I've used AI coding tools every single working day for about a year. Not for demos — for real, shipped, production work. Long enough to get past both the hype and the backlash. Here's the honest version, the one I'd tell a friend over coffee.
What actually got faster
The boring stuff. Boilerplate, glue code, the first draft of a function, translating an idea into a framework I half-know, writing the test I was going to skip. The "I know exactly what I want, I just don't want to type it all" tasks collapsed from an afternoon to a few minutes.
That part is real, and it's not small. A surprising amount of engineering is typing things you already understand.
What did NOT get faster (and got a little harder)
Knowing what to build. Deciding the tradeoff. Holding the whole system in your head. Figuring out why the thing is actually slow. Saying "no" to the clever solution and picking the boring one that survives.
If anything, AI made these more important, because it removes the friction that used to slow you down before you'd thought it through. It'll happily generate 200 lines of confidently wrong code. The bottleneck moved from writing to judging.
The skill that quietly became everything: review
A year ago, my main skill was writing code. Now my main skill is reading it — fast, skeptically, deciding in seconds whether to keep it, fix it, or throw it out. AI made me a senior reviewer of a very fast, very eager junior who never gets tired and never gets offended.
The engineers getting the most out of this aren't the ones who trust it most. They're the ones who trust it least by default, and verify quickly.
The honest failure mode
The trap isn't that AI writes bad code. It's that it writes plausible code, and plausible is exactly what slips through when you're tired. The worst bugs I've seen this year weren't typos — they were confident, well-formatted, completely reasonable-looking lines that were subtly wrong. You only catch those if you still understand the thing yourself.
So my one rule: never ship code you couldn't have written and can't fully explain. The day you do, you've stopped being the engineer and started being the rubber stamp.
So — replacement?
Wrong question. It's not replacing engineers; it's deleting the gap between knowing and doing. That rewards people who know things and have taste, and punishes hand-waving. The leverage is enormous if you bring judgment, and dangerous if you bring it instead of judgment.
I'm more productive than I've ever been. I'm also reading more carefully than I ever have. Both are true, and I don't think that's a coincidence.
If you code with AI daily: what's the one thing you refuse to let it do for you?
Top comments (0)