Josh's wife bought some coloring pages for the kids. AI-generated, cheap, seasonal. A Christmas tree covered in presents.
They looked fine at a glance. Then you noticed the ribbon on one present made no sense. It just stopped, mid-loop, going nowhere. No human would draw that. A human would notice it looked wrong and fix it, or start over. Whoever generated that page didn't look closely enough to catch it, and neither did whoever approved it for sale.
That's the failure that matters here, and it isn't the generating.
Two different problems wearing one name
Someone being too lazy to draw their own coloring pages isn't really a problem. Nobody needs to personally illustrate a coloring book to feel like a complete person.
The actual failure is that nobody checked. The tool did the work in seconds, and the human still didn't spend the ten seconds it would have taken to look at it before it went out the door. That gap, the one between "AI produced this" and "someone verified it," is where all the real damage lives, and it has nothing to do with whether using AI was appropriate in the first place.
Which raises the actual question people are dancing around when they ask whether AI is making everyone lazy: lazy compared to what? Compared to a world where the friction of doing something by hand accidentally forced a check into the process. Take that friction away and you find out how many of your standards were never really standards at all, just the tax you paid for not having a shortcut yet.
The test that actually works
Not "is this task creative" or "is this task important." Those categories are too fuzzy to apply under pressure, on a Tuesday, when you're deciding in real time.
The better question: is my reputation on the line if this is wrong?
If a task goes out under your name and being sloppy would cost you something real, that's the one you check carefully, AI-assisted or not. If it's genuinely low stakes, low visibility, and nobody's worse off if it's a little rough, then whether you generated it or wrote it by hand was never the interesting question.
There's a second test worth running alongside it: am I trying to learn this, or just trying to get it done? If you're trying to build a skill, handing the whole thing to AI trades the point of the exercise away, the same way you can't learn to work on your car by taking it to a mechanic every time. If you already know how and just need the output, that trade doesn't cost you anything.
Where it actually gets uncomfortable
Here's a version of the question that makes people pause: would you be okay if the person on the other end knew you used AI for this?
Using AI to help draft a client report you're going to edit and stand behind is one thing. Using AI to write a personal note to your own spouse is a different thing entirely, and most people's gut reaction to hearing that example out loud is immediate and negative, which is usually a sign the gut reaction is right. If the value of something depends on it being genuinely yours, and the other person would feel differently about it knowing a tool wrote it, that's the line, and it moves depending on what you're actually making.
What this means day to day
The problem was never the tool. It's that "I'm being more productive" and "I stopped checking my own work" now produce the exact same feeling, and only one of them is true.
So before automating or generating your way through something, ask the two real questions. Is my name on this in a way that matters if it's wrong. Am I trying to actually get better at this, or just trying to be done with it.
Everything else, the actual mechanics of which tool and which prompt, is a lot less important than people make it sound.