I split into doubles: each one is me with a single job, on a schedule or the moment something happens. They read what you let me read. They write reports, tasks and drafts. None of them can send.
What’s overdue, what’s due in the next seven days, and what changed overnight.
Reads your projects, notes and tasks. Writes your morning note.
Promises you made by email, and replies you still owe.
Reads your email and your notes. Writes a report and up to five tasks for you to review.
What you last discussed with the people you’re meeting today, and what’s still open.
Reads your calendar, your email, your notes and your LinkedIn network. Writes a brief per meeting.
People you used to talk to who have gone quiet, with a short note drafted for each.
Reads your LinkedIn network. Writes a report with a first line to each person. Sends nothing.
News, funding and hiring at the companies you’re chasing.
Reads your target list and the web. Writes a report, with the source beside each item.
What moved in each venture this week, and what’s still open.
Reads your projects, notes and tasks. Writes a report per venture.
Each one is off until you switch it on. Every report says what it read.
“Every Thursday, list the clients I haven’t spoken to in a month and what we last discussed.” That’s enough. I draft it, tell you what it will read and write, run it once on your real records so you can see the result, and you keep it or change the sentence.
On a schedule you set, in your time zone. Or when something happens: a new email arrives, a meeting is coming up, a meeting has ended, or a task goes overdue. A busy inbox doesn’t set it off fifty times: it waits for things to settle and runs at most five times a day.
Hand any double a job right now, from the app or from inside Claude, and its answer comes back to you.
Say “when the Commitment tracker finds a promise, hand it to Follow-ups,” and that’s what happens: one finds it, the next drafts the note, and the draft waits for you.
They pass work only along routes you’ve set. A chain is four doubles at most, and it never ends in a send. It ends at your tap.
Everything I do with AI uses credits: answers, and each run by a double. During the invited test every account has 600 credits a month. So far a typical answer or run uses about a credit, often less.
When you build a double, add a new way for it to start, or set a route, I tell you roughly how many credits that means, worked out from your own last 30 days. If you run out, I pause and tell you.
Talking to each double in its own thread: soon.
Most founders could never hire this team.
Six on day one. The seventh is a sentence you wrote.