I think there are tons of learnings to be shared about how to do this stuff, but it seems like it's all blocked behind arguments over whether AI is the best or worst thing ever, and penis-measuring contents about how to hold the tool. The net benefit is a very open question, and both the doom and gloom perspective and the AI booster perspective are valuable and have a lot of things right. But there's a dearth of information about what things work, what things don't, what happens in the process of using AI, how to adjust one's behavior and which of those adjustments is harmful even if effective.
But then, I'm part of the problem. I keep meaning to write up a series of experience reports, but it's a lot of work. More fun to vibe a new feature into existence, or to finally fix a UI wart...
If you have an LLM or human just start coding up something without nailing down those decisions first, then he/she/it will implicitly make those decisions arbitrarily in the moment (usually based more on pattern-matching than real weighing of alternatives) and the result will be a massive mess.
Ideally for an LLM, you'd hand-write a highly detailed DESIGN.md file to encode those decisions and a suite of test fixtures to enforce them.
If you explore Github, you're gonna see thousands of abandoned game engines, compilers for made-up languages. And that has been happening since before the LLM era.
I used to be part of an OS messaging board in the early 2000s and almost everyone had their own OS. A dozen people or so could even run Firefox! I remember (now legend) Terry bothering us to check out LoseThos or GodOS or whatever was its name, but quite a few people had OSs that could do more than that.
Not all software needs to be commercial to be useful, even if it's just for a learning experience. I have learned a lot from those experiments, even if they're not polished.
Yeah, but the people authoring those learned something.
> I used to be part of an OS messaging board in the early 2000s and almost everyone had their own OS.
Great example! I, too, once had my own toy OS, and browsed OSDev wiki nonstop. The thing is no one in the OS dev community were writing things they intended to place in front of actual users!
The difference now is that these incomplete projects:
a) Don't leave their owners any wiser than when they started, and
b) Are actually intended by their owners to be used by actual users.
> I have learned a lot from those experiments, even if they're not polished.
Would you have learned as much if you told a magic box "Make me an OS" and then slapped your name on it and uploaded it to GH?
Temple OS :)
This is fine so long as the author is learning something (questionable) and not polluting the commons with "I made this in a weekend" vibe slop.
Almost all of my prototypes are made this way. "I made this in a weekend", is more often, "I made this in a day". Like any trad-coded project, time to vibe code a backend vs frontend time is 1:N
This is exactly the same as demanding, that people who post their thoughts always post original, useful thoughts. It's just not going to happen. Making something easy will increase the total volume and the majority of that volume will be junk. It's inevitable. What is needed is a search engine for quality software. Perhaps LLMs can do that, since they are better at understanding concepts than generating them.
But it seems like we're finally starting to accept that "accidentals" like network effects, ownership, accountability, etc. are important. Of course, that's why many of us fled to technical corners in the first place - because the "accidentals" become tied up with things like nepotism, unfair and arbitrary judges from random humans who don't understand your merit, the need for bullshitting more than real technical value. Supposedly, anyways.
I never thought of this from this perspective but indeed it seems to be totally true.
I worked for a startup that stopped feature development and did a complete rewrite of a huge codebase. I was assigned to a side project during this time and missed the entire rewrite process. I came back to a completely rewritten codebase.
There was almost no learning curve, despite being in an entirely different language. The core architecture, data structures, and concepts were the same.
If you read the Bun blog post on how they did it, their rewrite was similar: The first step was getting it into a new language, not rearchitecting it from scratch.
I think they did it the right way based on my pre-LLM. Rewriting into a different language as fast and basically as possible is important for getting the team switched over quickly. Rewriting into a different language in X days is actually a good goal to minimize.
> If you read the Bun blog post on how they did it, their rewrite was similar: The first step was getting it into a new language, not rearchitecting it from scratch
They eventually might regret this, when they are trying to get rid of the last lines of unsafe code. Rust needs to be written differently before you can write safe things to perform fast when compared to other languages or unsafe code. It takes a lot of experience before you can see it.
If this were a major issue then you'd have to always nail your Rust architecture correctly from day 1 to handle future unknowns, and this isn't the case.
More importantly, LLMs are more than capable of figuring out how to rearchitect code and they have no problem making sweeping refactors for you, especially throwaway experimental ones that were way too expensive to do not long ago.
Short-termism at its peak. Code maintainers will learn the hard way how to set the boundaries between ai-generated code, and human maintainable code.
Also if you whip something up, you tend not to care for it as much as something you took the time to create in a "proper" manner.
What do you think of https://scramblequest.app and https://wordpeek.app ?
I have never looked at any of the code for them
edit: Yeesh, misread your comment as "word games" not "word clones" and was very confused about the claim. Probably should have noted my confusion and re-read.
Word is basically an operating system. Unix people like to make fun of Emacs for being one, but Word is basically one, too. And all the features in there are used, otherwise Microsoft wouldn't keep them around.
And .doc files are basically a memory dump.
In case of ccc (claudes C compiler) they just did not use any tests, though they are many existing C testsuite. They just stopped, when it finished compiling the kernel, still failing hundreds of tests.
Not really. Transpilers have existed since forever.
The hard part is all the edge cases. (And LLMs don't solve this problem; they probably akshually exacerbate it.)