Is AI Taking Over the Visual-Portfolio Side of Three.js?
A real question on Three.js Discourse last week: "Is AI taking over the visual portfolio side of three.js? What domains still need real 3D engineering?" The question is honest — every Three.js dev is watching AI tools eat parts of their job and wondering which parts are next.
Working answer below.
# TL;DR
- AI eats: portfolio-grade 3D scenes, marketing-page hero assets, simple product showcases, prototype levels.
- AI doesn't eat (yet): virtual tours with custom interactions, medical visualization, architectural walkthroughs at BIM scale, simulations, custom shader work, anything with hard perf budgets.
- AI augments: most other Three.js work — designers handle layout via AI, devs handle the unique 30%.
# What AI eats
Marketing-page 3D heroes. Product configurators (with caveats). Portfolio scenes. Simple animation. Small game-asset blockout. Generic 3D content that "looks 3D" without doing anything custom.
If your Three.js work is "client wants a 3D box on their landing page" — that's now a 90-second Spline or Yugma prompt away from done. The hand-coded version doesn't justify its cost.
# What AI doesn't eat
Three categories survive comfortably:
1. Custom interactions and shaders
If the project requires custom GLSL, multi-pass post-processing, audio-reactive geometry, or unusual gesture-driven cameras, AI tools don't go there. R3F + custom shaders is still hand-coded.
2. Performance-bound 3D
Mobile WebGL ceilings, Quest 3 framerate budgets, instanced 5,000-mesh scenes, custom culling. Generative tools don't optimize for hard ceilings.
3. Domain-specific 3D
- Medical visualization — DICOM volume rendering, segmentation, annotations.
- Architectural at BIM scale — Revit / IFC interop, dimensioned plans, code compliance.
- Simulations — physics-based rendering for engineering decisions.
- Cultural heritage — photogrammetric 3D with metadata-rich annotations.
These need expertise the AI doesn't have access to in its training distribution.
# What AI augments
The middle 50% of Three.js projects: designers handle layout via tools like Yugma, devs handle the unique 30% that AI can't compose. The split varies by project — sometimes 70/30, sometimes 30/70.
In the project workflows we see at Yugma, the designer drafts the scene in 90 seconds, exports GLB, hands to a developer who layers shaders / interactions / engine integration. Total: still half the time of fully hand-coded, with a cleaner first draft.
# The career take
If you specialize in "make a 3D website" and stop there, AI eats the floor under you. If you specialize in custom interactions, perf-bound work, or domain-specific 3D, you're fine — and probably have more work, because AI tools generate scenes that need devs to extend them.
The sharpest pattern: get fluent with AI scene tools so you can deliver in days what used to take weeks, then specialize in the unique 30% that AI doesn't touch.
# What we built Yugma for
Yugma exists to generate the 70% — the layout, the placement, the materials, the lighting. We assume the dev or designer handles the 30% that AI can't yet do. The tool is honest about that division.
Read the Yugma vs Three.js comparison → Read the where-3D-still-matters editorial →