Capture What the Board Decided Without Recording the Room

I sit on the board of ACT | The App Association, and as CEO of Visiting Media I report to a board of my own. Anyone who has spent time in either kind of room lately knows the moment. A tile labeled “Notetaker” appears in the grid, and a director who was halfway into a blunt sentence finishes it with something safer.

That flinch is the whole problem.

I wrote about how those bots get in, usually through one director’s calendar default, in AI note-takers in board meetings, and my answer there was to keep them out. This piece is about the harder half. Once the bot is gone, somebody still has to capture what the board decided, and the directors have to trust that capture enough to keep talking plainly while it happens.

What the minutes guides already agree on

The guides that rank for this question (Diligent’s long walkthrough, Atlas Governance’s step-by-step, BoardPro’s roundup of what its community of secretaries actually does) agree on nearly everything. Minutes record decisions, motions, votes and the actions that follow, plus who presented and what the board considered. They summarize. They leave out who said what. Verbatim transcripts are a risk, and so is anything privileged that drifts into the general record. BoardPro’s contributors want a draft within 24 hours, and Atlas wants it written soon after the meeting, then circulated, approved and stored.

Good advice. Almost all of it happens after the meeting, at the secretary’s desk, out of the directors’ sight. Diligent does suggest reading back a portion of the notes to check accuracy, in a single line of a very long page. I think that line should be the method.

Why the room goes quiet

A recording makes people careful for a reason that has little to do with the microphone. Nobody in the room knows what will be kept or who will read it, and any sentence might land in a summary emailed to the whole invite list, or two years later in a plaintiff’s exhibit binder. When the scope is unknown, the safe move is to say less.

Human minutes have the opposite property. Their scope is narrow and agreed in advance: what was decided, who owns it, by when, and enough of the reasoning that a director reading them two years later can follow why. The discomfort goes away when the room can see that narrow scope while the meeting is still going.

Close every agenda item out loud

At the end of each item, the chair stops and says the decision in one sentence, with an owner and a date. “We approved the revised budget, and the CFO brings the hiring plan back in March.” Then the chair asks whether anyone heard it differently.

Thirty seconds, maybe. It is also where most bad minutes get prevented, because a director who believed the board approved the plan and a director who believed it approved the plan pending one more look will discover they disagree right there, with everyone present, instead of three weeks later when the draft minutes go around and the memory of the discussion has gone soft.

Some chairs find it awkward. Do it anyway.

Put the decision line on the screen

None of the guides I read describe this part. Whoever holds the pen, usually the corporate secretary, types the decision sentence into the board portal or a shared document and puts it on the screen as the chair says it. Directors see the exact words going into the record. They can fix a verb. They can also see what is not being written down, which is everything else they said.

Visible capture and invisible capture feel completely different to the people being captured. The bot hears every word and keeps every word, and the room finds out what it made of them an hour later. A secretary typing one line on a shared screen also hears every word, keeps one line, and the room watches that line get chosen. I can’t picture a director lowering their voice for that.

The full minutes still get drafted afterward, the way the guides describe. The draft just starts from sentences the board has already looked at.

Keep the action register apart from the minutes

Minutes are the legal record of the board and should read like one. Follow-ups belong somewhere else: a short table with the task, the owner, the date, and whether it is done. When the two share a document, the minutes fill up with operational chatter and the actions get approved once a quarter and then forgotten. I would have the secretary build the register straight from the decision lines on the screen and put it at the top of the next meeting’s agenda, before anything new.

When the chair forgets

The secretary is allowed to stop the meeting.

Where my agents fit, and where they don’t

At Visiting Media a fleet of specialized agents chases the follow-ups I would otherwise forget, so I am the last person to argue that software has no place near a board. It has exactly one place in this process. After the minutes are approved, an agent can take the action register, nudge each owner as their date gets close, and draft the status column for the next pre-read. It works from a document the board signed off on. It never hears the room.

I care about that boundary more than the convenience. An agent reading a transcript is acting on what people said out loud on a Tuesday. An agent reading approved minutes is acting on what the board decided, and only the second carries any authority, which is the whole point of a delegation of authority matrix. It keeps the record clean, too. In AI agent audit trails I argued for storing a pointer to an agent’s input instead of the input itself, and here the input is a short document with an approval date on it. Nobody has to keep an hour of the board’s audio around so a reminder bot can do its job.

Try it at the next meeting

Tell the directors at the start that nothing is being recorded and that the record will be the decision lines on the screen. Close each item out loud. Save the last five minutes to read the lines back in order (the whole list usually fits on one screen) and let anyone object to a word before the meeting ends.

Then watch the director who went quiet when the bot showed up last quarter. If the blunt sentences come back, it is working.

The same habit travels to management’s meetings, where note-taking bots run all day on accounts nobody counted. That count belongs on the list of what boards should ask about AI, and a board that captures its own decisions this way is in a much better position to ask it.

Related Reading

• AI Note-Takers in Board Meetings: The Bot Nobody Invited
• What Boards Should Ask About AI: A Director's Oversight Agenda
• AI Agent Audit Trails: Keep Logging After the Agent Stops
• Board Governance for Tech Companies: Lessons from the Front Lines