A LinkedIn post crossed my feed this morning. It summarized Dario Amodei, Anthropic’s CEO, answering the question everyone keeps asking: should you still learn to code?
The summary leads with “coding is going away first.” The models are already doing it. Software engineering takes a little longer, but it’s going too. If you’re learning to code for job security, you’re learning the wrong thing.
I read it twice. It made me frustrated. Not because it’s wrong, exactly. Because it concedes the premise.
The premise is the problem
Let me be clear about one thing first. I’m not here to say Amodei isn’t smart, or isn’t good at his job. He’s both. He runs one of the most important companies on earth, and he thinks carefully about this stuff. The full interview is more nuanced than any six-bullet summary.
But the framing buries the thing that actually matters. “Coding is going away” treats coding as a synonym for typing syntax. It isn’t. It never was. I’ve argued before that the end of coding won’t be in 2026, and the reason is the same one I’m about to make here.
Coding was never about syntax. It was never about being an expert in some language. Coding is about solving problems. It’s about finding an elegant solution that brings value. Good code brings that value because it works better, most of the time, than bad code. But the goal is the solution. It always has been.
The language was always immaterial
You can serve billions of pages a day with Python. Or Rust. Or Go. Or Java. Instagram ran on Python at massive scale. The Linux kernel is C. Discord moved hot paths from Go to Rust because the problem called for it, not because one language is holy and the other is cursed.
The same problem gets solved well in any of them. That’s the whole point. If the language were the job, you couldn’t swap it out and still ship. But you can. The goal is the elegant solution to the problem you’re solving for. The syntax is just the medium you happen to be working in.
So when someone says “coding is going away,” what they actually mean is “typing syntax from memory is going away.” Fine. Good, even. I don’t miss memorizing flag orders for tar. And I still have to look up the syntax for crontab every single time. But that was never the craft. That was the friction around the craft. That friction going away, or just mattering less, is a net positive.
What the models don’t have
Agentic development doesn’t change the core purpose one bit. The models are getting very good at producing syntax. They’ve read all of it. They can write you a working function in any language faster than you can describe it.
What they can’t do is understand an elegant solution.
And I want to be precise here, because “they don’t understand elegance” is easy to wave away. The models can produce code that looks elegant. They’ve seen every clean abstraction ever published. What they don’t have is a stake in the problem. Elegance isn’t a property of the code sitting on the page. It’s a judgment about fit. Does this solution actually match the problem some real person actually has?
That judgment is the craft. It comes from having been burned. From shipping the wrong thing and watching it fail in production. From sitting with a user and realizing the feature they asked for isn’t the feature they need. The model has read about all of that. It has lived none of it. It has no relationship to your problem.
That’s the gap. And it’s not a syntax gap that the next model closes. It’s a different kind of thing entirely.
The two ways people get this wrong
Here’s what bugs me most. The people who don’t understand that software is about solving problems end up on both sides of the AI fight.
On one side, you’ve got the people misusing AI right now. They think the model is the engineer. They accept whatever it produces because it compiles and the tests are green. They’ve outsourced the judgment along with the typing. They generate a thousand lines they can’t evaluate, and they call it productivity. They confused the medium for the work, so they handed over the work and kept nothing. This is the classically defined “AI slop” wreaking havoc in startups and large enterprises alike.
On the other side, you’ve got the people too scared to touch it. They think using AI means giving up the craft, so they refuse on principle. But they’re making the same mistake from the opposite direction. They also think coding is the typing. So protecting the typing feels like protecting the craft. It isn’t. You can hand the syntax to a model and keep every bit of the judgment that made you good. That’s the whole move.
Both camps misread the same thing. The job was never the keystrokes.
What I actually do
I use these tools every day. I lean on them hard. They draft, they scaffold, they handle the boilerplate I used to grind through by hand. On my passion project, BrandCast, I move faster than I ever could have with traditional hand-written code.
But I’m still the one deciding what gets built and why. I’m the one who knows when the elegant-looking solution is wrong for the actual problem. I read every line that matters. I throw away plenty of what the model gives me, not because it doesn’t work, but because it doesn’t fit. That’s me doing the job. The job I was always doing.
The act of brutal simplicity is my primary goal when using Agents and creating agentic SDLC workflows.
I’ve written before about an agent that confidently broke my whole app overnight. That wasn’t a syntax failure. The code it wrote was fine. It failed because it had no judgment about what I actually wanted. That’s the gap, and it’s the same gap.
How the solution gets crafted is changing. Fast. The act of craftsmanship is not.
So should you learn to code?
Wrong question. Learn to solve problems. Learn to tell a good solution from one that just runs. Learn to sit with a problem long enough to understand what it actually is.
The syntax was always going to change. New languages, new frameworks, now models that write the syntax for you. None of that touches the part that mattered.
The tools change. The craft endures.
The summary that set me off is here. The source is Dario Amodei on Nikhil Kamath’s “People by WTF” podcast: watch the full interview or read the transcript. Both are worth your time. Tell me where I’m wrong on LinkedIn.