01The problem
A decision made in a conversation with the AI, a number heard on a call, something promised to a client: if they stay where they were born, a week later they no longer exist. To-do lists grow in three different places and none of them is the real one. Time goes into rebuilding what was already known.
02How it is used
- Work goes on as usual. At the end of each session the assistant writes what changed into the right note: client, project or procedure.
- At 8:30 the morning board arrives: a single page with the day's to-dos, taken from the notes and grouped by client and project.
- Whatever gets done is ticked off. The item goes back into the note it came from, marked as done, with the date.
- When something is needed, you ask. The notes are linked to each other: from a client you reach the procedure, from the procedure the source it came from.
03How it works
- Notesone card per client and per project, with everything inside but not spelled out: the details stay folded, with their date and source
- Linksevery note points to the others; the graph that comes out is the map of the work, with one colour per type of note
- Gardenerthe assistant distils the sources, proposes the links and keeps things in order; the decisions stay with me
- Morning boardevery morning the open to-dos become the day's list; every ticked item updates its note
- Written ruleswhat the assistant may rewrite and what it may not is written down, not remembered: how a session is closed, how a block is dated, what is never touched
04Results
The first attempt to split long notes into sub-pages was dropped: it moved the mess into the index. The current form, one card per client with the details folded, has held since September. The most useful rule came from a mistake: a tool had gone live and the note still said "to be launched". Since then every change is written in the same session, never the day after.