1. Executive Summary

This Blueprint sets out a community-developed, technology-agnostic specification for the Data Director, a proposed agentic AI tool to support researchers in preparing, documenting and depositing research data for publication in alignment with the FAIR principles. It is the formal output of the Data Director Agentic AI Working Group, established through an RDA-Microsoft global community consultation on agentic AI in research, and represents version 1.0: a final community-developed technical specification rather than a finished product.

The case for the Data Director rests on six interconnected gaps in current research data practice. Four core gaps, relating to limitations in researcher capacity, inconsistent metadata standards, uneven access to data support, and uncertainty navigating funder, journal and institutional policies, are directly addressed by the tool’s functional specification. Two further gaps, concerning the quality and provenance of data and metadata, and the governance of sensitive, ethically or culturally significant data, including Indigenous data under the CARE principles, establish boundary conditions for the tool's behaviour. Two acknowledged gaps, around infrastructure sustainability and the recognition of data outputs in academic reward systems, fall outside the Data Director's scope but are documented for completeness.

This problem definition is translated into 14 architecture principles spanning FAIR and CARE alignment, openness, security, human oversight, traceability, interoperability, accessibility, sovereignty and environmental impact. These are operationalised through 12 functional requirements, covering repository recommendation, FAIR-aligned metadata generation, persistent identifier support, Data Management Plan verification, metadata remediation, provenance tracking, and discovery and linking, alongside 17 non-functional requirements defining the quality standards any implementation must meet. A six-phase process flow and vendor-agnostic reference architecture show how these come together in practice, from project planning through to monitoring and reuse of research data after publication.

With no reference implementation yet existing, the Blueprint sets out forward-looking evaluation criteria, implementation guidance for research institutions, national research infrastructures and commercial or non-profit vendors, and a governance model proposing the RDA as steward of the specification.

Finally, the Blueprint is transparent about what remains unresolved. Its open design questions and roadmap for future Blueprint iterations should be read not as gaps in the work, but rather as a sign that the eight-week consultation has surfaced genuine tensions and future considerations. The breadth of these questions reflects the depth of community engagement and offers a basis for ongoing discussion as the Blueprint is developed further.