You’ve probably done it. Sat in front of Suno or Udio, typed a prompt, and felt your jaw drop as a song that sounds like it was written by your favorite artist pours out. You might have even thought: This is it. AI has conquered art. I am obsolete.
Then you asked it to write a Python function — something simple, like merging two sorted lists. And it gave you code that looked right, but when you ran it, it threw an error. You debugged it, gave it the error message, and it spat out a new bug. You gave up and wrote it yourself in five minutes.
The AI that made you weep with a melody can’t sort a list without breaking. That’s not a bug. That’s the whole point.
I’m talking about the cognitive dissonance that strikes every creator right now. On one hand, AI generates music that feels more meaningful than anything humans have ever made. On the other hand, it produces code that fails the most basic unit tests. If you’re a coder, you’ve felt that quiet superiority — I’m still better than the machine at this one thing.
That smugness is dangerous, but so is the awe. Both miss what’s actually happening.
The reason AI seems to have conquered music — the domain we thought was uniquely human, emotional, and spiritual — isn’t because AI has become a soulful artist. It’s because we have ridiculously low standards for what counts as “good” in subjective domains.
Think about it: when you listen to an AI-generated song, you’re not evaluating it against Mozart or Kendrick Lamar. You’re comparing it to the silence in your room, or to the mediocre pop song you heard in a coffee shop. The bar is on the floor. The AI doesn’t need to be brilliant — it just needs to be okay enough to trigger a dopamine hit of novelty. And novelty is cheap.
Code is the opposite. Code has a binary pass/fail. Either it compiles and produces the correct output, or it doesn’t. There is no “close enough” in production. An AI that writes a sorting algorithm that works 95% of the time is useless. A song that sounds 95% as good as a human hit? People pay for that.
The machine isn’t outplaying us at art. It’s outgaming our willingness to be easily amused.
This is the Gell-Mann Amnesia effect in action — you know a domain well enough to spot the AI’s failures (code), but you assume the same AI is magically perfect in a domain you know less well (music). The HN commenter who said they felt like a better coder than any model, yet considered the AI song “more beautiful than any song I have ever heard” — that’s the exact trap. They’re holding code to a rigorous standard and music to a purely emotional, uncritical one.
So what should you take away from this? Not that AI is overhyped. Not that AI is a threat. But that you need to calibrate your awe meter. Every time you see an AI demo that blows your mind, ask yourself: Would this pass in a domain where failure is unacceptable? If the answer is no, you’re just being played by a probabilistic parrot that figured out how to press your emotional buttons.
The real danger is not that AI will replace artists — it’s that we’ll start judging all AI capabilities by the easiest, most forgiving test we can give it. And then we’ll hand over control of the hard stuff because we got distracted by a pretty tune.
Stop asking what AI can do. Start asking what it can do correctly.
FAQ
Q: Isn't AI music actually impressive? Many people find it genuinely moving.
A: It can be impressive from a technical novelty standpoint, but moving ≠ correct. A song that makes you cry doesn't pass any objective test. Code either works or it doesn't. The emotional response to music is cheap for AI to trigger because it only needs to hit vague patterns humans find pleasant. That's not intelligence — it's pattern matching on a very forgiving target.
Q: What does this mean for developers and creators?
A: Stop using 'it blew my mind' as a metric for AI capability. For creative tasks where perfection isn't required, AI can be a useful tool — but for any task with a clear right/wrong answer, current models are unreliable. The practical implication: use AI for inspiration and drafts, but never trust it for production without a human expert verifying every output.
Q: So you're saying AI will never be good at code?
A: Not at all. AI will get better at code — but the bar is astronomically higher. A 1% error rate in music is imperceptible; a 1% error rate in code crashes a production system. The contrarian take: we should be <em>more</em> impressed when AI succeeds at code (even partially) than when it generates a pleasant song, because code requires actual logical consistency. Stop rewarding the easy wins.