Cooking with Constraints: A Designer’s Framework for Better AI Prompts | Figma Blog (opens in new tab)

Design and cooking both depend on preparation: clear inputs and intentional constraints lead to better outcomes. The article argues that AI models do not need politeness or emotional framing; they need precise instructions that reduce ambiguity. For product designers, structured prompting bridges the gap between probabilistic AI outputs and the repeatable, purposeful results design requires.

Prompting as Mise en Place

• “Mise en place,” or “everything in its place,” means preparing ingredients before cooking—and serves as a useful model for preparing AI prompts.
• Effective prompts should establish:
  • Clarity
  • Context
  • Constraints
• The author’s framework is TC-EBC:
  • Task: What should be built or accomplished?
  • Context: Who is it for and why?
  • Elements: Which features or components are required?
  • Behavior: How should the system respond to user actions?
  • Constraints: What technical, platform, accessibility, or product limits apply?
• This approach aligns with broader prompt-engineering guidance emphasizing defined intent, modular construction, and predictable results.

Why Vague Prompts Underperform

• A request such as “build an app that uses pantry photos to suggest recipes” leaves too many decisions to the model.
• Polite language and conversational phrasing can bury the actual task without adding useful information.
• The resulting prototype may include basic functionality but remain visually generic, uninteresting, and barely beyond a wireframe.

Applying TC-EBC to a Design Prompt

For a pantry-based meal suggestion app, the structured prompt specifies:

• Task: Build an AI-powered meal suggestion app using pantry and refrigerator photos.
• Context: Create a home-cooking assistant for households with dietary restrictions.
• Elements: Include camera input, pantry scanning, dietary settings, meal suggestions, and recipe cards.
• Behavior: Let users upload photos, scan inventory, apply dietary preferences, and receive recipes.
• Constraints: Make the experience mobile-first, support iOS and Android, provide accessible UI, and allow multiple household profiles.

This structure makes the request easier to scan and gives the model explicit guidance about the app’s purpose, interface, behavior, and limitations.

Design Requires Structured Uncertainty

• LLMs are stochastic, meaning their outputs are probabilistic and variable.
• Design, by contrast, depends on precision, consistency, and intentional decisions.
• Structured prompts help “collapse uncertainty into structure,” much as a design system provides reusable rules and guidance.
• The article presents the TC-EBC prompt as producing a substantially more purposeful prototype than the original one-shot request.

A practical recommendation is to treat prompting like preparation for a complex recipe: define the task, provide relevant context, list required parts and behaviors, and state constraints before asking the AI to generate a design.