There's a post going around right now. Short one, from a developer who said he's tired of talking to AI.
He described opening a GitHub discussion, getting a reply, realizing it was the same word-for-word answer ChatGPT had given him. Then another reply, same text again. Then he was messaging someone on Reddit and figured out mid-conversation that he wasn't talking to a person at all.
It's sitting at the top of Hacker News with 1,200 points and 600 comments. I read it at like midnight and I haven't stopped thinking about it.
The thing is, I do the opposite of what he described. I'm not the one sending AI answers to people. I'm the one who asks AI first, before I even let myself sit with the question.
Someone asks me to set up a pipeline. Before I think about what I actually know, I'm already asking Claude. I need a script. Straight to Claude. Some error I've seen twice before. Still Claude.
I've been building systems to make my process faster. And that part is working. But I'm starting to wonder if I've quietly trained myself to not start anything without checking in first.
Here's what I mean by that.
Before I used AI this heavily, there was a delay. I'd sit with a problem for a while. I'd try to figure it out, fail a bit, try something slightly different. It was slower. It was also when I was actually learning things.
Now the delay is almost gone. That feels productive, and a lot of the time it is.
But I'm noticing that when I do solve something myself, when I actually resist asking and push through the friction, it lands differently. I remember it. I have opinions about it. I understand why it works, not just that it works.
I don't know if this is a real tradeoff or just me overthinking it. Maybe both.
The post that went viral is really about something else under the surface.
It's not about AI being wrong. It's about not knowing who you're actually talking to anymore. It's about the feeling that the texture of an interaction has changed and you can't quite point to when it happened.
I think there's a version of that same thing that happens internally, just in your own head.
When I ask AI before I think, I'm not getting a wrong answer. I'm getting a fine answer. But it's someone else's answer, more or less. I never went through the process of arriving at it myself.
Over time that adds up. I'm not sure into what exactly. But it adds up.
I'm not trying to say AI is bad or that I should use it less. I work with it constantly and I'm going to keep doing that.
What I'm actually sitting with is the question of ordering.
Like, what if the rule was: try to figure it out yourself first. Even just for five minutes. Then ask.
Not because the self-derived answer will always be better. It won't be. But because going through that process, even briefly, means you understand the answer when it comes. You're evaluating it. You have a frame for it.
Right now I sometimes get an answer from Claude and just run with it. I trust it enough that I skip the part where I verify it with my own judgment.
That's probably fine most of the time. I'm not sure it's fine all of the time.
I'm still figuring out where I actually land on this.
Part of me thinks the speed is worth whatever I'm giving up. Part of me wonders if I'd be a better builder if I slowed down and let myself struggle more.
The weird thing is that the people building the best stuff I see are using AI constantly. And also, they seem to have very strong opinions. Very clear mental models. They're using AI as something to argue with, almost. Not just a place to get answers.
Maybe that's the difference.
Not less AI. But bringing more of your own thinking to the table before you sit down with it.
I want to try that for a while and see what actually changes.