How I keep track of dozens of conversations without taking a single note

A lot of my work is conversations. When I help an organisation adopt AI, the first phase is discovery, and discovery means talking to people: stakeholder interviews, leadership one-to-ones, vendor calls, team sessions, often across several sites and time zones. In a single engagement I might run dozens of them. The raw material of the job is talk, and the hard part has always been the same. Capturing it, making sense of it, and not losing anything across weeks of it.

For years I did that the way most people do. Scribble notes while half-listening, type them up later, hope I caught the important bit. It works, badly. You are either present in the conversation or you are taking minutes, and you cannot really do both.

Here is how I work now, because it has changed the job more than almost anything else I have adopted.

The setup

I work alongside an AI assistant that can see my working files and my meeting transcripts. The piece that gets conversations into that system is a transcription tool called Krisp. The two are connected directly, through an integration, and that connection matters more than it sounds. I will come back to why.

The loop

It runs like this:

1. I record the meeting with Krisp: a client call, an interview, a vendor pitch. Because it is recording, I can be in the conversation rather than scribbling.
2. Krisp transcribes it and gives me a clean, speaker-labelled transcript, plus notes and action items.
3. Because Krisp connects to my assistant through an integration, I do not copy and paste anything. I point the assistant at a meeting and it pulls the full transcript by reference, or searches across every meeting I have recorded.
4. The assistant turns that raw transcript into something usable: a structured synthesis of the themes, the decisions and the actions, an update to the running project memory, a drafted follow-up email.
5. Over an engagement, every conversation becomes part of a searchable corpus. I can ask "what did everyone say about onboarding" and get a real answer. I can see that three different people independently raised the same problem. I can track who committed to what.

That last point is the one that changed things. A single transcript is useful. Forty transcripts that an assistant can read across, connect and question is a different kind of thing entirely.

What it actually changes

I am present. I am in the conversation, not taking minutes. The recording remembers so I do not have to.

Nothing important gets lost. A full transcript beats notes I half-caught while thinking about my next question.

There is no friction between talking and processing. The integration is the quiet hero here. No copy-paste, no uploading files, no "I will sort this out later" pile that never gets sorted. The conversation flows straight to the place I will actually use it.

I have a second brain for the engagement. Dozens of conversations, captured, synthesised and cross-referenced. It is what makes a big discovery tractable instead of overwhelming.

If I am honest, the feeling is the best part. Holding conversations across a whole business, on a dozen different topics, and having something quietly keep track of all of them, connect them and tell me what needs doing, is astonishing. The relief of not having to hold it all in your head is hard to overstate.

The honest bits

I would not write this without the caveats, because the caveats are where people get caught out.

Transcription is not flawless. Speaker labels can be wrong, especially on a busy call, so I confirm who "Speaker 2" actually was before I rely on it. The odd word gets garbled. It is raw material, not gospel.

It still needs judgement. The synthesis is a draft. I check attributions and verify anything sensitive against the actual transcript before it goes anywhere. The value is the tool plus the human, not the tool on its own. That is the same rule that runs through everything I write here: AI does the production, your judgement is what makes it worth keeping.

Consent and confidentiality matter. Recording people carries an etiquette, and sometimes a legal dimension. Participants should know they are being recorded. Confidential material has to be handled with care, which means thinking about where transcripts of sensitive conversations are stored and processed. Better to be deliberate about it up front than to find out the hard way.

How to set up your own AI meeting workflow

You do not need my exact stack. The pattern is what matters, and it has three parts.

1. A capture layer. A transcription tool that produces a clean, speaker-labelled transcript. Krisp is what I use; there are others.
2. An assistant that can read the transcripts. The magic is in step three of the loop, the assistant working from the actual words rather than a summary. An integration that lets it pull transcripts directly, instead of you copy-pasting, removes the friction that kills most good intentions.
3. A discipline. Consent up front, a human check on the synthesis, and care with anything confidential.

Start with one recurring meeting. Record it, get the transcript into your assistant, and ask it for the decisions and the actions. Once you have felt the difference between that and your own notes, you will not go back.

If your real work happens in conversations, and for a lot of us it does, this is the highest-leverage change I can point you to.

FAQ

Do I still need to take notes if I record the meeting?

No, and that is the point. Recording with a transcription tool means you can be present in the conversation instead of scribbling. You still review the transcript and the synthesis afterwards, but you are checking, not capturing.

Is AI meeting transcription accurate?

Mostly, but not perfectly. Speaker labels can be wrong on a busy call and the odd word gets garbled, so treat the transcript as raw material and confirm anything that matters against the recording. It is good enough to work from, not good enough to trust blindly.

What do I need to set up an AI meeting workflow?

Three things: a transcription tool that produces a clean, speaker-labelled transcript, an AI assistant that can read those transcripts (ideally through an integration, so you are not copy-pasting), and a little discipline around consent and checking the output.

For the nuts and bolts of turning a transcript into clean notes and actions, I have written that up separately: how to use AI for meeting prep, notes and action items. And if you are trying to make this way of working stick across a team rather than just for yourself, that is the kind of thing I help organisations do.