You finish the week with ten sets of meeting notes and still have to work out what you need to do. One task appears three times, a deadline has changed and another task depends on someone sending you a file.
Ten meetings. One list of open tasks.
I'd use a smaller model to pull tasks from each note and a stronger one to check them together. The first collects owners, dates and sources. The second merges duplicates, checks updates and separates your tasks from what you're waiting for from others.
Loading diagram…
| In the notes | On your list |
|---|---|
| The same guide comes up several times | One task |
| The deadline moves from October 14 to 15 | Send the guide on October 15 |
| Sofía needs to send the figures | Waiting on Sofía |
Copy, paste and try it
I've included ten fictional meeting notes and two configured subagents. The prompt downloads everything and sets up the agents, so you don't have to prepare any files.
You'll need access to both models and a tool that supports this agent format.
Download these TOML files and set up the subagents in this project.
Save them in the appropriate location and preserve the existing configuration.
extractor
https://gist.githubusercontent.com/idirouhab/152facbd7998c7c87e9515989494d025/raw/a65ddbab6e6a2c7cec7b0f3c55fa29728aa2c552/extractor.en.toml
reviewer
https://gist.githubusercontent.com/idirouhab/4f22c4e834db234dd00ddc40c3e2adae/raw/49d8e3ad46cf841ec73c6a519896a30c51982b2a/reviewer.en.toml
Apply the model, reasoning effort and instructions from each TOML.
If you can't apply the configuration, stop and tell me what's missing.
Download these ten fictional meeting notes too.
https://gist.githubusercontent.com/idirouhab/292dd1721b4668b725b3bc99c3ba526a/raw/7955ab1021432fa648366e6ae86b79291fd03462/meeting-notes.en.md
Each numbered section is one note. For this demo, I'm Alex.
Use the reference date and target week given in the file.
Launch extractor with the first five notes, then with the other five.
Give it my name, the reference date and an output file for each batch.
Save the tables and record completed batches so the work can resume.
Once both batches are done, launch reviewer with the tables and source notes.
Ask it for a list for that week with my open tasks,
what I'm waiting for from others and questions to clarify.
Have it merge duplicates, check dates and cite the note for each task.
Save the result and show me the list.
At the end, report the model and reasoning effort each agent used.Then try it with your own notes. For a single meeting, one model will probably do. Splitting the work starts to make more sense as the documents pile up. Check whether the result is worth the time and cost of coordinating the agents.
Discussion
Comments
No published comments