Version Control for AI, Without the Jargon

"Version control" sounds like programmer territory. It's not. It's just the discipline of knowing which version of something did what, and when.

Think of textbooks. The 3rd edition and the 4th edition are different books — different chapters, corrected errors, new material. If a student asks about page 47, your first question is: which edition? AI systems work the same way. The model you ran in March and the one you run today are different editions, even if nobody announced it.

So keep a version log. Every time something meaningful changes — a model update, new training data, a rewritten prompt, a settings tweak — write down what changed, when, and who approved it. One line per change. That's version control.

Why does this matter? Because when something goes wrong, the first question is always "what changed?" If you can answer it in thirty seconds, you look competent and the fix is fast. If you can't, you're guessing in the dark while the problem continues.

Your future self, debugging a weird output at 4pm on a Friday, will be enormously grateful to the present you who wrote it down. Editions matter. Log them.

I explain frameworks; I don't give legal counsel.