Meta releases Llama 3.2 with 'Prompt-as-Code' for reproducible design prompts

AI · 5 min read

Llama 3.2, Meta's latest model release, ships with a built-in Prompt-as-Code system that records prompt construction, transformations, and post-processing steps as structured, versionable artifacts. This helps teams track how specific prompts created design outputs and roll back or iterate reliably.

The Prompt-as-Code format integrates with common design tooling via plugins that attach prompts to Figma frames, component libraries, or asset entries. Designers can replay or tweak prompt pipelines and compare outputs across model versions, enabling scientific A/B testing of generative recipes.

Meta also published a repository of community-shared prompt templates optimized for UI elements, iconography, and branding. While privacy and data governance questions remain for enterprises, the company stressed that Prompt-as-Code files can be stored on-premises or encrypted in cloud repositories.