Scope, respected by nobody
I love using AI. Genuinely. When I’m the one in the prompt window: iterating, correcting, throwing away the bad bits, keeping what works: it’s one of the best tools I’ve picked up in years. The problem isn’t AI. The problem is when someone else hands me their AI output and treats it like finished work.
Scenario one: you give an assignment to your team
You brief someone on your team. Clear scope. Clear deliverable. “Write me 200 words on X.” “Proofread this paragraph.” “Fix this one bug.” “Give me three headline options, nothing else.”
They come back the next day with something that is technically on-topic and completely wrong for the job. Eight hundred words where you asked for two hundred. A rewrite where you asked for a proofread. A full dashboard redesign where you asked for a button colour change. Polished, confident, and now your problem.
You ask what happened. They say they used AI. As if that explains it. As if “I ran it through ChatGPT” is a substitute for reading the brief.
If a junior employee did this without AI, you’d have a word. “When I ask for a proofread, I want a proofread. Not a rewrite. Not a redesign. The thing I asked for. The size I asked for.” With AI in the middle, people somehow think the rules changed. They didn’t.
Scenario two: the client hands you slop
Same pattern, different direction. A client sends you their “finished” copy, their “approved” structure, their “ready to build” spec. You open it. It’s AI slop. Generic tone. Vague claims. Subheadings nobody asked for. Scope creep baked in from the first paragraph.
Now you’re not starting from a brief. You’re starting from someone else’s unedited chatbot output, and you’re expected to make it good: at your rates, on their timeline, without being rude about how it got there.
That’s not collaboration. That’s you doing the job twice: once to reverse-engineer what they actually wanted, once to do the work properly.
Why AI slop hits different from bad human work
Bad human work usually smells bad immediately. You know. You push back early.
AI slop is fluent. It’s formatted. It reads like someone tried. That makes it harder to reject without sounding picky: even when it’s wrong in all the same ways AI is always wrong when nobody’s steering: scope ignored, meaning drifted, specifics sanded into generics.
When you use AI yourself, you catch that drift. You reject the rewrite. You cut the extra sections. You keep the one useful line and bin the rest. When someone else hands you the output raw, all that filtering work lands on you, or worse, nobody filters it and the slop ships.
The rule I wish more people would follow
AI is a power tool. Fine in the hands of the person doing the job. Dangerous in the hands of the person trying to skip the job.
If you’re using AI on work you’re handing to someone else, you’re still responsible for the result. Read it. Edit it. Make sure it matches what was actually asked for. Don’t paste the first output and call it a day.
If you’re receiving work from a client or a colleague, you’re allowed to say: this isn’t what I asked for. The existence of AI doesn’t make scope optional.
And if you’re the one briefing AI directly: same rules as always. Tell it what it’s not allowed to do. “Do not rewrite. Do not add. Do not restructure. Return only changes as a list.” “Find typos. List them. Do nothing else.” I type that last sentence more than I’d like. But at least when I’m the one typing it, I know what’s coming back.
Useful? Yes. Magical when you’re driving? Often. Trustworthy enough to hand off unedited to someone who trusted you? Absolutely not.
For the same problem at industrial scale: across an entire e-commerce data pipeline. Read Don’t put AI in the middle of your pipeline. Spoiler: it does not go well.
English 
Dutch