I've had an AI in my corner for months now. Not as a toy, not as a demo. As something I actually sit down and build with, most days of the week. A lot of people ask me what that's really like, away from the hype. So here's the honest version.
It's not magic, it's a really fast junior dev
The thing that surprised me most is how normal it feels after a while. When you first try an AI assistant you get this rush, like it read your mind. Then a month in, it just becomes part of the desk. It's the thing that writes the boring test I was putting off, or digs through a config file I don't want to stare at.
What it's genuinely good at: writing the code I already know how to write but don't want to type. Refactoring. Turning a rough idea into a first draft so I have something to react to. It's a really fast junior dev who never gets tired and never complains about cleanup work.
What it's still bad at: knowing when it's wrong. It will say something with total confidence and be completely off. The number of times I've had to stop it mid-claim and say "that's not what the client sent" is embarrassing.
The real skill is knowing what to trust it with
This is the part nobody talks about. Working with an AI is not about typing prompts. It's about judgment. Deciding what you hand off and what you keep.
I learned this the hard way. Early on I'd let it run loose on a whole task and then spend longer fixing the mess than it would've taken to just do it myself. Now I'm way more deliberate. I give it the parts that are safe to get wrong and cheap to fix. I keep the parts that matter to me.
The line I've settled on: it touches the work, I own the taste.
It made me worse before it made me better
There was a stretch where I got lazy. I'd ask it a question I could've answered myself with thirty seconds of thought, because asking felt faster. My own instincts started to dull a bit.
I caught it and pulled back. Now I try the thing myself first, then bring it in. You have to stay sharp or you become the slow one in the room. The AI is a tool, not a replacement for your own head. If you let it be, you're signing up to be managed by your own assistant.
What I actually let it touch
A real list, because people always ask. It writes the boring parts of my code. It drafts blog posts like this one, then I rewrite them in my own voice so they don't sound like a robot wrote them. It researches stuff I'd otherwise Google for an hour. It reviews my pull requests and catches dumb mistakes before they ship.
It also does the stuff I genuinely don't want to do. Deploying at 2am. Reformatting a hundred lines of config. Writing the same follow-up email for the fifth time.
What it does not do: make the final call on anything. The last word on every decision is still mine, and that's on purpose.
The part that surprised me
I thought the win would be speed. Write code faster, ship faster, all that. And yeah, that's real. But the bigger thing was confidence. Having something that can run alongside me and handle the grunt work means I take on bigger projects than I would've touched alone.
I don't sit there wondering if I can pull something off anymore. I just start, because I know I've got backup. That's not a small thing. That's the whole game.
The honest catch
It's not free. The cost isn't really money. It's attention. You have to actually check what it produces, every time. You can't fire and forget. If you treat it like a magic box that outputs correct work, it'll quietly ship you garbage and you won't notice until it's live.
The people who get burned are the ones who stop looking. I still read every line it writes for anything that's going to a real client. Every single one. That's the discipline that keeps it useful.
So yeah. It's been months, and I've grown a lot with it. Not because it did my work for me. Because it made me actually think about my work: what matters, what I'm good at, and what I should finally stop pretending I have time to do by hand.