🧭 THIS WEEK AT BuildProven
Howdy, on vacay next 2 wks, but here goes.
More and more advanced models are coming out and the complexity of using AI is growing. That would seem a bit contradictory but it is now taking quite a bit to learn the newest tools, techniques, models etc and to keep a system you have running well.
Example, if you’re using claude code, you might want to create a monthly or quarterly automatic check - ask Claude to simply create an automation - “create periodic automation, and check my token usage and optimize it appropriately. Additionally, check skills, commands and plugins usage and optimise”.
This week, we’re following up last week starting to use Claude Code.
as usual, would love it if you could share this newsletter, give feedback via the poll at the end, or best, just hit 'reply' and tell me what you need/want with AI. I will definitely reply and try to help best i can.🧰 Worth Your Click
Here are a few things I found recently:
How tech workers are doing in 2026 - I’d say not great given huge # of layoffs and AI disruption
bit detailed but interesting on what ppl are using Fable to do
OpenAI released GPT-Live - next gen voice models
Anthropic created a new /checkup command - looks useful - my setup
🗺️ FEATURED INSIGHT
Last week you installed Claude Code and made it change one file, then a whole folder. This week: about 200 meeting notes, one prompt, one table of who-owes-what, in about 4 minutes.
Here's the job. I had a folder of roughly 200 meeting-note .md files, standups, client calls, planning sessions. I wanted one thing out of them: every action item, with a date, an owner, and whether it was still open. One prompt did it, and I got back a single table I could read top to bottom.
No searching. No scrolling through Q2 to find the thing someone promised in April. One prompt, one table, done.
You don't need to learn anything new for this. It's the same tool from last week, pointed at a bigger, messier folder, doing the thing it's quietly best at: reading a pile of files you already own and handing you back exactly the slice you asked for.
Start where you left off
Open Terminal, move into the folder your notes live in, and start Claude Code, the same two lines from last week:
cd ~/Documents/meeting-notes
claudeIf your notes are spread across sub-folders (one per client, say), that's fine, Claude Code reads down into sub-folders too. Point it at the top folder that holds all of them.
One habit worth keeping from last week: get your bearings before you ask for anything. Type this first:
Don't change or write anything yet. Look at this folder and its sub-folders, tell me how many .md files you can see, and show me the format of two of them so I understand how the notes are structured.This is the seatbelt. Meeting notes are never as consistent as you remember some have "Action:", some have "TODO", some just have a name and a verb buried in a paragraph. You want to see that variety before you ask Claude Code to make sense of it.
The one prompt
Here's the whole job in a single prompt. Paste it, adjusting the folder description and the date range to match yours:
Read every .md file in this folder and its sub-folders.
Extract every action item, anything someone committed to do, was assigned, or agreed to follow up on.
For each action item, give me four things:
1. Date, the date of the meeting the note came from (use the date in the filename or at the top of the note; if there's no date, write "unknown").
2. Owner, the person responsible. Use the name as written in the note. If no owner is named, write "unassigned".
3. Action, what they committed to, in one short line.
4. Status, "done" only if the note explicitly says it was completed; otherwise "open".
Output ONE markdown table with columns: Date | Owner | Action | Status.
Sort by date, oldest first.
Do not invent action items that aren't in the notes, and do not merge two different commitments into one row. If you're unsure whether something is an action item, include it and put a "?" in the Status column so I can check it.Two things in that prompt are doing the real work, and they're the difference between a table you trust and one you have to redo:
"Do not invent action items, do not merge two different commitments."
Without this line, a model will happily tidy your data into something cleaner than reality.
The "?" in the Status column.
You're telling it that "I'm not sure" is an allowed answer. That one instruction turns the grey-area rows into a short review list.
What it hands back
The output is a single markdown table as example:
Date | Owner | Action | Status |
|---|---|---|---|
2026-04-08 | Priya | Send revised SOW to the Northwind account | open |
2026-04-08 | Marcus | Confirm the Q3 budget line with finance | done |
2026-04-15 | unassigned | Decide whether to renew the analytics vendor | open |
Make it a Monday habit
You just built something you can run again. Next quarter, next folder, same prompt, paste it, read the table, review the "?" rows. Two ways to keep it:
Save the prompt in a note titled "action-item extract" and paste it whenever you need the sweep.
Ask Claude Code to write it to a file at the end: add "When the table's right, save it to
action-items-Q2.mdin this folder" and you've got a dated record you can diff against next quarter to see what actually got closed.
Do it this week
You can do this on your own notes in the time it takes to make a coffee. Point Claude Code at a folder of them, a client's, a project's, or the last quarter's worth. Run the get-your-bearings prompt, then the extraction prompt. Read the "?" rows. You'll have a table of who-owes-what before the kettle's boiled.
Weekly build logs from a 25-year program manager who codes with AI.
— Brett
👉 Hit “Reply” and share your experience — I read every one!
Picture by xxx on Unsplash.