The Golden Angle Meets Vibe Coding

Cademy xyz announced a free live webinar in mid-March 2026 on the McNeel blog: Fibonacci patterns in the Grasshopper workflow, inspired by the Bang & Olufsen Beosound A9 speaker. Technically elegant — phyllotaxis-based 2D point clouds transformed into 3D geometry with correctly calculated draft angles. Sounds like an afternoon of Grasshopper legwork. But the real question this raises is different: How would an architect today build such a workflow if they don’t memorize every component?

The answer emerging from the AI corner in 2025–2026 has a name: Vibe Coding. Andrej Karpathy — ex-Tesla AI, OpenAI founding member — coined the term: write code without fully understanding each step, by telling an LLM your intent and iterating until the result works. For studio scripts in Grasshopper or Python, that sounds tempting. But does it actually work?

←TODAY: Grasshopper users combine phyllotaxis scripts manually or copy Stack Overflow snippets — hours for geometry an LLM suggests in minutes.
→3012: In Zurich-3012, the atelier generates surface textures via intent-layer; the parametricist validates manufacturing tolerance, not the code.
Fulcrum: The value lies not in the generated code itself, but in knowing why the draft angle is 1.5° and not 3°.

What vibe coding delivers in the AEC workflow — and where it stops

The Cademy webinar (29 March, 17:00 CET, free registration via McNeel blog) illustrates what’s actually difficult about a Grasshopper script: not the Fibonacci formula (angle = n × 137.5°, radius = c × √n), but the decision chain that follows. How dense? Which cell geometry? Which wall thickness for CNC milling or injection molding? This is where vibe coding becomes either a productivity tool or a time sink.

Karpathy’s method works well in the AEC studio for three use cases:

• Boilerplate geometry: Point clouds, grids, simple transformations — an LLM (GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro) delivers working GhPython code on the first or second try.
• Documentation scripts: Rename plans, sort layers, dump Rhino data into Excel — repetitive, error-prone, ideal for an LLM-generated RhinoCommon one-liner.
• Rapid algorithm prototyping: Phyllotaxis distribution? Five minutes of prompting, then manual refinement. If you know the math, you can judge the result immediately.

What vibe coding doesn’t solve: manufacturing logic. The draft angle in the Beosound A9 workflow isn’t an aesthetic detail — it’s an injection-molding requirement. If you don’t tell the LLM you need 1.5° for an ABS part, you get vertical walls and a prototype that sticks in the tool. Responsibility for manufacturing knowledge stays with the human.

The McNeel community solved this pragmatically: discourse.mcneel.com and the Grasshopper forum have circulated phyllotaxis definitions (including from David Rutten himself) for years, serving as starting points. Vibe coding complements this community resource — it doesn’t replace it.

Atelier: In the PAZ context, we recommend the approach from the →3012 framework: tell the LLM first your intent («Fibonacci pattern, 450 points, golden angle, production-ready for CNC, min. wall thickness 2 mm»), then the constraints, then the output format. Iterate at most three times. If the code doesn’t hold the constraints after three iterations, switch to manual definition — that’s not failure, that’s process hygiene.

The honest assessment

The Beosound A9 example from Cademy xyz is well chosen: complex enough to expose traditional CAD limits (the webinar claim «parametric outperforms traditional CAD for complex texturing» is fair here), yet concrete enough to work through live in 90 minutes. Karpathy’s vibe-coding thesis fits this workflow — with the crucial caveat that phyllotaxis mathematics and manufacturing parameters are domain-specific knowledge no LLM invents from context.

Anyone using Grasshopper seriously builds a hybrid workflow in 2026: LLM for boilerplate and iteration, manual for constraints and validation. Anyone who thinks they can switch entirely to vibe coding gets pretty renderings and wrong draft angles. That’s expensive.

Sign up for the Cademy webinar on 29 March, watch the draft-angle step closely — and ask yourself afterward which part of it you’d trust to an LLM. That answer is your personal vibe-coding calibration point.

Source: McNeel Blog (Rhino)