and chiefs of staff stopped wanting to be the human stenographer and started wanting to be the human curator — and the curator is a better, higher-leverage role.

What an AI-generated brief actually looks like

For a typical 60-minute executive meeting, the output is about one printed page and is structured as:

Summary (3 sentences, top of the page)

The Q3 forecast came in 6% under plan, driven primarily by a delayed launch and an unfilled VP Sales role. The team agreed to pull two engineering features from the H1 roadmap and reallocate to the GTM motion. Final Q3 numbers and reforecast will be presented at the next ELT in two weeks.

Decisions

• Pull feature_a and feature_b from H1 roadmap; reallocate to GTM.
• VP Sales backfill is the #1 hiring priority for Q3.
• Marketing budget moves from quarterly to monthly review cycle.

Action items

• Priya — share Q3 reforecast by Tuesday.
• Marco — draft the revised H1 engineering plan and circulate for ELT review by Friday.
• Sarah — open the VP Sales role this week and shortlist three search firms.

Discussion notes (full diarized transcript with timestamps, ~6 pages, collapsed by default)

The most useful row is the action-items list, which is also the one that humans most reliably get wrong when typing minutes live. The human attention cost of typing has a known failure mode: under load, you write down the decisions and miss the action items. The AI doesn't have that failure mode.

What it does not replace

• Governance minute books that need a specific evidentiary format. (Most AI briefs are now structured well enough to use as the source-of-truth, but a human still signs and finalizes.)
• The judgment call about what's important. The AI surfaces the decisions; the EA or chief of staff decides which ones need follow-up emails, which need a Slack post, and which need to be quietly buried.
• Real-time facilitation. AI doesn't replace the person who keeps the meeting on time. That's still a human role.

The new EA / chief-of-staff workflow

The pattern that's worked in every team we've seen:

1. Pre-meeting. EA confirms the meeting will be recorded. (Most platforms now make this a one-click default.)
2. In the meeting. EA does not type. EA listens, asks one or two clarifying questions when something is left ambiguous, and notices what's not being said.
3. Post-meeting (within 30 minutes). EA uploads the recording. The AI brief is ready by the time the EA's coffee is.
4. Curation pass (5–10 minutes). EA rewrites two or three sentences, adds names where the AI labeled "Speaker 3", deletes one or two action items that were aspirational rather than committed.
5. Distribution. EA sends the brief to the meeting participants, posts the action items to the project tracker, and archives the file.

The total per-meeting EA time goes from ~70 minutes (1 hour in the meeting + 10 minutes for minutes) to ~75 minutes — almost identical — but the value of those 75 minutes shifts from stenography to curation. That is a much better use of an EA's day.

What to evaluate before you switch

If you are responsible for this decision, three questions:

1. Does the vendor sign a BAA / DPA and give us our data? If yes, you can pilot. If no or vague, hold off.
2. Is customer audio used for foundation-model training? The answer should be no, in writing.
3. Where does the audio live? US-only, EU-only, or in-region — for many enterprise customers, the answer matters.

If you can clear all three, the pilot is straightforward: pick one recurring exec meeting, run it for four weeks, and compare the AI brief against the hand-typed minutes at week four.

We've never seen the hand-typed minutes win.

AudioToNotes is built for this workflow. Join the waitlist for early access.