Hacker News Reader

Comment by pixel_popping | Hacker News Reader

Let's talk about it again in 5 years, but 1-2 years from now, at the very least, coding will be over in the sense that the best models will do it better than the best (or the 99.99%). I don't think I'm hallucinating no, when my own work went from coding+managing+bunch of other stuff to just orchestrating and my output is just insanely higher and I literally have a bunch of friends that went from coding 8h a day to just "pretending to code" and just using a bunch of agents and get paid the same salary for working 30min a day, that's real, not an hallucination.

View on HN

How are you/they instructing those agents? If you are writing detailed spec.md and reviewing those results, you are _still_ programming. Just in pseudocode effectively. I’ve seen enough session transcripts and detailed prompts that would have been easier to just write the code instead!

> in 5 years

That's literally the same argument that the blockchain gurus made, and each following year it was still 5 years in the future. I'm getting strong Real Soon Now™ vibes.

Bitcoin was never actually valuable for the average person except if they got lucky by timing the speculation bubbles right, or if they were buying illegal drugs online.

Lots of AI tools already add actual value and they're only getting better. Every software dev I know uses Claude at some level. Whether it will be the next trillion dollar unicorn might be overhype, but in terms of demonstrating its general utility, it's already there. No need to wait 5 years.

It's really 2 very different things, only the "shilling" might be deja-vu.

Haha. My favorite shillings will always be NFTs and the "let's do email job pointless meetings in Second Life and call it Metaverse". Although the former is/was pure shilling, and the latter is/was a horrible pipe dream shilled as useful.

A Machine Learning + Large Language model is one popular technology. Decentralized blockchain tokens is another popular technology. Besides hype, there's not much in common with those technologies.

> Every software dev I know uses Claude at some level.

here is my leve:, try to get claude to properly perform some boring boilerplate for me until the tokens run out or i get super angry and then do it myself in a rage.

I find the best use of chatbots re: software engineering is more of an office hours with the professor than it is a fellow colleague; Use it for specifics. Throwing things over the fence to it often causes a reach into the bag of garbage spaghetti code that is half the internet content.

I've had luck with 'explain this bullshit code i need to update that no-one has touched in two years and none of the committers even works here anymore'. Getting it to actually add the feature, a bit of a crapshoot.

what stack do you work in, and how are you prompting it?

Ah, the old 'you're using it wrong'.

Either it's smart or it's not.

In my experience it IS very smart for 90% of use cases. I'm curious to identify how someone could be using it in a way where that's not the case.

I hated AI tools a year ago and now I use them every day. If someone's opinions of AI were last updated > 6 months ago I don't think they've really given it a shot.

thank you.

If you said "This car is stupid, I can't drive it" and it turned out you weren't putting the key in the ignition...

my dude, i stopped listening to your pompous ass, take a hint.

Clearly you didn't ;)

No, AI has real, immediate, large-scale industrial and scientific applications. Cryptocurrency is a massive disappointment, but AI has genuine potential value.

> genuine potential value

That's some beautiful mental and verbal gymnastics.

That's some cynical ignorance.

common, that's very different, that's something current with practical use-cases that are already being implemented across all companies, I don't even know why we compare this with blockchain, blockchain is just some fancy resilient DB with proofs in the end.

or elson and fsd.

Reply on news.ycombinator.com