AI Adoption in Analytics Engineering: A Dependency-Aware Framework for Context, Verifiability, Risk, and Progressive Autonomy

Abstract

AI Adoption in Analytics Engineering presents a dependency-aware engineering framework for deciding how much responsibility generative-AI use cases can safely carry in analytics-engineering environments. Rather than treating AI adoption as a sequence of organizational maturity stages, the framework focuses on the engineering conditions required for individual use cases to operate reliably. Fifteen recurring use cases are organized across four peer dependency surfaces: Context & Governance, Engineering Assistance, Analytical Intelligence, and Optimization & Operation. These surfaces are intentionally non-sequential and may be developed in parallel. The framework introduces AI context debt as a practitioner framing for how absent, stale, or implicit engineering knowledge becomes load-bearing when AI systems consume enterprise context. This framing is explicitly positioned against prior work on technical debt, ML technical debt, and tacit/explicit organizational knowledge rather than claiming those underlying concepts as new. The manuscript also identifies executable ground truth—including SQL, schemas, configuration, lineage, tests, and execution metadata—as an important source of independent assurance for AI-assisted analytics engineering. A decision framework combines use-case value, context readiness, verifiability, and consequence of error. The Understand → Recommend → Generate → Decide → Act continuum describes responsibility allocation between humans and AI systems; it is not proposed as a new universal autonomy taxonomy. Evidence boundary: This is a framework paper. The fifteen-use-case taxonomy and dependency surfaces are practitioner-derived and conceptual and have not been validated through a statistically powered multi-organization study. The proposed evaluation describes a future empirical validation approach rather than established causal evidence.