Hey,

It has been quiet here for a while. This newsletter used to be about Google Sheets and finance, the formulas and dashboards and small tricks that save hours every month, and then I got heads-down building things and the posts stopped.

What I was building changed how I think about this work, so I wanted to share it, because it feels like the kind of shift that is worth writing about.

For the past while I have been wiring AI into my finance work, the actual day-to-day work with the usual deadlines attached to it, and somewhere in the middle of that I kept running into the same realization, which is that writing the code is not the hard part anymore, because I can describe what I need and it gets built, and the hard part has moved somewhere else entirely, and that somewhere is coordination.

When everyone on the team can build, and everyone's agents can build too, the question stops being whether we can make something and becomes how we keep from building the same thing twice, because two people will ask their agents for the same report and three agents will update the same sheet and someone's work will get overwritten because nobody knew it was in progress, and it is basically the shared-drive mess except that now it happens at machine speed.

What actually works comes down to two things.

First, keep the work in one real place, which is GitHub if you are technical and a Google Drive folder synced to your machine if you are not, because the point is the same either way, which is one source of truth for the actual files instead of five copies scattered across chat threads.

Second, keep a shared record of context, which means every project with its status and its history and its decisions lives in one database that every person and every agent can read, so anyone can pick up exactly where the team left off and nobody has to ask what was decided about the Q3 numbers because it is written down.

Here is my actual setup, which is a Google Drive folder synced locally that holds the deliverables, with an Apps Script dashboard and database sitting on top of it that keeps context and status in sync, and the rule that makes the whole thing work is that an agent has to claim a project in the database before starting it, like writing your name on the whiteboard, so with no claim there is no work and two agents never end up doing the same task twice.

It is simple and it works, which is usually how you know something is right.

I am not a software engineer, I am a finance guy who builds things, and what this whole stretch taught me is that the people who win with AI will not be the best coders, they will be the best coordinators.

TL;dr: building got cheap and coordinating got expensive, so keep one source of truth for the files and one database for context, with a claiming rule so that work never doubles up.

More to come, I am writing regularly again.

- Francois

P.S. One question I keep coming back to is this: if you could hand one repetitive part of your week to an AI agent tomorrow, what would it be? Hit reply, since I read every response, and your answers will shape what I write next.