L
🟢 Working with AI · Part 1 · Your first hours

What goes wrong on day one

Most trouble with AI is not the AI. It is a handful of ordinary-looking moments that almost everyone walks into, in roughly the same order, in their first few hours. Each one below names the moment, tells you what to do in it, and says what you get in return.

Your files · protect them firstMoney · set the lineTime · stop a runaway job45 habits

How do I stop AI changing or deleting my files?

What actually happens

You hand over a spreadsheet of client contacts and ask it to tidy the phone-number column. Back comes the whole sheet, rewritten, with two columns you never mentioned now in a different order. The version you had is gone. Nothing warned you, and the save reported success.

Why it happens

Opening a file to write into it empties the file first, before a single character goes in. If the writing then stops halfway, through a crash or a full disk, you are left holding an empty file instead of your old one.

From the outside an empty file and a saved file look identical. You find out when you open it, which is usually much later. A delete run from a command line has no recycle bin behind it either.

What to do
  • Work on a copy, never your only file. → Duplicate it first and hand over the duplicate. Ten seconds, and it makes every ambitious change safe to try.
  • Take the copy now, not this morning. → A copy from six hours ago is six hours of work waiting to be lost. Freshness matters as much as having one.
  • Copy both halves of a pair. → A document and the sheet it draws from, a page and the file behind it. Recovering one while the other is gone is its own bad afternoon.
  • Ask for the change in one place first. → One section, one column, one page. Look at it. Then let the same change run everywhere.
  • Name what to keep, not only what to remove. → Everything except this one is read more widely than you meant. Say the files by name.
  • Check there is room on the disk before a big write. → A disk that fills mid-write leaves a file of nothing and reports no error at all.
  • Treat a test run as a real one. → Plenty of jobs write to the same place whether you asked for three items or three thousand. Read where the output lands.
  • Never empty a growing record to run it again. → A second run adds what is missing and skips what is there. Emptying it first throws away every earlier run.

See what each ask leaves behind

Same job, two ways of asking. Pick one and see what is left when it goes wrong halfway through.
Your original
—
The work
—

Pick one above.

What this gives you

Not caution. Room. Once the original is safe you can ask for the ambitious version, the one you would never risk on your only copy. Most people ask small because they are frightened of the file, and small is what they get back.