Hey all,
not sure if this is an ADHD thing, but it feels like it is. So, I’m a mid/senior level engineer who’s been coding professionally for a while now. Before LLMs dropped, I used to get a legit dopamine kick from fixing even the tiniest bugs and getting things to compile. Tedious debugging and diving into docs? Bring it on! It was all part of the fun.
But ever since LLMs came along, that spark has kinda vanished. I feel like my skills are fading by the day. It’s like I can’t bring myself to code manually or look things up anymore. I know exactly what to ask LLMs and how to fix issues, but the thought of doing it all manually? No thanks.
Now, the only time I get that dopamine hit is when I can implement entire features that should take days in just a few prompts. Anything less feels like a waste of time. I hate feeling so dependent on it, especially since I know the code it spits out isn’t always top-notch. I know how to fix it, but I just can’t bring myself to do it. Especially tools like Cursor with its agentic coding make it even harder for me to leave prompts behind.
Coding used to be my passion, and now it feels like LLMs has messed with that while boosting my output. Anyone else feeling this way? Any thoughts or advice?
You must log in or register to comment.
I’m a senior dev and the amount of bullshit code that LLMs generate is just unacceptable if you are copy pasting. You should take the output as you would take a junior coders output and code review it the same way and then use your own version in the commit. Don’t be lazy to just use some of the ideas and completely refactor the output if needed. You will find that LLMs accuracy drops dramatically when you go from easy/common -> average/uncommon -> hard/rare problems. They also don’t care much for optimization.
Train yourself to use the AI only for doing grunt work. I used it yesterday to split 1 cypress test into 3 separate ones for example. I force myself to solve the problems and don’t lean too heavily on AI (it doesn’t increase output anyway in my experience). Copilot gets tons of shit wrong that I end up having to manually write.
- Get a passion that isn’t work.
- I have the opposite problem. I for some reason refuse to use AI. I worry that I’m becoming one of those tech dinosaurs that refuses to learn something new because it isn’t what they’re used to (but I’m only 36). I am opposed to using AI for many reasons but I’m also slightly worried this will harm my job prospects in the future if I ever have to find a new job.
It’s a crutch. Some perspective introspection questions: Is it really boosting your productivity? If the LLM can do most of the nonsense being asked of you, maybe it is time to consider a new role with more challenge and skill required?
Mid career is the time to take on larger projects or deeper technical fixes/augmentations now that you have a solid foundation of skills. Without this kind of experience you’ll be one of those “seniors” that has the title because of tenure, but not because merit/skill.
I am undiagnosed ADHD and vibe coding just kills the magic. No longer am I in control. Can’t achieve flow if I am not the one actually doing the work. Vibe coding is really being a ((n) engineering, project, etc) manager and that takes a different, uninteresting skill set plus patience to deal with the computer’s fuck ups. I refuse to vibe code at work (fintech). Hell, I can’t even use the LLM autocomplete nonsense in Intellij because it confabulates API like a motherfucker.
While the thinking sand is amazing relative to unskilled, for those of us that know better, it is a really shitty junior dev that sometimes can’t even write code that compiles and will randomly fuck up other working code. And I just don’t have the time or the patience to deal with that shit.
To answer your question: It is boosting my output, not my productivity. I feel like I am always following the same pattern. At first, when I take on a complex project, it’s challenging and forcing me to do a lot of research and upfront work. At this stage I enjoy doing the manual programming part. But after getting the initial implementation productive, and starting to do maintenance and implementing feature requests, I loath doing manual coding. And this issue became worse with the introduction of LLMs. I think you might have a point - I need to seek more challenging work, I don’t want to become one of those sensiors because of tenure - you’re points are valid.
Agree, vibe coding kills the magic. I also can’t achieve a flow while doing it. But somehow I’m already too much dependent on it, so I lost the ability to get in a manual coding flow.
I’m thinking about ditching all those tools altogether and going back to neovim with a simple LSP, without any auto complete and AI tooling. But I’m scared of getting too frustrated and jumping right back into the AI tools.
Maybe stepping back slowly could help. Like if there are three AI tools that you usually rely on, this week only use two of them, then the next week only use one, etc.
I feel like the debugging dopamine comes from the fact that it’s usually a well scoped problem that’s easier to understand than implementing a new feature while still being important. At least that is how I explain it to myself. It sounds like that is taken away from you, since the llm kinda opens up the scope in programming.
I think you are right. Scoping is definitely a factor.
yeah.
it feels kinda like “what’s the point”. i’ve been doing this for 10 years and i love the problem solving parts, but the stuff around it takes more and more effort to do, especially when an llm could do boilerplate in a quarter of the time with some manual checking. if it can do that, why can’t it just do all of it.
Junior/mid here. AI helps me solve annoying problems, get through the boring monotonous stuff faster and do more of the fun high level or creative stuff. (I’m mostly a frontend dev)
