What we need is more than just OSS templates. We need fully LLM-managed "software libraries". These would be like package managers on steroids. They would be a huge hierarchy of LLM-coded modules. When you ask for an app, your LLM assembles it from the modules that already exist. If you ask for a novel feature, it codes it up and places it into the library.
So, for your VST player example. You would ask for a barebones version with just the features you need, and your LLM would assemble it for you from preexisting library modules. Then you ask for sound font support. If there is an existing sound font module, it would download it and add it into your app. If there wasn't, it would create one and put it into the library, so the next time someone asks, they can quickly get a VST player with sound font support.
I’m an engineer but not a software engineer and it’s been pretty interesting trying out vibe coding and seeing the barriers of entry being blown apart.
I’ve been progressing at learning the piano quite a lot faster than I expected with this. I’ve been debating open sourcing it but I’m not sure I want to take on the work of managing it
However, it seems to me like there must be some middle ground, between sharing architecture "ideas" and full on code PRs.
I myself maintain various FOSS projects; and incoming PRs are marginally useful. generally i just close, and copy with claude following my own & project taste (crediting initial contributor).
I've seen this idea: https://v-it.org/
Hate to say it but this is genuinely just a skill issue.
Resorting to building a DAW tailored to your own lack of understanding because you can't be bothered to learn tools that are extremely widely used and well-documented is the most extreme case of yak-shaving I've ever seen.
These has been this ladder of copium that "oh human still has input here for X, oh shoot it doesn't with Claude 4.5, well human is still needed for Y, oh shoot...".
You are suggesting that you can provide context that the model can't on its own. This is a search problem, every iteration it gets better.
We are at the stage in our company where "/goal optimize this" gets better result than any biased human context engineering can provide.
How exactly do you propose that the model should search my thoughts? Of course I can provide context that the model doesn't have. I can provide context that does not exist anywhere on the Internet (unless I got hacked).
> We are at the stage in our company where "/goal optimize this" gets better result than any biased human context engineering can provide.
First, a "this" must exist to optimize. Second, you have to be happy enough with it that optimization is the only problem.
I fully agree with them that the easiest solution is to ask an agent for exactly what you want, and iterate from there. Evaluating and installing and learning off-the-shelf software is often more of a pain than doing it from scratch.
And the same library is built without the "seed" or baseline?
1) If the UX/UI is important for you, the agent can have access to your other work, or to your other conversation and build your tool with that in mind. This is just another search space question. If you have a documented, predictable taste, the models can efficiently produce what you need.
My uni supervisor/professor used to complain about my slop papers, then I started generating them in his writing style by just pulling down all of his papers and blogs, and he doesn't mind it now.
2) In general, the LLMs are great for steering opinion. There is good research on steering political opinion with LLMs. Your taste and opinions can be formed by the LLMs to like the solution that it provides you.