Every call you have writes itself up

Why It Works

Your AI is only as good as the data you give it access to. It is more than clever enough to write up a call, draft the follow-up and remind you what someone said three weeks ago, and it cannot do any of that for a conversation it was never in. A call is the most a client will ever tell you in one go, and almost all of it evaporates by the next meeting. Keep the transcript and the rest is a job an assistant could do, so give it to one.

Step One, Transcribe Every Call

I use the free version of Wispr Flow. Install it, turn on the notetaker in its settings, and it records every call from your machine and keeps the transcript, with the two sides of the conversation separated. In the Claude app, add Wispr Flow under connectors and your chats can read those transcripts directly. In Claude Code, paste the line below and it finds the transcripts on your machine itself. Either way the test is the same, ask it about your last call and see it answer from what was actually said.

Any notetaker whose transcripts land somewhere your AI can read them does the same job. Wispr Flow is the one I use because the free version keeps everything on my own machine.

Step Two, The Evening Review

Here from the video? This is the line. It asks which setup you are in first, then interviews you for where your contacts live and what a note you would actually read looks like, writes the standing instruction with the never-send rule inside it, and runs once against your most recent call so you see tonight what every evening looks like. Each call becomes a note, an updated row for the person, and a follow-up sitting in your drafts, written from what was said rather than from what you remember.

Read the first few notes properly and correct what it got wrong, in the note and in the instruction. It learns what you count as useful faster from your edits than from anything you could tell it in advance.

Step Three, The Daily Schedule

A write-up you have to remember to ask for is the same job you were not doing before. In Claude Code this line has it create the schedule itself, and from then on every call you had is written up by the time you sit down the next morning. Mine runs at nine each evening and leaves its note of what it wrote and flagged in the morning brief. In the Claude app it will tell you honestly that it cannot start itself, and give you the one line to send each evening instead.

One Honest Bit

The write-up is only as good as the transcript. Mumble a name and it will be spelt wrong, a figure said quickly can come out as a different figure, and once in a while the two speakers get swapped, which is why the prompt has it flag anything it could not hear rather than guess, and why you read the row before you trust it. The follow-up never sends itself. I press send on every one, and I want it that way.

You are recording people. Tell them at the start of the call, accept that some will say no, and treat the transcripts as theirs as much as yours, which is one reason I keep mine on my own machine. And the follow-up only sounds like you if the AI has read how you write, so the drafts will feel like a polite stranger until it has.

Where This Goes Next

The follow-up lands in your drafts folder alongside every other reply you owe, and your inbox already has the replies written is that routine. If the drafts came back sounding like nobody, the folder that writes like you is the fix. The note of what it wrote each evening lands, for me, in the meeting my AI books with me every morning.

This is the same idea as the onboarding document in it only knows what you tell it, applied to every conversation you have. If the idea of an agent is still fuzzy, an AI agent is a chat with a job is the plain English version and the start of the series.