Filth Assembly is alive! (I swear it's not AI slop)

It came to me in a dream.

Well, today’s module idea and hyperfocus to grind my website together at least. I was entering an old car dealership, which had made conciderable CapEx efforts to attract customers: The whole facade with massive concrete pillars were surrounded by atomic water streams, onto which spectacular holographic advertisements were projected.

Suddenly the whole extravaganza collapsed, water was coming down more like a tsunami than misting controlled streams of before. Whole place was flooding and true facade with dirty grey concrete revealed itself. We went inside never the less.

The owner was nice enough, but soon something went wrong with him and my friend’s ne fiance, which she doesn’t have in real life. They were leaving me out of something. Dream went recursive, almost nightmarish and begun to loop with ever so slightly worse vibes each iteration.

In reality I was listening to podcasts in my sleep like every night. Conversation between Lex Friedman and Stephen Wolfram begun to bleed into the quasi-nighmare. Great, metaphysical ponderings about Ruliad was just what this clusterfuck of a dream needed.

Wake up, I say, wake UP! Your feeble mind can’t handle this shit right now.

So I woke up, got up and took a leak. Dreams like that tend to linger into reality, so I watched as reality warped and glitched some in the toilet.

Ruliad. It’s Stephen Wolframs concept about a possibility space, in which every rule that can exist, exists. We, the people, observe a subset of a ruliad, that constructs our world, laws of physics etc. But the Ruliad in whole contains every rule that can be.

Somehow this nighttime metaphysical pondering led yet again to virtual eurorack modules.

What if there was a modulation source, in which there is 8 slots for rules. Or formulas, rather. X axis is time and Y axis is control voltage. $Y=X$ would produce linear rising phasor. $y = x²$ would produce exponential curve. Slots could be sequenced by triggering. They could be randomly populated out of greater rule pool, or Ruliad, by triggering.

Ruliad. God damn that’s a great module!

I get my glasses and present my idea to ChatGPT Astra. It promptly expands the idea with additional formulas and EOC (end of cycle) output. Great! Soon enough the idea is complete enough to prompt for a detailed design document. Chat knows what I want to build and how I want it. The document is one-shotted and perfect. I save it for further review.

I hate AI slop

Who doesn’t? But on the other hand I love my ChatGPT/Codex. I like coding as well, but I have rather severe case of ADHD, so most code projects die out with my dwindling interest before anything substancial is made.

I know, bad excuse, there are many developers with ADHD who make it just fine. But I have so much going on… Work at the ambulance factory, shoeing horses, other hobbies, family! I need to produce stuff fast, if ever want to see it completed.

So this is rather conflicting. I want to publish this stuff, because I really enjoy making it and I believe many people would find my modules useful. But I most certainly do not want to publish AI slop.

Codex, make me a Call of Duty but better, no mistakes, go.

Now that’s my definition of slop in coding context. Some would call my stuff slop, sure.

I mean I don’t code at all. It’s all vibe coded. Most of the algoritms as well, I’m pretty bad at math to be honest.

But! There is one thing I’m good at and I’m not giving that to AI ever. That’s panel design. I’ve been a professional web designer and graphic designer for so many years in my life, that I get vector-based graphic design and even UI stuff. I love it.

So here’s my workflow:

I get an idea, like the Ruliad one described above. I chat about it with Astra. Usually Astra can point out stuff that I forgot or didn’t even think about. Pretty soon the idea feels solid and I prompt for design document.

I’m still amazed how great it works. All is right there, every little detail written out just the way I like it. I don’t get it how it’s so good at this. I mean sure, it’s a frontier AI model but it seems to read my mind.

Many details are filled in that I didn’t ask for in prompt like behaviour, testing, compatibility with other modules etc.

OK so I got the design document. I save it as markdown in empty project folder. Above that is rather long AGENTS.md which describes the overall Filth Assembly design philosophy and practices, version numbering, naming conventions and so on. I prompt my local Codex agent in Omarchy using GPT-6.1-Sol default. Sol is crazy efficient with tokens, so hitting my session limit is very rare.

Codex works for some 12 minutes, and usually first version works just the way described in the design document. Some tweaks later I got first prototype like I wanted it, and I can move on to Inkscape and make new panels instead of the temporary ones.

Time for work

All this happened between 4-7 am. Oh and I installed Astro (with an o) locally as my to-be website framework. I love it. I set it up so it takes in markdown files and publishes them as blog posts or other pages. Did I mention how much I love md? It’s the best thing since CSS, man. I ordered Codex to implement my style that I made some weeks ago in this new Astro-powered website. Boom, first prompt, simple as that. How is this so simple and just works? There’s got to be a catch XD

Well, it’s 7:20 and I must get to work and wrap some vinyl to ambulances, see you next time!