Developers aren't being replaced, but coding might be
The problem with AI coding isn't that it's bad. It's that it doesn't know when it's bad. The AI says it fixed the bug. It didn't. It moved the bug ten lines down. Or it changed something you weren't even working on. Or it just lies.
I've been there. A lot.
The quicksand thing
Early on, AI coding feels great. Quick wins. You're flying. Then you hit the details and it all turns to mud.
I've watched people (myself included) get stuck for hours while the AI confidently regenerates the same broken code. "I've fixed the issue," it says. It has not fixed the issue. You point out the error. It apologizes, changes something, breaks something else. Three hours later you're mass-reverting commits and wondering why you didn't just write the damn thing yourself.
So I stopped trusting its word. Now it has to prove it worked.
How I stopped babysitting
My setup: the AI doesn't get to declare victory. Tests run automatically. I have subagents whose only job is checking if problems are actually solved. A reviewer that scans for regressions before anything gets committed.
It's a lot of overhead. But without it you're basically a parent watching a toddler with scissors.
Most people skip this part. They treat AI like a junior dev who needs constant supervision, and then they're surprised when supervising it takes all day. I'd rather spend a few hours setting up guardrails than spend every hour being a guardrail.
Are developers being replaced?
I keep seeing this question. Short answer: we won't need as many people who write code. We'll need more people who can figure out what code to write.
The world never runs out of problems. Solve one, two more appear. Developers have always been lucky. We got paid to be problem-solvers. But some developers never actually liked that part. They just wanted to write code. To nerd out over syntax and architecture patterns.
Those people are probably in trouble.
The rest of us are still needed. The best developers I know aren't fast coders. They're the ones who can stare at a mess and see the actual problem buried underneath. AI can write code all day. It can't tell you which code matters.
What actually changed
I spend more time now defining what "done" looks like. Writing test cases before I write prompts. Setting up systems so I'm not manually checking every output.
It's the same work, honestly. Just with a faster, dumber assistant who needs more structure.
What's your setup? I'm still figuring mine out.