I've spent this year building products with a lot of AI assistance — five of them, end to end. So when people ask whether AI-assisted work "counts," I'm not answering in the abstract. I've had to answer it for myself, every day, in front of a real codebase. Here is where I landed.
The pattern is old, and it always goes the same way
Every time a tool democratises a scarce skill, the people whose identity was built on that scarcity call it cheating.
Photography was "not real art." Word processors meant you weren't a "real writer." Calculators meant you didn't know "real math." Google meant you weren't "really smart." Stack Overflow meant you weren't "really a programmer."
Every one of those arguments was made. Every one of them lost. And in each case, the people who adapted earliest captured the most value in the shift. The argument we're having about AI right now is the same argument — compressed from fifteen years into eighteen months.
But "it's just the same old gatekeeping" is too easy an answer. Because this time there's a version of the cheating argument that is actually correct, and it's worth separating from the version that isn't.
The weak version is gatekeeping
"You used AI, so it doesn't count" is not an ethical position. It's incumbency dressed as ethics — people protecting the scarcity their value was built on.
A carpenter who uses a power saw isn't cheating. They're producing better work, faster. The cabinet is what gets judged, not the calluses. Nobody asks a finished piece how many of its cuts were made by hand.
If the work is good, "you used a tool" isn't a critique. It's an admission that the critic is behind on the tool.
The serious version is real
Here's the version that's true: if you use AI to produce output you cannot evaluate, you are shipping unreliable work that will fail in non-obvious ways — and you may not even know it's wrong.
A student who uses AI to write an essay they can't critique is submitting something they don't understand. A developer who uses AI to write code they can't debug is shipping a system they can't maintain. That is a genuine problem.
But notice what the problem actually is. It isn't the tool. It's the absence of judgment over the output. The tool is neutral. The real question is whether the person holding it can tell good from bad — and that single distinction tells you exactly which skill still matters.
You can only catch slop if you know what "not slop" looks like
"AI slop" is real — output that is fluent, confident, and wrong, or technically correct but tone-deaf to the context. I see it constantly. And here is the thing about catching it: you can only recognise slop if you already know what good looks like in that domain.
Which means knowledge and taste didn't become less valuable. They became the whole job. What changed is where you spend them. You're no longer the writer — you're the editor. No longer the one typing the code — the one deciding the architecture and reviewing the result. The craft moved up a level. It didn't disappear.
The people who think AI made expertise obsolete have it backwards. AI made expertise the only thing separating usable output from confident garbage.
"Vibe coding" and the level you actually need to understand
The critique of "vibe coding" — generating code you don't understand and shipping it anyway — has merit at the bottom end. People who build systems they can't reason about will build systems that fail mysteriously.
But the answer isn't "don't use AI." It's "understand what you're building at the level that matters."
A pilot doesn't understand aerodynamics at the physics level. They understand how to fly, and how to recognise when something is wrong. The required depth of understanding shifts; it doesn't vanish. The dangerous move isn't using the tool — it's not knowing which level you're allowed to stop at, and stopping too early.
The question that's coming
In five years, no serious interview will ask "did you use AI." That question will sound the way "did you use a calculator" sounds now.
The question will be: can you defend the output? Can you stand behind what was produced, explain why it's right, and catch it when it's wrong? Work samples, interviews, and professional judgment are already shifting from "did you make this unassisted" to "can you evaluate this."
That's the bar I hold myself to. I let AI do an enormous amount of the work. I do not let it make a decision I can't defend. The leverage is real and I'd be foolish not to use it — but the moment I ship something I can't explain, the tool has stopped helping me and started exposing me.
Use the tool. All of it. Just make sure you can always answer the only question that's going to matter: can you defend the output?