A year ago I started using AI coding assistants seriously — not just for boilerplate, but across the full stack of things I build: React Native apps, Node backends, PHP services, infrastructure. Here's what I actually found after twelve months of real use.
What genuinely got faster
Scaffolding and boilerplate. New API endpoints, migration files, UI components — the mechanical parts of building disappeared. What used to take an hour now takes five minutes. This alone is worth the price of any subscription.
Debugging unfamiliar code. Dropping a stack trace or an unfamiliar codebase into context and getting a working explanation is genuinely useful. I debug faster in code I don't own, and faster in my own code when I haven't touched it in months.
Context switching across languages. I work across React Native, Node, PHP, and occasionally Swift. The AI handles the syntax differences so I can stay in the problem space instead of spending ten minutes remembering how PHP handles array destructuring.
What didn't change
Architecture decisions still take the same amount of thinking. The AI is confident about a lot of things it shouldn't be confident about. Knowing when to push back — and noticing when the generated code is subtly wrong — requires the same engineering judgement it always did.
Code review got more important, not less. Fast generation means fast accumulation of technical debt if you're not paying attention. Every time you accept a block of code you don't fully understand, you're borrowing against your future self.
The honest number
I ship roughly 2–3× more per month than before. Some of that is AI. Some of it is that I've gotten better at scoping work and saying no to things that don't matter. Hard to separate the two — and I suspect that's the point.