A data engineering essay proposes that people define data semantics and promises to each consumer, while an engine merges or reorganizes pipelines rather than relying on manually fixed tables and task sequences. Incorrect joins involving historical customer identifiers can double-count revenue despite passing format checks, making approved business rules and independent verification on identical inputs essential. Changes are activated against specific versions and contracts, and people resolve ambiguous definitions. Related analysis likewise identifies agreement on shared organizational meaning, rather than the cost of generating computations with AI, as the bottleneck. This is a design proposal, not a finished product announcement.