The long-form arguments

A paper here is one whole argument with its evidence attached, for the reader who has to hand something to a colleague or a review board rather than forward a link a week.

Markers of an AI-Agent-Native Data Platform defines Autonomous Data Engineering Readiness — ADER: the set of capabilities a data platform has to provide before agents can be trusted to change production data, and a platform's completeness against that set, scored from 0 to 1.0. Ten markers, grouped under four pillars — Context, Control, Choice, Cost. Its finding is about the set rather than about any vendor. Mature platforms provide a strong subset, and two markers go unmet almost everywhere: deterministic quality gates bound to the agent's loop, and a single primitive that carries both build and operate. The paper names each marker, says why an agent needs it, and points to where it is done well today. It publishes no per-platform score and no ranking, and it is the only one of these papers that exists today.

The other two titles name what each would argue. Isolation for data work — what it takes to make being wrong survivable when the artifact is data rather than code, and what that costs on each platform. Re-engineering a legacy estate — what an uplift that reads out of Spark and writes into dbt and dlt involves in practice.