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.

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.

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.

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.

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)

This is how I picture you as you’re saying that: https://youtu.be/JeNS1ZNHQs8

Has it genuinely boosted output? It’s been a while since I used copilot, but the code always required checking. More importantly, the interactions between the code provided by the prompt and the other code it’s working with. Has that all gone now and you can just straight rely on what it spits out?

1. Get a passion that isn’t work.
2. 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.