These independent studies ask two practical questions: How well does a design system stay aligned from design through production? Can an AI coding agent use it to produce working software?
GitHub’s Primer
Primer is strong, but it needs better proof that Figma, code, and product patterns stay aligned.
Evidence through
A strong foundation with incomplete proof that the system stays aligned at scale.
The agent produced working software in most runs, but one missed form behavior shows why automated tests and human review still matter.
IBM’s Carbon
Carbon is mature, but its strongest guidance is not yet backed by system-wide proof.
Evidence through
A strong system with mixed proof that design, code, product flows, and governance stay aligned at scale.
The agent produced working software in most runs, but two missed accessibility connections show why automated interaction and accessibility tests still matter.
AWS Cloudscape
Cloudscape has a mature public system, while a purposive agent benchmark found uneven compliance with its full answer, evidence, and executable requirements.
Evidence through
A mature public system with mixed proof that design, code, product flows, and governance stay aligned at scale.
The agent met every blocking benchmark requirement in 24 of 60 frozen-evidence runs. This purposive result is not a production reliability estimate.
U.S. Web Design System
USWDS is mature in code, while its design kit remains the main readiness gap.
Evidence through
A mixed result with strong code, guidance, components, layouts, and governance, but a dated design kit and no public machine-readable design-to-code mapping.
The agent passed every token, button, validated-field, and governance run, plus two of three complete-form runs. One incorrect workflow label blocked the remaining form.
The Prototype-to-Production Diagnostic identifies where design, code, product behavior, and team practices drift, then turns the findings into an achievable improvement plan.