Figma's Task-Shape Routing for Model Selection
TRIGGER
Using a single default model for all design tasks led to suboptimal results—some tasks needed fast structured execution while others benefited from reasoning or example-following, but without explicit model switching, users couldn't match model strengths to task requirements.
APPROACH
Figma Make exposes model selection (Claude Sonnet 4.5 as default, Gemini 3 as alternative) and provides guidance on when to use each. Claude: layered structure prompts, nuanced context, architectural TC-EBC prompts. Gemini: narrow precise tasks (rename layers, swap icons, tune layout), tight constraints, fast execution. GPT: reasoning with examples, exploratory reframing, 'show don't tell' demonstrations. Users explicitly choose models based on task shape rather than using a single default.
PATTERN
“Hit-or-miss results using one model for everything. Claude excels at structured frameworks, Gemini at narrow precise edits, GPT at example-based reasoning. Match model to task shape instead of using a single default.”
✓ WORKS WHEN
- Platform supports multiple models and explicit switching between them
- Tasks vary enough in shape that different models provide meaningfully different results
- Users can recognize task patterns that map to model strengths (narrow vs. exploratory, structured vs. example-based)
- Switching cost is low enough that per-task model selection is practical
✗ FAILS WHEN
- Single model handles all task types adequately—switching adds friction without benefit
- Users lack mental model for which tasks suit which models
- Model differences are subtle enough that selection feels arbitrary
- Workflow requires continuity across tasks where model switching would break context