Three times in two weeks. Three different people told me you don’t need to know how to code anymore. Each time with genuine excitement, like they’d discovered a cheat code for the entire profession.
I’ve been writing code with AI for two years. The pattern I see is the exact opposite.
The developers getting the most out of LLMs aren’t the ones who stopped thinking like engineers. They’re the ones who think like engineers — and that’s exactly why they extract more from AI. Not knowing how to code isn’t a competitive advantage. It’s a ceiling you can’t see until you hit it.
Where engineering actually shows up
Building features. I can describe precisely what I need in a prompt. Which edge cases to cover. How a new piece should integrate with the existing codebase. That’s not prompt engineering. That’s software engineering through a different interface.
Debugging. When something breaks, I give the AI way more context than a pasted error message and an enter key. I know which files matter, where the state lives, what changed recently. That context is the difference between a useful answer and a hallucinated guess.
Choosing solutions. I understand the difference between building an MVP and delivering a feature for an enterprise client. Different constraints, different tradeoffs, different architecture. AI doesn’t know which one you need. You do.
The long game
This is where the “coding is dead” crowd really falls apart.
I know what a poorly designed database costs when you discover it after a year in production. I know that a vibecoded UI is not a production application. A working interface is not the same as well-designed entity relationships, secure data handling, and architecture that won’t collapse under the first real traffic spike.
You can’t prompt your way around these things. You learn them by building systems that break, debugging them at 2 AM, and redesigning them so they don’t break the same way again.
What actually changed
Syntax knowledge? Matters less. Fighting the compiler? Matters less.
But database design, system architecture, error handling — none of that went anywhere. These are the areas that separate someone who uses AI from someone who builds with AI.
The bar is higher than it was a year ago. For everyone. Not everyone has noticed.