Almost Timely News: 🗞️ Creating A Keynote Speaker Second Brain (2026-09-13)

Almost Timely News: 🗞️ Creating A Keynote Speaker Second Brain (2026-09-13) :: View in Browser

The Big Plug

✍️ Enroll in my new course, AI for Writers and learn how to make AI write better.

Content Authenticity Statement

75% of this week’s newsletter was made by me, the human. There’s a purely AI-generated glossary of all the technobabble nerd terms at the end since this week’s newsletter is VERY technical. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.

Watch This Newsletter On YouTube 📺

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What’s On My Mind: Creating A Keynote Speaker Second Brain

Last week was fun and high level, looking at what I’d tell a PR practitioner. This week, let’s get down and nerdy.

A few weeks ago, Wil Reynolds shared a short video of working in public, as he built a system that looks at his master slide catalog and helps him assemble new talks. The backbone was a simple cosine similarity measure, which I created a short YouTube video about when he said he wasn’t sure what the heck cosine similarity was.

I liked the idea, but I have different problems than Wil - I have a TON of slide decks across many, many different topics and my slides (thanks to folks like Garr Reynolds, Nancy Duarte, etc.) are mostly images because I don’t want words on a slide competing with words coming out of my mouth when I’m on stage. Our language processors in our brains can only handle one language stream at a time, so your brain has to decide whether you want to read or listen, and if a slide is a wall of text, then you’ll tune me out.

When I quickly eyeballed how many decks I have on my topics over the past couple of years, I have something like 7,000 different slides. Yes, many of them are duplicates, but many are not and they vary wildly by content type. Before I can engineer a system to help me put together a talk, I have to know what I even have.

That’s what I built this week, and I’ll share with you how I did it.

Part 1: Planning

Before we dig into the tech, we have to know what the heck we’re doing. Why am I doing this? What benefits am I after?

To do this, I typically use Katie Robbert’s 5P Framework by Trust Insights™ as it’s the best way to think through this sort of thing and decide what’s important. The 5P Framework by Trust Insights™ is purpose, people, process, platform, and performance.

• Purpose: To create a meta-index of every slide I’ve presented in the last 2 years so that I can quickly put together new slide decks with exactly the right materials. Given a session description and an interactive interview, I’d have an AI tool assemble a manifest and then a Google Slides deck of a new deck quickly, from a short session to a two day workshop. It will also highlight where I have gaps that I need to fill.

• People: Me in terms of who’s doing the work, and my audiences (you) as the beneficiaries. I have to make sure whatever tool I assemble creates slide decks that audiences actually benefit from.

• Process: Right now, when I assemble a slide deck for a talk, I typically look at the most recent talk I’ve given on the same general topic and modify it from that. That’s been reasonably effective but it takes a long time AND if there’s a particular quirk I want to focus on, this process misses that. My new software should take in the session description, an interactive interview with me about what I want to present, then hit up the master catalog to find the appropriate materials based on themes, topics, and keywords - with the understanding that some slides may be thematically appropriate but wildly out of date. When we hit a situation like that, it has to flag that slide for me to outright replace.

• Platform: For an application like this, we’re not assembling net new slides per se, merely indexing the data that already exists. That means we don’t need expensive, state of the art AI - local AI like Qwen3.6 running on a Mac is more than enough to do indexing and cataloging. I’ll build the infrastructure with a frontier model, but I’ll run it locally.

• Performance: In the short term, a working application is the definition of done, which means a knowledge catalog of all my slides. In the long term, time savings AND increased audience satisfaction are the benchmarks, but for today, a working application itself is our measure of success.

If you sit down and go through the 5P Framework by Trust Insights™, it’s pretty much your entire project plan in a nutshell. This plus an interactive interview from the AI coding tool of your choice is enough to build requirements, a technical spec, and a workplan.

Part 2: Slides Prep

In generative AI, there are three fundamental kinds of language model, a straight up text model (LLM) that only does text, a vision language model (VLM) that does text and images, and omni models that do text, images, video, audio, etc. Each type takes progressively more resources to run and takes longer to generate outputs, so if I wanted to index all of my slides, I’d need a way to convert the slides to basic images first and then have AI write up each slide. this way I can use a smaller model that only needs to be able to see images and not a huge omni model that would cost me an arm and a leg

Doing slide conversion to images doesn’t require AI. Every desktop/laptop computer has what’s called a shell, or a terminal, or a command line/command prompt. Every computer can run hundreds, if not thousands, of free, open source software packages to do all kinds of tasks like read slide decks and convert them to images.

For my task, there‘a a free office suite called LibreOffice that can open and read PowerPoint files (PPTX) really well. LibreOffice can also save decks as images, and it has a command line version called soffice that you can write scripts against, meaning you don’t have to go through the painful manual labor of opening each slide deck, hitting Save As and saving it as images. I wrote a very simple shell script for my Mac that says find all the .pptx files in this directory and open each one in soffice and save it as images in a subfolder.

This took probably under 15 minutes for it to do - it’s boring document conversion.

Could I have done this with AI? Sure. Tools like Claude Code and ChatGPT Work can open PPTX files and convert them as well, but behind the scenes they’re doing exactly what I did, only they’re chewing up your usage limits to do it. I’d rather skip the middleman (middlemachine?) and do it myself, or barring that, have AI write the script for you, but then you run it yourself. The bonus of doing it this way is you have a piece of static, reusable software for the next time you do the task. And if you have a good process down, you