Design system adoption failures predate AI and stem from recurring patterns: engineers preferring to build components themselves, cold-start friction for rotating engineers, contribution gaps, and fragmented cultures from acquisitions. AI amplifies these existing problems by accelerating drift, inheriting the contribution problem at the skill-file layer, and triggering replacement anxiety. Practical approaches that have worked include lowering contribution activation energy with explicit documentation and guided first contributions, snapshot publishing to private registries for safe experimentation, moving design system engineers into the dev org for credibility, and using a custom MCP as a centralized non-negotiable source of truth paired with skill files for platform-specific deviations. Adoption reports scanning codebases can surface disengaged teams and start productive conversations. Ultimately, adoption is a culture problem requiring ongoing human conversation — tooling makes those conversations easier but cannot replace them.
