I kind of miss when we weren't assumed to be using LLMs all the time

I use LLMs all the time — probably significantly more than last year, especially since agentic coding got decent. They are a huge help, and I do work faster, especially when dealing with legacy codebases — shout-out to the 1M+ LOC ColdFusion monolith I need to consult on a regular basis!

And yet, I feel like something was lost. I get stuff done, but it does not feel like my achievement in the same way it did a couple of years ago.

Partly, it’s a sort of imposter syndrome. LLMs feel like a very productive collaborator, though I still feel like I’m steering a savant most of the time. I know I will get used to it, even though I probably will — and should — remain uncomfortable passing off tasks mostly done by LLMs as my own. All the more reason not to get sloppy or offload thinking to the machine.

But there’s another aspect to this paradigm shift, one driven from the management level down to software engineers. We are assumed to be using LLMs, and expected to be more productive. To the point that I get asked, “Are you using LLMs for that?” every time I mention a delay in task execution.

I mean — I am using LLMs! But this is my work, and I’m responsible for the results. Everything is fine and great when LLMs speed up my work, but if I act on their assumptions blindly and things get borked up, can I shift the blame to an LLM? I haven’t tried that yet — maybe because I almost instinctively always own my mistakes — but I don’t think that would really fly with management.

So in the end, it all comes down to “work faster and better,” which I suppose has always been management’s message.

Still, it is so cringe to see that message repackaged as “AI will make you 10x more productive — embrace it, and everything will be swell.” I mean, you trust me with huge, complex distributed systems, and you think I can’t tell whether something makes me more efficient or not?

Maybe that’s the issue: I just want people to stop denigrating software engineering. It is complex work, and it is not only about coding. Spitting out 10x more code does not amount to a 10x increase in efficiency. Pretending that is the case is, at best, delusional — and hypocritical at worst.

Anyway, I hope this is just another hype cycle — the first one that directly affects my life, yes, but one that will eventually subside.

And if not — well, let’s see how it goes. The only thing I know for sure is that being an LLM operator is not an exciting career prospect.

This post has been proofread by GPT-5.4 - I tweaked grammar here and there based on its findings. I am pleased to say they were not many, so this particular skill has not withered away yet!