AI-generated prototypes often produce inconsistent results due to undocumented design decisions and hard-coded values in design systems. A practical approach from Atlassian's Hardik Pandya outlines three layers for making design systems AI-ready: structured Markdown spec files containing spacing rules, color choices, and component guidelines; a token layer that constrains AI to named variables; and an audit script (e.g., FigmaLint) that flags hard-coded values and keeps specs in sync with updates. The core principle is treating design decisions as infrastructure — every decision must be documented and fed into spec files that AI reads before generating prototypes. Real-world examples from Atlassian, IBM Carbon, and others illustrate the approach.
