In a meeting with our Marketing team today, my colleague Ted said, “I see you got some new note-taking software.”
Yes, but not exactly.
In the age of AI, I’ve gotten into the habit of starting every project with a README.md file, a GitHub repository, and Claude Code. Whether it’s actually a software project or not. It feels like the most natural way to work with AI currently.
So I started doing that with meetings.
I create a GitHub repository, a markdown file, and begin building an agenda with Claude Code. During the meeting, I keep a Claude Code window open and type anything I see as relevant, having first instructed Claude to update the document and sync with GitHub for every piece of new information.
When the meeting is done, I have a nicely-formatted markdown file in GitHub. From there, I copy and paste the pretty content into an email and send it out. Right away. Not later, which I’ll never get around to.
Here’s an example of what a finished agenda looks like (entirely fictitious—names, organizations, and details are made up):
Operations Meeting Notes
Date: January 6, 2026
Attendees
Sarah Chen (VP Ops), Marcus Webb (Facilities), James Liu (IT)
Follow-up: Badge Reader Installation
- Vendor confirmed Jan 15th; technician arriving 9 AM
- Action: James to verify network drops are active beforehand
New Business: Q1 Safety Training
- Compliance deadline March 31st (94% last year, targeting 100%)
- Split sessions across three weeks; add evening option for second shift
- Action: Sarah to send calendar invites by Jan 10th
This agenda was created with the assistance of Claude Opus 4.5 via Claude Code by Anthropic.
Â
So what’s the difference between this and typing into Word?
For starters, LLMs handle markdown files with ease. This isn’t coincidental—markdown has become one of the most efficient and LLM-friendly ways to structure content for AI. These models were trained on vast amounts of markdown from platforms like GitHub, and the format’s clean, hierarchical syntax mirrors how humans naturally organize information. The structure tells the model how concepts relate to one another: what’s a main idea, what’s a subpoint, what’s a list of items.
Second, Claude Code can take my stream-of-consciousness typing and build a well-organized agenda and notes out of it. I don’t have to do it in any particular sequence. The less the meeting follows the agenda in sequence, the more helpful this ability is.
For recurring meetings, I create a dedicated repository with a separate markdown file for each occurrence. The collection becomes a project unto itself. Follow-up items naturally flow from one week to the next, and Claude can reference previous agendas when building the next one.
I knew I was on to something when I sent out notes quickly following an IT meeting and got this response from our CEO:
“Bob— I like this format of meeting notes. Thank you.”
Sometimes the best software isn’t software at all.