Sandra spent twenty minutes arguing with AI about a floating shelf.
Not a metaphor. An actual shelf. The bracket was pulling away from the wall and she needed to know how to fix it.
She didn’t just describe the problem. She sent photos. Multiple photos. She circled the issue so there was no ambiguity about what she was looking at.
AI gave her five solutions. Each one sounded reasonable. Each one required screwing into something that wasn’t there.
She knew enough to question it. Knowing when to question AI matters just as much as knowing how to use it. Most people don’t.
So she changed her approach. Instead of asking AI to solve it, she came up with her own ideas and asked if they would work.
That’s when it got useful. AI validated her thinking, helped her spot a flaw, and confirmed which option made the most sense.
Same tool. Different use. Completely different outcome.
That’s the thing about AI — it sounds confident whether it’s right or wrong. It doesn’t hesitate. It doesn’t flag uncertainty. It just keeps going.
Which means knowing when to question it matters just as much as knowing how to use it.
When AI genuinely struggles
Anything physical or visual it can’t actually see. AI can read a description of your cabinet. It can’t see your cabinet. If the answer depends on what’s physically there — materials, structure, damage — it’s guessing. Even with photos, it can miss what’s obvious in person.
When you need someone to be accountable. AI won’t be there if something goes wrong. For anything with real consequences — medical decisions, legal agreements, structural repairs — use AI to prepare, then take it to someone responsible for the outcome.
When the situation is changing in real time. AI works from a snapshot. It doesn’t know what just changed. In live situations — crises, negotiations, developing issues — the context can shift faster than AI can keep up.
When you need someone who actually knows you. AI only knows what you tell it in that moment. It doesn’t know your history, your relationships, or the full picture. For decisions that depend on that — a family situation, a career move — a real person still matters.
When something doesn’t feel right. If the answer doesn’t match what you’re seeing, trust what you’re seeing. Sandra knew the screws would be going into air. The right move was to stop, not to keep asking AI for a better answer.
What AI is actually good for here
When AI can’t solve something directly, it can still help you think it through.
Describe your idea. Ask if it will work. Ask what could go wrong. Use it to test your thinking, not replace it.
In these situations, AI is often more useful as a second set of eyes than a decision-maker.
The rule that covers most of it
Use AI to think through a situation. Don’t use AI to replace your judgment about a situation.
The shelf example is low stakes. You notice the answer doesn’t make sense and you move on.
The higher the stakes, the more important it is to verify.
Try this: next time AI gives you an answer that doesn’t feel right, don’t ask it the same question again. Instead ask: “What am I missing here?” or “What would make this not work?” That’s when AI gets genuinely useful.
What to read next
→ How to Use AI Safely Without Overthinking It
→ The One Skill That Makes AI Actually Useful
→ Or visit the Decision Hub