I still remember the first time I used GitHub Copilot and just sat there for a second, kind of stunned. I’d typed a comment describing what I wanted, half expecting garbage, and it just… wrote the function. Not perfectly. But close enough. I remember thinking, okay, this is going to change things, and not really understanding yet how much.
A few years later, here we are. Most days I barely write code from scratch anymore. I describe what I need, I get a draft, and then my job becomes figuring out if the draft is actually good or just looks good. That’s a different skill than the one I spent a decade building, and honestly some days I’m not sure I’m better at it yet.
The boring stuff got way less boring
The first thing that changed was the small annoying stuff. Writing a parser for some weird CSV format. Remembering the exact syntax for a regex lookahead. Setting up boilerplate for a new pipeline. None of that requires real thinking, it just requires knowing the syntax, and AI is great at that. I used to lose whole afternoons to this kind of thing. Now it’s minutes.
What I didn’t expect was how much it would also help with stuff I thought only a human brain could really do, like reading through a messy old codebase someone left behind years ago and trying to figure out why it’s structured the way it is. I can just ask now.
My actual job changed shape
Here’s the part that took me longest to notice. I used to think my value was in writing good code. Turns out a lot of my value was always in deciding what code should exist in the first place, and that part never went away. If anything it matters more now.
Knowing the syntax for something used to be worth a lot. Now it’s worth almost nothing, because the model already knows it. What’s worth a lot is judgment. Can you tell when a suggestion is technically fine but a bad idea long term. Can you spot the moment where the “clean” answer is going to cause a mess in six months because it ignores some annoying constraint that only exists because of a decision made three years ago. AI doesn’t know about that decision. I do. That’s the value I bring now, more than the typing.
Code review feels different
This one is a little uncomfortable to admit. I review pull requests now where I genuinely don’t know if the person who wrote it understood every line, or just kept accepting suggestions until the tests passed. Most of the time it’s fine. The code works. If a change looks too smooth, too easy, I’ll ask more questions, not less. It sounds backwards but an unusually clean looking diff is sometimes the one that needs the most poking.
I worry a little about the people just starting out
I learned how to actually debug by writing bad code and suffering through fixing it myself. That slow, annoying process is honestly where most of my instincts came from. I’m not totally sure what replaces that for someone starting their career right now, when a lot of the early struggle can just get skipped over with a good prompt.
Maybe it doesn’t matter and new engineers will build good instincts a different way, by reviewing and steering instead of grinding through syntax errors like I did. Or maybe we’re quietly cutting out a step that mattered more than we think. I don’t know yet. I don’t think anyone really does.
What hasn’t changed at all
I want to be fair here, because it’s easy to make this sound like a bigger deal than it is in some ways. Talking to a product manager and figuring out what they actually need, even when what they’re asking for doesn’t quite make sense, that’s still entirely on me. Knowing which problems are even worth solving, same thing. AI hasn’t made me better at any of that. It just made me faster at the part of the job that, honestly, was never the hardest part to begin with.
And there’s something I try to protect on purpose now. Every once in a while I’ll turn off the autocomplete and just write something by hand, slowly, the boring way. Not because I’m precious about it. Because I noticed that the slow way still teaches me things that reading a finished answer doesn’t.
Where that leaves me
I don’t think the job is going away. I think it’s shifting, away from producing code and toward deciding if the code that got produced is actually right. That’s not necessarily easier. It takes a strange mix of confidence and humility, confidence to override a suggestion that looks okay but isn’t, and humility to accept one that’s better than what I would have written myself.
I’m not nostalgic for typing out boilerplate all day, I really am not. But I’ve started being more careful about which parts of the work I let the AI take over completely, because it turns out some of the slow parts weren’t just slow. They were how I got good at this job in the first place.
Anyway, that’s where I’m at right now. Ask me again in a year and I bet half of this is already out of date.