My hot take is that using Cursor is a lot like recreational drugs.
It's the responsibility of the coder/user/collaborator to interact with the code and suggestions any model produces in a mindful and rigorous way. Not only should you have a pretty coherent expectation of what the model will produce, you should also learn to default/assume that each round of suggestions is in fact not going to be accepted before it is. At the very least, be prepared to ask follow-up questions and never blindly accept changes (the coding equivalent of drunk driving).
With Cursor and the like, the code being changed on a snippet basis instead of wholesale rewriting that is detached from your codebase means that you have the opportunity to rework and reread until you are on the same page. Given that it will mimic your existing style and can happily explain things back to you in six months/years, I suspect that much like self-driving cars there is a strong argument to be made that the code it's producing will on average be better than what a human would produce. It'll certainly be at least as consistent.
It might seem like a stretch to compare it to taking drugs, but I find that it's a helpful metaphor. The attitude and expectations that you bring to the table matter a lot in terms of how things play out. Some people will get too messed up and submit legal arguments containing imaginary case law.
In my case, I very much hope that I am so much better in a year that I look back on today's efforts with a bit of artistic embarrassment. It doesn't change the fact that I'm writing the best code I can today. IMO the best use of LLMs in coding is to help people who already know how to code rapidly get up to speed in domains that they don't know anything about. For me, that's been embedded development. I could see similar dynamics playing out in audio processing or shader development. Anything that gets you over that first few hard walls is a win, and I'll fight to defend that position.
As an aside, I find it interesting that there hasn't been more comparison between the hype around pair programming and what is apparently being called vibe coding. I find evidence that one is good and one is bad to be very thin.
It is kind of like reviewing PRs from a very junior developer that might also give very convincing but buggy code and has no responsibility. Seriously don’t see the point of it except doing copy paste refactoring or writing throw-away scripts. Which is still a lot so it is useful.
It needs to improve a lot more to match the expectations and it probably will. It is a bit frustrating to realise a PR is AI generated slop after reviewing 500 of 1000 lines
This is actually supporting my point, though (again IMO).
There's a world of difference between a very junior dev producing 1000 line PRs and an experienced developer collaborating with Cursor to do iterative feature development or troubleshoot deadlocks.
Also, no shade to the fictional people in your example but if a junior gave me a 1000 line PR, it would be part of my job as the senior to raise warning bells about the size and origin of such a patch before dedicating significant time to reviewing it.
As a leader, its your job to clearly define what LLMs are good and bad for, and what acceptable use looks like in the context and environment. If you make it clear that large AI generated patches are Not Cool and they do it anyhow... that's a strike.
I think the best usecase by far is when you don't know where to start. The AI will put something out there. Either it's right in which case great, or it's wrong, and then trying to analyze why it's wrong often helps you get started too. Like to take a silly example let's say you want to build a bridge for cars and the AI suggests using one big slab of paper maiche. You reject this but now you have two good questions: what material should it have? and what shape?
I did start by disclaiming a hot take, so forgive my poetic license and unintentional lede burying.
What I'm trying to convey is a metaphorical association that describes moderation and overdoing it. I'm thinking about the articles I've read about college professors who are openly high functioning heroin users, for example.
Every recreational drug has different kinds of users: social drinkers vs abusive alcoholics, people who microdose LSD or mushrooms vs people who spend time in psych wards, people who smoke week to relax vs people who go all-in on slacker lifestyle. And perhaps the best for last: people who occasionally use cocaine as a stimulant vs whatever scene you want to quote from Wolf of Wall Street.
I am personally convinced that there are positive use cases and negative use cases, and it usually comes down to how much and how responsible they are.
It’s really just another instance of ‘this is why we can’t have nice things’ - because the people who don’t have their shit together are always going to ruin it for the people that can handle their shit. And further, because the people who don’t know any better will fixate on one aspect or the other - drugs are the devil, drugs are god - AI is the devil, AI is god - because they’re bad at nuance and can’t just leave it alone.
It's the responsibility of the coder/user/collaborator to interact with the code and suggestions any model produces in a mindful and rigorous way. Not only should you have a pretty coherent expectation of what the model will produce, you should also learn to default/assume that each round of suggestions is in fact not going to be accepted before it is. At the very least, be prepared to ask follow-up questions and never blindly accept changes (the coding equivalent of drunk driving).
With Cursor and the like, the code being changed on a snippet basis instead of wholesale rewriting that is detached from your codebase means that you have the opportunity to rework and reread until you are on the same page. Given that it will mimic your existing style and can happily explain things back to you in six months/years, I suspect that much like self-driving cars there is a strong argument to be made that the code it's producing will on average be better than what a human would produce. It'll certainly be at least as consistent.
It might seem like a stretch to compare it to taking drugs, but I find that it's a helpful metaphor. The attitude and expectations that you bring to the table matter a lot in terms of how things play out. Some people will get too messed up and submit legal arguments containing imaginary case law.
In my case, I very much hope that I am so much better in a year that I look back on today's efforts with a bit of artistic embarrassment. It doesn't change the fact that I'm writing the best code I can today. IMO the best use of LLMs in coding is to help people who already know how to code rapidly get up to speed in domains that they don't know anything about. For me, that's been embedded development. I could see similar dynamics playing out in audio processing or shader development. Anything that gets you over that first few hard walls is a win, and I'll fight to defend that position.
As an aside, I find it interesting that there hasn't been more comparison between the hype around pair programming and what is apparently being called vibe coding. I find evidence that one is good and one is bad to be very thin.