To be fair, the "beeps and lights a led" part is very important in case of a dead disk in an array. The only time, I disrupted a service in production for longer than a few hours was when I didn't get the email of a degraded raid 1 array and the second disk died a few weeks later.
I don't want to turn this into an, oh look at mine.
It's just always been very behind the curve. I don't think you should use anyone's honestly. You should create your own shell base, and point to skills you find might add value, and add them. Then as you learn things about the shape of how things work, what you end up doing is rewriting them to follow your rules from what you learned.
What you'll find is, you'll have significantly better skills and systems.
It's great for feature development. It queries you about you intentions, creates a spec for you to review and modify, creates a plan for it to follow, then farms out the work to subagents.
Crap on it all you want, but it makes LLM work predictable.
It does not make LLM work predictable. Try it on the same type of project twice and see. You think you're saving time but youd accomplish similar results with 2 or 3 skills.
I think this depends a lot on two things: how well-defined your requirements are up front, and how well-defined the problem space/domain are.
If there's some clarity and structure with clear expectations, it seems to produce fairly reliable, consistent results, and the process of validating specs with Claude can yield all kinds of unexpected dead-ends or holes in the idea you had. I find this very helpful.
But I'm curious too, what are the skills you'd use instead? superpowers are just skills themselves.
How is an implementation spec that helps it author a plan to follow not considered predictable? What are these skills you are referring to specifically?
I'll correct myself. 1 Skill for helping refine your plan/spec with propositional yes/no questions. After that no more skills needed to emulate superpowers honestly. Workflows do the rest and claude will do that for you. ultracode is 90% of superpowers
Three years? Vibe coding became a thing roughly 1.5 years ago, around Sonnet 4 release. How would you even compare models from 2-3 years ago to today’s? Is this satire?
> Vibe coding is a software development practice assisted by artificial intelligence (AI) where the software developer describes a project or task in a prompt to a large language model (LLM) which generates source code automatically.
Based on this definition, yep, I've been vibe coding since 2023. The products were less sophisticated, and I was copying and pasting one function at a time, but it worked. More important: it was something I couldn't do on my own.
The modern version is that bonafide engineers accept AI-generated code. It's a good thing I'm not a bonafide engineer.
That initially looked pretty interesting but a quick peek through issues and folks are complaining that it’s opinionated on git workflows and overriding user instructions otherwise. No thanks.
Edit: a deeper look at the issues and there are many examples of it not behaving as intended. Seems superstitious at best.
Definitely not superstitious.
It's just an opinionated workflow, you can see the steps further down in the repository.
It's just spottily enforced because it's just written to the context in markdown - it's basically one of the very first attempts from the very beginning of Claude code.
But superstitious would mean it's just in your head, essentially - but this has an effect. It does create questionairs, documents every decision and plan etc. Wherever you want that is up to you, I personally didn't... But it's also definitely very much causing an big difference in behavior from Claude Code
"It's still pretty obvious and still much worse than human prose"
You have no idea how many false positives and how many false negatives you have in your judgement. It is indeed impossible to differentiate between badly written human text and somewhat good written llm text.
> You have no idea how many false positives and how many false negatives you have in your judgement.
The default LLM style is pretty deterministic. "Not X, Not Y. Just Z", etc.
There are phrases and cadences which are very rare in human prose (<5%) but unusually common (+95% occurrences in LLM prose). It is not unreasonable to look at content which is 95% LLM tells and conclude that that an LLM authored it.
I have noted, IRL, that those people who read very little, and only read when they have to (work docs, etc) are literally unable to tell that a piece of prose sounds like an LLM even when it has about 12 occurrences of "Not X. Just Y" or "Not X, Not Y. Just Z" in as many paragraphs.
It's really similar to special effects in tv/movies: Whenever there's something new it's hard to tell, but as we get used to it and start seeing the patterns it becomes easier to tell. The older the special effects are the more obvious it is. Quite often you can just kind of tell if something was special effects or practical effects without being sure why. And there are always some really well done ones that slip past everyone, or odd lighting that makes it look fake (a thread a day or two ago on here about a legal case involving a photograph, some people in the comments thought it was a painting).
I can still detect AI-written prose with pretty high confidence (I average about 85% accuracy after a bunch of rounds while testing, though I'm rushing because I'm testing), but it has gotten harder. Opus 4.8, GLM 5.2, GPT 5.5, and DeepSeek v4 Pro, are able to fool me sometimes. Mistral writes so weird it almost trips me up, but then I go, "oh, that's just Mistral". Qwen sometimes accidentally switches to Chinese (filtered out of the game, just a thing I found in testing). I'm doubling the corpus, currently, with a focus on the frontier models that have been able to fool me more often, so it'll probably get a little harder. GPT 4o is disgusting, just horrible saccharine nonsense (and, I guess that's the model that's caused the most psychosis with its sycophancy).
And, I have to concede that the best models prose has gotten better, especially when they're prompted to write in a specific style, like literary fiction or an encyclopedic article about a specific topic. Still flowery and verbose, though, and never really getting to the point (the AI doesn't want anything, and it shows).
They didn't sued Thomann and they didn't show up. They sued some Chinese guy who didn't show up and used that default judgement against everyone including Thomann.
They don't seem to understand, that there is no way out of it except giving the guy all the Lego sets or 200k dollars and saying "we are very sorry" multiple times.
They're still trying to save their face and money and it might ruin the whole franchise.
I switched to MacOS a few months ago from sway and I really try to be as open as possible to the mac way of life, because I don't want to fight my OS. But, boy, mission control is unusable crap. I was really shocked how dumb everything around this feature was made. Things that were possible a few years ago, are not possible anymore. Like switching to desktops/workspaces by keyboard. Or the grid.
With the app "AltTab" I can at least switch between my apps without using the mouse and with raycast I can position windows, but it is painful how much slower switching and positioning things in MacOS is, than in any tiling window manager.
reply