You're not alone in this.
And you're not imagining it.

The autocomplete got so good I stopped reading the code it generated. I'd tab through, see green, merge. Then one day I had to debug something in my own PR — something I hadn't written, just accepted — and I couldn't explain a single line. Not one. That's when I realized: the autocomplete hadn't made me faster. It had made me optional.

Sunday nights started feeling different. Not the normal weekend-is-over anxiety — something worse. I'd open my laptop to preview the week and immediately feel behind. Not because there was more work. Because every task already felt answered before I started it. Why am I even showing up?

I used to bisect in my head. Narrow down where a bug was, trace the logic, find the exact commit. Then I started asking AI to bisect for me. After six months, I tried it manually and couldn't do it. I tried three times. The skill was just... gone. Not forgotten — unbuilt. I hadn't practiced it in so long the pathway had degraded.

A junior on my team used to ask great questions. Then they got good at AI. Now they ship features and can't explain why the tests pass. They can tell me the AI said it was correct. That's it. I'm watching someone skip the part where you build the model — the part that takes years — and the model is not building.

My velocity went up 3x and my confidence went down 2x. I'm shipping more than ever and understanding less than ever. My manager sees the velocity. I see the gap between what ships and what I would have shipped. Nobody else sees that gap. But I feel it every day.

As a staff engineer, my job used to be holding the architecture in my head. Knowing why the system was built this way, where it would break, what the next constraint was. I could have long conversations about this — in the architecture review, in 1:1s, in design docs. Now I can hold less of it. The AI handles the parts I used to think about. And the parts I used to hold are getting harder to hold.

I took a job interview — just to see where I stood. An algorithm problem I've solved a hundred times. I couldn't get to a solution without AI suggesting the approach within 30 seconds. I stopped the interview. I knew the answer to the problem. I couldn't get there without watching someone — something — suggest the path. I haven't taken another interview since.

2am, prod is on fire. I open the AI and it gives me a runbook. I follow it. The symptoms change. I ask for the next step. It gives me another. After an hour, the incident is resolved and I couldn't tell you what happened. I just followed a tree. A $50B company was depending on me and I couldn't tell them what happened because I didn't know. I just navigated.

I haven't written code without AI in 14 months. Not one line. I can feel that I'm not an engineer anymore — not in the way I used to be. When I try to start something on my own, the resistance is total. It's not writer's block. It's like the machinery that used to just... run... has stopped running. I know what I want to build. I can't get there without the safety net. And the safety net is the thing that won't let me learn to do it myself again.

Got laid off in February. Started interviewing. Realized, sitting in a technical screen, that I couldn't solve the problems I was expected to solve. Not because I didn't know the concepts. Because the pathway from concept to solution — the working-through-it part — I hadn't done that in over a year. I could explain what AI would have done. I couldn't do it myself. That was the most frightening thing I've ever experienced in this career.

Week one: terrible. Everything took three times as long. I couldn't stop reaching for the shortcut. Week two: slightly less terrible. Week three: I debugged something on my own — traced it through three files, found the root cause — and the feeling was unlike anything I'd felt in a year. Not just "I solved the problem." More like: "I am still an engineer." That's the thing I didn't know I was losing.