Voice to Summary to Action

🧑‍💻 The Manual Way

You record a voice memo walking to your car after a great call. You captured everything — the budget, the timeline, the name of the person who actually makes the decision. It’s all right there on your phone. And then life happens. You drive home. You make dinner. The memo sits there for a day, then two, then a week. When you finally listen back, you can’t remember which call it was from. The context is gone. The action items you nailed in the moment — the ones that would have moved the deal forward — are buried under seven newer memos you also haven’t listened to. Voice memos become a graveyard of good intentions.

🤖 The Automated Way

Record your thoughts. That’s it. That’s your entire job. Walk to your car, talk into your phone for two minutes, and put it away. By the time you start the engine, the transcript is done. By the time you pull out of the parking lot, your CRM has the call notes, the follow-up email is drafted and waiting for your approval, and the action items are sitting in Notion with deadlines attached. You just review and hit send.

How It Works

1. Voice memo submitted via phone shortcut, Slack, or file upload You record a voice memo on your phone and it hits the workflow automatically. Could be an iOS Shortcut that sends the audio file, a Slack message with a recording attached, or a simple file drop. No extra apps to learn. You use what you already use.
2. OpenAI Whisper converts speech to text with speaker identification Whisper handles the transcription — accents, background noise, fast talkers, industry jargon. It identifies who’s speaking if there are multiple voices on a call recording. You get a clean, timestamped transcript in seconds. Cost: fractions of a penny per minute of audio.
3. Claude Haiku parses the transcript into structured output The raw transcript goes to Haiku with a prompt tuned to your workflow. It pulls out action items, decisions made, people mentioned, deadlines committed to, and any open questions. It knows the difference between “we should probably do that someday” and “I need this by Friday.” Haiku is used because this is a fast, structured extraction task — no need for a bigger model.
4. Actions fire in parallel: CRM, email, task manager The structured output triggers three things at once. Your CRM gets the call notes attached to the right contact record. A follow-up email gets drafted with the key points and next steps already written. Notion (or Asana, or whatever you use) gets tasks created with owners and due dates. All from one voice memo.
5. Summary sent to you with links to everything created You get a single notification — Slack, email, or SMS — with a summary of what was captured and direct links to the CRM record, the draft email, and the tasks. One glance tells you everything is handled. You approve the email, and you’re done.

What It Replaces