Experiment 04Summaries that stay accurate
Meeting notes to actions
Turns rough notes into decisions, actions with owners and dates, and open questions, each linked to its line.
Simulated preview — local rules, no live AI. This experiment runs in your browser with illustrative data. Nothing you type is sent anywhere. Please use made-up text, not real personal or confidential details.
Try a sample
One point per line works best. Owners are read from “Name to…”, “Name will…” or @name.
Summary
Validated · simulatedAttendees: Anna, Ben, Carla
Decisions (2)
Agreed to launch with online booking, payments later
Line 4
Keep the current domain
Line 5
Actions (3)
Send the sitemap draft
Owner: BenDue: FridayLine 6
Collect photos from the clinic
Owner: CarlaNo dateLine 7
Share the brand guidelines
Owner: AnnaDue: next weekLine 8
Open questions (3)
Need to check whether the booking system has an API
Line 9
Who signs off the copy?
Line 10
Hosting: tbc
Line 11
2 lines left out as context or small talk.
What this shows
Summarising without inventing: owners and dates come only from the notes, and missing ones are flagged.
The rules here match patterns; they don’t understand language the way a model does. They exist to show the shape of the workflow honestly: fixed output, validation, and a person deciding.
In a production version
- A model works from transcripts or notes and returns the same structure, each item citing its source.
- Actions are proposed into the team's task tool for someone to accept, not created silently.
- Attendee lists come from the calendar, so owners can be checked against who was there.