The difference between a one-off AI session and a repeatable workflow is one file.
I have a LinkedIn analytics dashboard built entirely with Claude. Drop new xlsx exports in, run Claude, get an updated dashboard. Takes two minutes.
What makes it repeatable is not the code. It is a CLAUDE.md file that sits in the project folder and acts as the project brain.
It describes the folder structure, the data schema, what each file contains, and what to do when new data arrives. Every session picks up exactly where the last one left off. No re-explaining. No drift.
Without it, you get a useful one-off result. With it, you get a workflow.
This pattern works for any recurring AI task. The file is the continuity layer. It stops you from starting from scratch every time.
I shared the exact CLAUDE md file I use for the LinkedIn analytics project. It is linked in the first comment if you want to take it and adapt it for your own setup.