Last time I promised you a story about two versions of the same file. Here it is, and it's a short one, because the fix turned out to be short too.
Keep in mind that this happened while everything was already "organized." I had my working method in place, my files in folders, my agents reading from them. And still, a small mess crept in.
The problem: two drafts, side by side#
When I wrote my first blog post, the draft went through a couple of rounds. The first version, then a better second version. Both lived in the same working folder, right next to each other, with names that were almost the same.
That's normal. We all do it. Draft one, draft two, draft final, draft final for real this time...
Here's the part that matters. My agents don't scroll around a folder the way you and I do. They check a short list first, the one that says "this is the current version of that thing." And that list was still pointing at the older draft.
Nothing broke yet. But the next agent to pick up that post would have started from the wrong file, confidently, without anyone noticing. Same as the drift story from last time: no error, no warning, just quietly wrong.
The fix: one folder per piece of work#
The fix was embarrassingly simple, again.
Every post now gets its own folder. Everything about that post lives inside it: the notes, the outline, every draft, the free download, the checks. One post, one place.
And when a draft gets replaced, it doesn't get deleted. It gets a line at the top that says it's been replaced, and by which file. The old version stays as history, but nobody, human or AI, can mistake it for the current one.
We moved five files that day, fixed the pointers, and left a small table showing where things used to live, so older notes still make sense.
Why this matters more with AI#
A flat folder works fine at ten files. At two posts a week, with notes, drafts and downloads for each, it turns into a few hundred files a year. You can still find things, because you remember. Your agents can't. They find files by name and by pattern, so the more files that look alike, the better the odds they grab the wrong one.
Structure isn't neatness for its own sake. For an AI, it's the difference between the right answer and a confident wrong one.
I'm still learning here. This one is small, but small messes are the ones that compound quietly.
If you want to see how I keep the bigger picture from forgetting, start with Why Your AI Forgets You, and the Folder That Fixes It.