AI content and forum conduct in an unmoderated space
Note added 23 June 2026: For the hardly enforced topic hygiene, the discussion about CAD Augumentation and Autonomy that originally began here, will restart to its own private thread: [link tba].
Current thread has drifted into a wi(L)der and polarizing discussion about AI-generated content, forum norms, and moderation. I have therefore retitled it to better reflect what the discussion became.
With no active moderation available, I have left the original post as written and changed only the title and this note, for clarity. The wider discussion can continue here on its own terms, without treating it as off-topic.
Apologies for the repeated edits. I am still learning the forum interface. Gladly, the removed/revised messages are still available in the history and should now make sense in the updated context without blaming off-topic.
Original topic title: “From Words to a Working Prototype: AI-Driven Design Environments”
There’s a kind of magic that turns up in stories from all over the world: someone says the right words, and a thing simply appears. A table lays itself with food. A word becomes light. A wish becomes something you can pick up and hold. For almost all of human history that stayed firmly in the realm of fairy tales. If you wanted to make something, you made it: patiently, by hand, learning the craft over years.
That is quietly changing. It is becoming possible to describe what you want in plain words and watch it take shape in front of you. This topic is for exploring the superpowers that machine learning and high-performance computing can bring to a creative person, inside whatever design environment they already like.
If I opened with the technical end of this (agents, skills, recursive self-improvement), it would read like English but land like noise: much of the vocabulary and semantics are new, or project-specific. So forgive a slightly longer introduction than usual.
Making has been “code-driven” for a while
A lot of what we already do in the space is driven by code, not by drawing. When you cut on the laser or print on the Bambu, the machine is not following your sketch. It is following G-code, a plain list of instructions: move here, at this speed, heat to that temperature, laser on, now off. It is the control layer underneath digital fabrication, and it has been there all along.
Parametric CAD took the same idea up a level. Tools like Fusion, or Rhino with Grasshopper (which a few members here have already explored: see the Grasshopper workshop and Rhino 3D & Grasshopper threads), let you describe a part with numbers and relationships instead of placing every line by hand. So “describe what you want, let a machine produce it” is not new. What is new is the language we are allowed to describe it in.
Talking to your tools in plain words
- Vibe coding (a phrase Andrej Karpathy popularised in 2025): building something, whether software, geometry, or a run of G-code, by describing what you want in natural language, letting an AI write the underlying code, and steering by the result rather than reading every line.
- MCP (Model Context Protocol): an open standard, a bit like a universal plug. Think USB-C, but for AI. It lets a language model reach out of the chat box and actually operate other software: call a CAD program’s commands, read a file, run a slicer, through one shared interface.
- Agentic codespaces (Claude Code, Codex, Antigravity, OpenClaw and others) are not chatbots but environments where the model can read and write files, run commands, generate scripts, do research in batches and work in a loop, pulling in serious cloud compute. Point one at almost any design app with a scripting interface (FreeCAD, Rhino, Fusion, Blender, Bambu Studio, LightBurn) and it can generate and modify geometry, G-code, firmware, documentation and standards, responding to your voice commands with your own voice if that’s what you want it to do.
These tools inject model inference, a dose of supercomputing we are already paying for in rising memory prices, and a degree of autonomy into design environments that were previously yours alone to drive by hand.
Is “vibe coding” real coding?
Does describing software in natural language make you a programmer? Linus Torvalds, who started Linux, recently pushed back on the hype. His point, roughly: people boast that “99% of my code is written by AI,” but 100% of all code has always been written by compilers. Tedious learning of syntax was the gatekeeper standing between many creative minds and their potential.
I find that clarifying rather than discouraging. If the machine takes the mechanical part, the question becomes: what is the craft that stays ours? I do not think there is a settled answer, and I would like to hear yours.
Where I have got to
- I started by wiring GPT-3.5 into Kali Linux through AutoGPT. That was a terrible idea.
- For the last few months I have been in OpenClaw, Antigravity, Claude Code and Codex, offloading the mechanical parts of the work onto the machine, where I think they belong, and keeping the judgement for myself.
- Along the way I have got AI talking to FreeCAD, Rhino, Fusion, Blender, Bambu Studio, LightBurn and others: generating and modifying geometry, G-code and firmware, scripting motion, controlling light, and the rest.
- I’m currently building an agentic Claude-Codex hybrid, a recursive self-improvement CAD “metaskill” system: tools that augment and, step by step, automate the design → manufacturing loop.
None of this replaces making. It moves the most tedious parts out of the way, leaving more room for the dopamine-rich ones.
Why I am posting
To gauge interest, and to find people already working this way. If there is appetite, I am happy to run regular, self-taught workshops in the space and to exchange experience with anyone down this road.
There is already interest here: @Christian has worked with parametric design in Rhino and Grasshopper; @jamesnwalker ran a workshop on generative modelling in Grasshopper; @nkmdk007 was after an easier way to design and visualise a product; @Brendon_Hatcher has been building computer-vision and AI tools on the Raspberry Pi. If that is you, or near it, I would value your take.
A few questions:
- Please chime in if you are already scripting or automating your design tools (skilling, workflowing, fan-out, MCP, all of it). What has worked, and what has burned you?
- Which design environment would you most want to be able to “talk to” in plain words?
- Where does this help, and where does hand skill stay irreplaceable?
- Is vibe coding real coding to you? Does it make someone a programmer?
- Would a hands-on workshop series be useful, and what would you want from the first session?
Pull up a chair. I can share progress as I go.
Max Craft
Phantom Quantum 2026
All Rights Reserved.
