Sharp et al. (2025) introduce "agentic inequality" as a framework for analyzing disparities in access to AI agents across availability, quality, and quantity dimensions. They propose the Context Access Divide (CAD) as a complementary dimension operating at the individual interaction level.

  • The CAD distinguishes between Dynamic Context Retrieval, where systems autonomously retrieve context, and Manual Attachment, where users must identify documents.
  • For knowledge-intensive workers, manual curation imposes cognitive burden that reproduces inefficiencies AI aims to eliminate.
  • A probabilistic model demonstrates that manual attachment leads to combinatorial collapse in task-success probability as corpus size grows.
  • Dynamic retrieval architectures are structurally insulated from this collapse compared to manual methods.

The authors formalize contextuality as a dimension of AI-mediated inequality and analyze its implications for knowledge-work stratification and AI platform governance.