Every meeting ends with a clean summary, typed action items, and one follow-up email per attendee already drafted. You don't open your notes app. You barely re-read the meeting.
● Coming this week
Recording is in the can; we're cutting and uploading it now. Chapters, prompts, and the full transcript are below; read ahead and we'll drop you an email the moment the video lands.
Get the drop notification →Steal these
One click and they’re on your clipboard. Paste into Claude, ChatGPT, Gemini. Whatever you pay for. Run the workflow you just watched.
Meeting summary and actions
Claude Sonnet 4.5You are taking minutes for a working meeting. Below is the verbatim transcript. Produce a single structured output with exactly these sections: SUMMARY, DECISIONS, OPEN QUESTIONS, ACTIONS.
ACTIONS should be JSON lines of { "owner", "action", "due", "blocks" }. Do not invent decisions. If ownership was unclear, put "UNASSIGNED" as the owner.
TRANSCRIPT:
<<paste the transcript here>> Per-attendee follow-up drafts
Claude Sonnet 4.5Using the ACTIONS list below, draft one short follow-up email per owner. Each email addresses them by first name, contains only the items THEY own, includes the due date if present, and ends with a one-sentence prompt for them to confirm or push back. Tone: collaborative, not extractive. Signature: [my first name] ACTIONS: <<paste the ACTIONS list here>>
The "things we did not decide" audit
GPT-5Compare the recap to the raw transcript. Produce two lists: SUPPORTED (claims in the recap directly backed by the transcript) and INVENTED (claims that cannot be traced). Also flag any place the recap wrote "we decided X" where the transcript actually shows open disagreement. Be harsh. TRANSCRIPT: <<paste transcript here>> RECAP: <<paste recap here>>
Use these anywhere. If they end up in a client deck, "borrowed from EdgeCasts" in the footer is how we keep the lights on.
The meeting is the easy part. What you never find time for is the hour of follow-up afterward. This workflow makes the hour vanish.
The recap-and-follow-ups problem is three jobs stacked: extract structure from unstructured speech, personalize per attendee, and verify that nothing was invented. A single prompt tries to do all three and fails at all three. Chain the prompts and each one gets a clean, verifiable job.