My first automation wasn't a design. It was copy, paste, and see what happened. I built it in n8n, a program where you wire tasks together step by step without having to program everything yourself, with Gemini beside me as a co-pilot. I'd paste an error message, get a suggestion, paste it back, and try again. That's how, by trial and error, I got the first automated reports working.
That fumbling wasn't a detour, it was the lesson. Every error I pasted back taught me something about how the parts work together. After a week I understood more about how data flows from A to B than a course had ever taught me, simply because I had to make it work.
At some point the flow worked on my laptop. But a laptop closes. A report that has to be ready every morning can't wait until I turn on my computer. So it moved to its own server, a VPS, that keeps running day and night. From that moment the system worked even while I slept.
That brought the less exciting work that makes the difference. The server had to be secured, the automations had to keep running, even after a restart, and I wanted to be able to see in the morning what had happened overnight. None of it is exciting. All of it is the difference between an experiment and something you can build on.
The last big step was in the storage. At first I wrote everything to Google Sheets, because I knew it and it worked right away. Until it didn't: at thousands of rows it got slow and clumsy. I moved the data to Supabase, a real database. Since that switch the system can handle far more, and I can look up much more precisely what happened when.
And it didn't stop at those morning reports. When I started building my AI assistant Kai, I ran into a limit. The standard way to let the AI talk to Notion, the program that holds all my tasks and knowledge, was slow: a simple overview of ten tasks cost eleven separate requests. So I built my own flows around it that handle such a question in one go. More followed, each for a chore I didn't want to do by hand every time. The machine grew along with what I needed.
The most visible example lands in my inbox every morning. A briefing on the war in Ukraine and on the tech world, put together while I slept, from sources the server fetches and summarizes on its own overnight. No longer a loose trick, but the visible tip of the machine underneath.
Curious what rolls out every morning? The daily briefing that runs on this machine.
