[gdilla]

Yup, sw engineering is a slow march to being commoditized. Some things will remain hard (only because it's cutting edge and pushing the limits of something) but known patterns and services will be just-yell-at-ai to stand up. A lot of businesses can run on the latter, i guess - but at that point the challenge is having a viable business, not the software development of X.

[reverendsteveii]

our industry has existed on the cutting edge doing what's hard since its inception. it's just that there was a time when sending a piece of text across a wire was hard. Now that's easy, so we do more with the tools that make that easy. When what's hard today becomes easy we'll do that quickly with the tools that make it easy and then do more hard stuff. We can say we've achieved AGI when the tools are doing better on their own than a tool plus an engineer would do, and I think that's a long way off.

[svachalek]

Exactly. This is how it's always been. LLMs make it easy to spit out boilerplate code, which drives the price of boilerplate down to free. But good engineers will add a lot more value to that which raises the bar for everyone. The things you can create with an LLM become boring and worthless (honestly they mostly already were before coding agents came out) and the hirable skills become everything else that engineers need to do.

[reillychase]

Good take

[reillychase]

I still think that might be oversimplifying what software creation is which is being able to explain to a computer what it is you want. I think of Cursor as Python was to C. It's a higher level language but you still have to be able to think like a hacker, which will always be a rare skill.

[lotyrin]

And the best hackers at any level abstraction will always be the ones that actually understand what's going on in every lower layer in order to diagnose when the abstraction is failing them. Anyone that thinks you can be up at the level of vibes without understanding how an LLM thinks, without knowing how to review and factor your vibed Python or whatever high-ish level language, can make it performant without knowing if or when to write something in lower levels like C or need to be using a library where all the hard work is in something like C, make it secure without understanding how that gets turned into instructions for an incredibly obedient but ignorant machine (like the LLM is but in the exact opposite ways, buffer overruns and free before use and stuff)... It's a holistic practice. The guys that produce code and don't know which parts are happening in the browser or in the client, think they can trust the values of cookies not to be tampered with and junk are able to be productive, probably more so with LLMs these days but they simply can't make quality software and never will be able to. Corporations love them because nobody's accountable to resiliency (securty, quality, reliability) until something actually breaks and those guys can get thrown under busses easily when and if that happens because they're cheap cogs. Hackers love them because we'll always have work to do to improve (or compete with, or exploit) what folks like that make.

[reillychase]

I love it! This is what I was trying to say but you said it much better.

[henry2023]

The more software there is, the more maintenance and willingness to build more software will be.

On top of that, LLM output is so mediocre that even marketing firms are doing most “copy(s)” by hand.

[asdev]

sw engineering will be at an even higher premium if you've seen the code AI creates. AI will raise the bar to entry for sure though