Figma's TC-EBC Framework for Reducing Prompt Iteration Cycles
TRIGGER
LLM outputs for design tasks were unpredictable and required extensive iteration because natural language prompts introduced ambiguity—models spent cycles interpreting vague instructions like 'keep it simple but exciting' or 'maybe like how Apple does it' instead of executing clear intent.
APPROACH
Figma's design advocacy team developed TC-EBC (Task-Context-Elements-Behavior-Constraints), a structured prompt framework for Figma Make. Input: decomposed prompt with explicit sections—Task (what to build), Context (why and for whom), Elements (specific UI components), Behavior (user flows and interactions), Constraints (platform, accessibility, profiles). Output: more complete first-pass prototypes with richer visual design and better scaffolding. The team also created a GPT assistant trained on TC-EBC to help users restructure vague prompts into the framework format.
PATTERN
“Ambiguity costs tokens twice: once guessing intent, again iterating to fix. Structured templates (Task-Context-Elements-Behavior-Constraints) force disambiguation upfront, producing complete first-pass outputs instead of multiple regeneration cycles.”
✓ WORKS WHEN
- Task has well-defined output structure (UI prototypes, feature descriptions, code scaffolds) that can be specified in advance
- You're generating the same category of output repeatedly and can reuse template structure
- Constraints are knowable before generation (platform, tone, length limits, component requirements)
- Model supports long-form structured input without degrading output quality
- Team can invest time learning the framework—payoff comes from repeated use, not one-off prompts
✗ FAILS WHEN
- Exploratory tasks where constraints should emerge from generation (brainstorming, open-ended research)
- Constraints aren't known upfront and need discovery through iteration
- One-off prompts where framework overhead exceeds iteration cost
- Tasks requiring model creativity that rigid structure would suppress
- Prompt length already near context limits—adding structure sections reduces space for actual content