[banana_feather]

This just does not match my experience with these tools. I've been on board with the big idea expressed in the article at various points and tried to get into that work flow, but with each new generation of models they just do not do well enough, consistently enough, on serious tasks to be a time or effort saver. I don't know what world these apparently high output people live in where their days consist of porting Conway's Game of Life and writing shell scripts that only 'mostly' need to work, but I hope one day I can join them.

[throwaway4aday]

Not to pick on you but this general type of response to these AI threads seems to be coalescing around something that looks like a common cause. The thing that tips your comment into that bucket is the "serious tasks" phrasing. Trying to use current LLMs for either extremely complex work involving many interdependent parts or for very specialized domains where you're likely contributing something unique or to any other form of "serious task" you can think of generally doesn't work out. If all you do all day long are serious tasks like that then congrats you've found yourself a fulfilling and interesting role in life. Unfortunately, the majority of other people have to spend 80 to 90 percent of their day dealing with mind numbing procedural work like generating reports no one reads and figuring out problems that end up being user error. Fortunately, lots of this boring common work has been solved six ways from Sunday and so we can lean on these LLMs to bootstrap our n+1th solution that works in our org with our tech stack and uses our manager's favourite document format/reporting tool. That's where the other use cases mentioned in the article come in, well that and side projects or learning X in Y minutes.

[AndyNemmity]

I use it daily, and it's a time and effort saver.

And writing shell scripts that "mostly" work is what it does.

I don't expect it to work. Just like I don't expect my own code to ever work.

My stuff mostly works too. In either case I will be shaving yaks to sort out where it doesn't work.

At a certain level of complexity, the whole house of cards does break down where LLMs get stuck in a loop.

Then I will try using a different LLM to get it unstuck from the loop, which works well.

You will have cases where both LLMs get stuck in a loop, and you're screwed. Okay.. well, now you're however far ahead you were at that stage.

Essentially, some of us have spent more of our life fixing code, than we have writing it from scratch.

At that level, it's much easier for me to fix code, than write it from scratch. That's the skill you're implementing with LLMs.

[alexpotato]

> I don't expect it to work. Just like I don't expect my own code to ever work.

This line really struck me and is an excellent way to frame this issue.

[zarathustreal]

Any hints on why you’re adding so many newlines into your comment?

[walterbell]

Possibly entered on a narrow mobile device where one sentence can wrap into multiple "lines" that visually approximate paragraphs.

[kredd]

You get used to their quirks. I can more or less predict what Claude/GPT can do faster than me, so I exclusively use them for those scenarios. Implementing it to one's development routine isn't easy though, so I had to trial and error until it made me faster in certain aspects. I can see it being more useful for people who have a good chunk of experience with coding, since you can filter out useless suggestions much faster - ex. give a dump of code, description of a stupid bug, and ask it where the problem might be. If you generally know how things work, you can filter out the "definitely that's not the case" suggestions, it might route you to a definitive answer faster.