Why I am not in favour of AI meeting notes

It might be more true to say I abhor them, but, as with all strong global statements one soon finds fault. What I do believe is that, as with the majority of unfocussed LLM usage, they allow for the abdication of responsibility, and the absence of clear intent in action.

Here is what I've observed:

1. Participants do not have to concentrate on the meeting, because they believe the note taking is handled.
2. I see people multi-tasking when they should be paying attention, or politely declining the meeting.
3. The meeting is had, agreements are reached, everyone believes the notes have been accurately taken.
4. The transcription and notes are rarely verified, or, if they are they are read sparingly, without attention to detail. Errors are rarely caught.
5. People have an LLM ingest the meeting notes, and write a plan for the next phase of the operation.
6. Plans are made on errors, that are compounded.

The errors I've seen run the gamut of work assigned to the wrong person, sewing confusion and wasting time with the correction, incorrect contract valuation, and outright hallucinations of important technical details (eg Intune versus Iru, two very different device management platforms).

Every so often this is not an issue, but compound it at the scale of intensity that a modern business operates at, and it becomes a serious issue.

What should we do, then?

I love writing by hand - denken mit dem hand - to quote Leuchterm, and believe that a pen or penciil and paper is a peak technological device we cannot improve on. Here's what happens when I hand-write my meeting notes:

1. I am responsible for my own notes: if I fail, I have wasted the time of myself and others, and time is the one currency we cannot control. I have focus.
2. A point is made, the hand is slower than the mind: I request a pause while I take the note.
3. I may read the note back to confirm accord.
4. At the end of the meeting, I have a tight, accurate list of actions from the meeting. Mine, other's, verified, ready to go, error-free.
5. By dint of writing, I rarely have to refer to my notes: the pause, the focus on the writing, forces me to focus and build the memories.

Where I value AI meeting notes

I have therapy every fortnight. I record each session, transcribing with diarisation using ElevenLabs. I upload each transcript to a chat in a Claude project, and Claude writes a summary of the transcript, which is saved to the project memory and files (a RAG).

What's the difference?

With therapy, I'm not tracking data, eg a $3 or $5m contract: I'm tracking sentiment shift over time. It's a course-grained guide to my shifting feelings, something I believe LLMs excel at. This is not the abdication of professional responsibility in the face of strategic or tactical planning, but a chemical ebb and flow of a personal tide.

Both my therapist and I take advantage of it: we will call out to Claude to highlight a piece we want to retain for later focus. Careful, intentional use of the tool.

As the sessions happen every other week, not perhaps six or more times a day as happens in a professional environment, I have the time to review after for completeness and accuracy. This, in itself, is a valuable review of the material.

So, cancel all AI meeting notes?

No, but it is critical that the tool is used with intention, and not abdication of responsibility.