Comment by poisonborz
5 days ago
This would have been a very smart and useful article up until 3 years ago. Now with AI doing this work is a minute, with most if not all considerations baked in, if some strange quirk would need it. Frameworks and dependencies made and replaced by custom "ground-up" creations is now a plausible reality. Not necessarily useful but doable and most importantly, testable in a fraction of time. We should readjust our sensibilities to that.
By default AI doesn't bake in all considerations. By nature of how it works it behaves like a human, i. e. making similar mistakes and oversights... I feel like this could somewhat ironically be shown with exactly this task. Let's let it make button and see how many ways it gets it wrong.
> Now with AI doing this work is a minute ... Frameworks and dependencies made and replaced by custom "ground-up" creations is now a plausible reality.
"You only have to want it and believe in it, then it will succeed."
- Count Ferdinand von Zeppelin
"The airship [is] the future of commercial air transport in general."
- Dr. Hugo Eckener
"The age of the airship is no longer a dream of the future; it is a reality of the present. Giant dirigibles will soon cross every ocean, making obsolete the slow surface ships of today."
- Popular Mechanics, December 1929
There were hundreds of such misplaced tech hopes, and hundreds that did become true - all those "visionary" quotes of newspapers talking about video calling and remote libraries. Would have anyone advocated a Rust rewrite of Bun before? The community would have frowned upon the idea, saying the effor would never be worth it. And now here we are.
I'm not saying use custom html elements because we can. But if somehow a html button is a valid technical problem, it can be reasonably done well for the scope of that app.
> Would have anyone advocated a Rust rewrite of Bun before?
That’s hardly a “tech hope” or “visionary”. Dirigibles continue to be interesting and fascinating to new generations, even if just as a concept; the language used to write a JavaScript runtime is uninteresting and unimportant in the scheme of things.
> The community would have frowned upon the idea, saying the effor would never be worth it. And now here we are.
Here we are where? Is the Rust release out yet? Didn’t it have tens of thousands of `unsafe` blocks? What’s so great about it? At this point in time, even implying the transition was a success is the same as calling the Hindenburg or the Titanic a success. You’ve seen the thing exists but have no idea how it’ll perform in the real world under real scrutiny and real scenarios. Perhaps that wouldn’t matter had much of Bun was just an internal project used by just one company, but it isn’t.
There’s an interesting economic point to be made about the specialization of labor and the benefits that arise from that.
Just like it is more efficient to have a food system than to have everyone feed themselves from their backyard (if they have one), maybe someday people will realize that it will be more efficient to build things once and re-use.
Similarly, every argument for “AI makes it cheaper so we can do it now” falls apart under “AI also makes it cheaper to not do it”.
What do you mean with testable in a fraction of time?
Generated tests can help, and if we go into that direction we can now certainly afford to introduce more proven code (Lean/Roq/Frama-C…), but that will still not be wild reality proof until it faces the whole user base and their widely different environments.
And large load of code is still large load of code.
Abstraction and reusability is good even when copying is getting cheap.
In a same vain, why create functions or classes, when AI can rewrite the code in place and keep track of rules.
It hasn't even been a few weeks since I had AI use a button + onClick handler for navigation instead of a simple anchor tag...