Coding is Dead?

I keep hearing this at conferences. Saw it three times in my feed last week, and someone will actually still say it.

Coding is dead. AI writes the code now. Learn to prompt.

And the annoying thing is, I get why it lands. I personally use Cursor. Tasks that used to eat my whole afternoon happen n twenty minutes now. Last month i scaffolded an entire admin dashboard before lunch and still had time to get annoyed by something unrelated.
The productivity thing is real, I'm not going to pretend otherwise.

But I keep thinking about something that happened maybe four months ago.

I was building async file processing; job queue, and the AI produces a genuinely clean code. Readable, well-structured, the kind of thing you'd be happy to find in a codebase. I ran the tests. Worked perfectly. I was basically done.

But I didn't ship it, not because there was a bug, but because I just felt something was off with the code. At that particular moment I wasn't sure what it was, and in deciding to dig in, I found a race condition buried deep in the pattern, this would have completely destroyed data under concurrent load, when now users start using the application. Of course now we're in production lol.

Now, this is the thing nobody is saying clearly enough: AI doesn't replace judgement. It generates plausible code very fast, but plausible is not same as correct. The distance between those two things is exactly where engineering actually lives, and you only develop the ability to measure that distance by spending time on the wrong side of it.

There's something almost cruel about how this plays out. The developers who are loudest about "anyone can build anything now" are, without exception, deeply technical people.
I am excluding the "i am a software developer" kind of people we see on youtube in this lol, atleast i hope we know real builders.

Now, the actual builders, have internalized the fundamentals so completely that they can safely delegate the execution and still supervise it well. They're not replacing their knowledge. They maximizing it by using using tools as a multiplier. When they say "you don't need to learn to code" they're describing their own situation incorrectly. They needed to learn to code. They did, and now they're describing what came after.

The people who will take the hit are newer developers who'll skip the foundations, build things that mostly work, and have no vocabulary for what's wrong when they break. Not wrong in obvious ways, wrong in the ways that only show up at scale, or under edge cases, or when two systems interact in ways nobody predicted.

You can refactor your way out of technical debt, but for mental model debt; not knowing why something is broken? Not having the pattern-matching to even suspect the right area? That doesn't refactor. You just have to go back and do the work you skipped.

I'm granting the honest version of the counterargument: the tools are getting better fast. Mistakes that require an experienced engineer to spot today might be caught by better static analysis in two years. There's probably a real floor drop happening at the routine end of engineering work. The established patterns, the well-documented domains, the stuff that follows predictable shapes. That is real.

But someone still has to make the judgement calls on architecture. Someone still has to know when the schema that looks clean today will be a nightmare to query in eighteen months.
Someone still has to push back when a product decision creates a technical constraint that will cost ten times as much to fix later. That's taste, and taste accumulates slowly, through building things and breaking them and knowing why.

To everyone early in tech career, using AI tools is right, they're genuinely useful for learning faster. But don't use them as a substitute for breaking things yourself. The thing you're really trying to build isn't the app. It's the mental model of why the app is the way it is. You can only get that by being the person who made it wrong first.

The floor didn't disappear. It just got easier to accidentally fall through.

*Note: By the way, AI credits are getting expensive by day, so you better know how do the thing by hand, hahaa;*