Transcripts and the quest for sovereignty

I have an obsession with transcript sovereignty. Have I taken it too far?

When Google Meet started transcribing calls properly a year or so ago, I was obsessed. “Can you believe it, it summarises action points and everything!” If I didn’t get that sacred summary at the end of the call, to me the information was lost. A true waste.

18 months later and I refuse to read summaries. I’m now a cynic wearing a tin hat preaching about transcript purity & interpretation. 

What’s changed?

🔸🔸🔸

My initial obsession was close; it’s not the summaries which are gold, it’s the transcripts. This is the raw data which holds the key information. Everything you need is in the transcript, and the more you have the richer the data. The more you gather though, the more important the interpretation becomes.

Have you ever read meeting minutes and thought that’s not what I heard. AI summaries are the same thing, except it’s more important than that. If I were to defer all my interpretation of meetings to an AI, who is the COO? The summary is the narrative. Who is setting the narrative?

Although the AI summaries are good, and they can pick out what’s important in this call, there is no coherent thread between meetings. There is no joining of dots, it’s all just in isolation. And that’s where it stops. 

Why are we doing this and who is accountable? I need people to understand why. I don’t want to lead an organisation where it’s all tick boxes and do as you’re told - by a human or AI. 

When reading these auto-generated summaries, I found myself getting annoyed. They simply generate a growing number of arbitrary lists. Where are the lessons, the principles, the core of what we’ve been saying? Ultimately this is what will teach people to operate independently with initiative & alignment, not another checklist.

I began thinking tirelessly about how I can gather transcripts, clean them and interpret them. I want to apply my own lens to the interpretation, not have it decided for me. If I were to advise someone on how to approach this, I’d tell them the most important part is the interpretation of the transcript. That’s the key, the rest is workflow optimisation which can follow once you’ve done the one thing.

So I became obsessed with how to clean the transcripts.

🔸🔸🔸

If I’m using ChatGPT one week & Claude the next, then everything needs to be in a universal format in a centralised location. In the early days I’d have deep chat threads and then realise that it doesn’t exist outside that narrow thread. It’s not searchable. It’s not shareable. It’s not cross platform. Every conversation is in isolation. 

And although you can have conversations across chat threads now, you can’t define the weighting. 

How does it know which is the prevailing thought that you had? Is the most recent conversation more or less important than a conversation I had last month? 

This drives me the most insane. I once said I’m ruthlessly effective and now it quotes this back to me whenever it can, particularly when I’m asking it to review a transcript for advice on what I can work on. 

This is exactly the sort of question a ruthlessly effective leader should be asking…

The irony of this is not lost on the man who spends hours a week fixating on transcripts.

The next level of my insanity is token consumption. This transcript library is going to only keep growing. The more data, the more expensive it will become to search. How do I make these various transcripts from different sources as few bytes as possible? 

So I got AI to write me a script to clean and file the transcripts. And then I got that script to append the clean transcripts to a master transcript so it’s searchable. And now it does it depending on the topic of conversation and the role I’m playing. 

I pride myself on simple solutions and focusing on what’s important.

I use ChatGPT for iteration. Google Drive for storage. Google Apps Script for execution. Gemini when I forget that I use ChatGPT. OpenClaw for fun. Codex within OpenClaw. Antigravity within Cloud Shell.

I’ve built a complex web which I need to untangle. My model agnostic approach is at least allowing me to play with all the tools though, and I’ve still got Claude Cowork to come...

🔸🔸🔸

I’ve obsessively cleaned the data, now I need to do what I should have done in the first place and focus on interpreting it.

Stoirm