Not exactly a tool, not exactly an assistant. LLMs give you access to categories of labour, Craig Mod wrote recently, that individuals like him and me would otherwise be priced out of.
Specialists you couldn’t otherwise afford to work with, doing tireless work that you might not ask a human to do without feeling bad about it.
Since reading Mod’s account, I’ve set up a small staff of my own:
Archivist
Copy editor
Gallery curator
Coaching group assistant
Rights and backlist director
Each of these is a conversation window I keep open, much as you might keep a WhatsApp thread with a colleague, except the colleague is Claude wearing a different hat each time. I find it helpful to imagine them as distinct individuals, in chats with their name attached.
Each specialist chat is embedded within a folder on my Mac – Claude can read whatever’s inside.
Before any session, that “staff member” reads its own state document – a plain text file I keep updated, saying what’s been decided, what’s still open and what happens next.
If I asked one assistant to do all of this, it would either forget half of it or drown in the rest. Splitting the work means each specialist needs only to hold a small world in its head.
The underlying text files remain small, and because I work this way I use relatively few AI tokens. Fewer still if I keep “compacting” conversations to remove excess data in the chat’s cache. (I can do that because what matters remains in my own text files, not in some distant Claude server.)
What ties it together is a “Chief of Staff” chat, to see the whole thing at once. This runs all the state documents, cross-checks them against each other and produces a single briefing.
Today, that briefing unearthed a deadline I’ve been letting shape my week without keeping it in sight: a Micro-Memoir early bird price that nobody has even heard about yet. I’ll move it back a week. Half my anxiety went with it.
***
This is a note-to-self, posed here for anyone to see. See also: my whole site in text files.