Journal · The AI Architect · 2026-08-11
Copy real files before your AI touches them. A backup is a seatbelt — you put it on before you drive.
Copy real files before your AI touches them. A backup is a seatbelt — you put it on before you drive.
The tension isn’t whether AI tools are trustworthy. It’s that the moment you feel confident in a tool is exactly the moment you stop protecting yourself from it. You’ve run the assistant on a dozen small tasks, it’s been accurate and fast, and by the thirteenth task you hand it your real project folder without a second thought. That’s not carelessness. That’s how trust normally works with people and processes. The problem is that an AI editing your files doesn’t accumulate judgment the way a colleague does — it can be excellent nine times and confidently wrong the tenth, with no change in tone to warn you.
Backups are about confidence, not distrust
There’s a quiet misconception that making a backup means you don’t trust the tool. It’s the opposite. A backup is what lets you say yes to bigger, more useful requests. If you know your original files are sitting untouched somewhere safe, you can let the AI restructure a whole document, rewrite a batch of code, or reorganize a folder without flinching. Without that safety net, you end up being overly cautious in small, annoying ways — approving only tiny edits, re-reading every line before it happens, slowing the whole collaboration down because the cost of a mistake feels unbounded.
Try this: before your next AI-assisted task that touches more than one file, copy the whole folder to a new location with today’s date in the name. Not a vague “backup” folder you’ll forget about — something specific, like project_2024-05-14_before_ai_edit. That small act of naming it by date makes it easy to find later and easy to trust you did it.
Make the backup step automatic, not a decision
The reason people skip backups isn’t laziness — it’s that “should I back this up?” is a decision, and decisions require attention you may not have in the moment. You’re focused on the task, not on risk management. The fix is to remove the decision entirely. Build a habit or a script that runs before any AI session that touches real files, so backing up isn’t something you weigh each time — it’s just what happens first, like putting on a seatbelt before you even think about the drive ahead.
Try this: if you’re comfortable with basic scripting, write a two-line script that copies your working folder to a dated archive and then opens your AI tool. Run that script instead of opening the tool directly. If scripting isn’t your thing, keep a checklist taped near your screen with one item: “Copy files first.” The format matters less than making the step come before the impulse to skip it.
Know what a backup does and doesn’t protect
A backup protects you from a bad edit, a deleted paragraph, a corrupted file, a rewrite that went sideways. It does not protect you from a bad decision you make deliberately — if you tell the AI to delete something and it does, that’s not a backup failure, that’s a choice you’ll want to reconsider before making it. Understanding this distinction matters because it keeps you from either over-relying on backups (assuming they’ll save you from every kind of mistake) or under-using them (assuming they’re only for catastrophic failures).
It also helps to think about how many backups you keep, not just whether you have one. A single backup protects you from one bad session. If you’re doing several rounds of AI-assisted changes over a week, keeping a dated copy before each significant session means you can step back not just to “before,” but to whichever “before” actually had the version you wanted. Storage is cheap. The folder you wish you’d kept is not something you can recreate after the fact.
Try this: once a backup is more than a few weeks old and you’re confident you no longer need it, delete it. This keeps the habit sustainable — you’re not drowning in fifty dated folders, and you’re more likely to keep doing it if it doesn’t feel like clutter.
None of this requires special tools or technical sophistication. It requires treating the backup as the first step of the task, not an optional extra you’ll get to if there’s time. In the book, I go into more detail on setting up backup routines that fit different kinds of work — writing, code, research — and how to recover cleanly when something does go wrong. But the habit itself is simple enough to start today, with whatever file you’re about to hand over next.
Go deeper. The full method is in The AI Architect. New here? Start with the free companion pack, or explore the series.