If you work with an AI long enough, the moment always comes. You get a setting wrong. You break a pile of data. You do something that feels impossible to undo.
What happens next depends entirely on whether your mind goes blank and you freeze, or whether you can calmly put things back together.
The other day I managed a fairly spectacular one: I accidentally created 94,080 wrong settings in a single move. Today I want to share what that incident taught me about how to lean on an AI when you are panicking, so you get back on your feet by the shortest possible route. This is one of those things that really does help when the moment arrives.
- What happened: one innocent checkbox on a bulk import
- Tip 1: Do not dress it up. Show the panic, and the screen, exactly as they are
- Tip 2: Ask for the options and a recommendation, with reasons
- Tip 3: Before an irreversible action, ask for a beat of checking
- The takeaway: how you ask decides how fast it ends
What happened: one innocent checkbox on a bulk import
I had installed a plugin. Somewhere in the middle of the setup there was an item asking, “Do you want to import your existing settings?” — and the box happened to be checked. I did not think about it much. I just carried on.
Later I opened the list of settings. The screen was filled with a list that went on and on and on. The count: 94,080.
And every single one of them was wrong. They all pointed in the opposite direction from what they were supposed to. If they had actually been live, the whole site would have been broken.
That is where the biggest lesson of the day starts.
Tip 1: Do not dress it up. Show the panic, and the screen, exactly as they are
When you make a mistake, the natural instinct is to paint over the whole screen with the worst thing your imagination can produce. “What do I do, maybe everything is gone, maybe this is the end.”
The important thing here is the opposite: do not run on guesswork. Show the screen as it is right now to your AI and start with a plain question of fact — “is this a bad state?”
I showed my AI partner Kuro (Claude) the situation, more or less in tears. This is what came back:
First, relax. Those 94,080 settings are currently in a disabled state. So right at this moment, no harm is reaching your site. The number of times any of them actually fired is zero across the board. Nobody has stepped on one, which is proof that the disabling is working. You are fine for now.
My breathing steadied.
Panic, most of the time, is nothing more than imagining the situation to be bigger than it actually is. So the first move is to get the AI to confirm the facts only. Is real damage happening right at this second, yes or no? Once that is clear, your head starts working again.
Tip 2: Ask for the options and a recommendation, with reasons
Once you are calm, it is time to deal with it. And here, rather than just asking “what should I do?”, it works far better to say: “List the options I have, give me the upsides and downsides of each, and then tell me which one you recommend and why.” That way you can act while actually understanding the decision.
My partner came back with this:
There are two options. Leave them disabled and ignore them, or delete all of them. I recommend deleting. Not one of these wrong settings has any value, and if you leave them there you are living with a landmine — the instant someone accidentally enables them in the future, all of them go off at once. Delete them and that risk becomes zero.
“There is no need to live with a landmine in the house.” The reasoning convinced me, so I chose deletion without hesitating.
Do not just swallow the AI’s conclusion. Ask for the reasons, satisfy yourself, and then move. That is the trick to making decisions you will not regret later.
Tip 3: Before an irreversible action, ask for a beat of checking
Deleting is hard to undo. My partner was careful here too. “Before we delete, let us check whether any of the correct settings you made yourself are mixed in with these,” it said, and it inspected the state before going ahead.
Luckily, the situation allowed me to delete only the 94,080 wrong ones. Bulk delete. In an instant, the landmine was gone.
After that, I added one correct setting by hand, quietly checked that it behaved properly in a separate window, and it worked.
Before an action you cannot take back — “delete”, “send”, “confirm” — ask your AI to pause for one beat and check before proceeding. That is how you avoid a second round of damage.
The takeaway: how you ask decides how fast it ends
The biggest lesson of the day is this: the more convenient a feature sounds — “import”, “bulk load” — the more you should refuse to tick the box without looking at what is inside. One move made to save myself effort turned into 94,080 landmines. For settings that matter, adding them one at a time by hand, however tedious, ends up being the safest route.
And if you do mess up: first, do not panic on guesswork — show the current screen and confirm the facts. Second, ask for the options, the reasons, and a recommendation. Third, ask for a check before any action you cannot reverse. Remember that order and most trouble gets cleared up without a panic.
Maybe the real strength of working with an AI is exactly this: in the moment you have screwed up, it does not blame you, and it walks you back by the shortest route.
I have also written about what to do when the AI itself stops mid-task, over in what to do when your AI freezes. And if you want to see how I ask an AI to dig through data and find the cause of a problem, that one is here.
