Skip to content

Data Director Agentic AI Tool Blueprint

[!NOTE] About this edition. This is a Markdown edition of the Data Director Agentic AI Tool Blueprint, version 1.0, an RDA Recommendation published at https://doi.org/10.15497/RDA00157 (Zenodo record 22275620, file SEPTEMBER2026_v1.0_ENDORSED_DataDirectorAgenticAIBlueprint.pdf, MD5 38eaa6a999ee6a911e9f155f634b3889). The Blueprint is the work of the Data Director Agentic AI Blueprint Working Group and is licensed under CC BY 4.0. The published PDF is authoritative.

The text is reproduced verbatim. This project made the following changes:

  • The document is split into one file per chapter and appendix. Appendix headings are raised one level so that each file opens with its own title.
  • The seven figures are redrawn as Mermaid diagrams from the originals. Their legend and page footers are left out; node shapes and colours follow the legend: rounded green boxes are entry points, the pale yellow band is the phase header, diamonds are decisions, dashed boxes are optional steps, R-numbered notes are functional requirements and Guidance note marks guidance.
  • Tables are set as Markdown tables. Text that was on several lines within a cell is separated by line breaks, and a cell's bullet points are marked with "•".
  • Page numbers, running headers and footers, and the page numbers in the table of contents are removed; a table of contents is generated at the end of this page.
  • Cross-references to sections, tables, figures and appendices are links, and each requirement identifier in Section 7 has an anchor (for example 07-functional-specification.md#r3).

These files are generated by scripts/blueprint/convert.py; do not edit them by hand.

RDA Data Director Agentic AI Blueprint WG Recommendation: Data Director Agentic AI Tool Blueprint

RDA Recommendation

DOI: 10.15497/RDA00157

Authors: Data Director Agentic AI Blueprint WG

Published: 3 September 2026

Version: 1

Abstract: The Data Director Agentic AI Tool Blueprint v1.0 is a community-developed, technology-agnostic specification for an agentic AI tool to support researchers in preparing and depositing research data for publication in alignment with the FAIR principles. Produced by the Data Director Agentic AI Working Group, an RDA-Microsoft collaboration, it defines 12 functional requirements, 17 non-functional requirements, and 14 architecture principles, alongside a six-phase process flow, a reference architecture, implementation guidance, and evaluation criteria. The Blueprint is version 1.0, final version for endorsement by the RDA Council, and is transparent about what remains unresolved, offering a basis for ongoing community discussion as the specification develops.

A research software engineer at a large academic computing centre is engaging with the Blueprint in two ways. First, through their own research on research software metadata, where the Blueprint’s grounding of quality checks in existing validators, and its recognition that schema compliance alone does not confirm semantic correctness, maps onto ongoing work on cross-record metadata consistency; the Blueprint’s framing will be applied and findings reported back. Second, through dissemination within their institution, introducing the Blueprint to research software engineering and data support colleagues and gathering informal feedback. This is a personal commitment and does not commit the institution, nor extend to deploying a Data Director instance.

Impact: The Data Director Agentic AI Tool Blueprint addresses a systemic gap in research data practice: the uneven access to skilled data support that determines whether research data is prepared, documented, and deposited in a way that makes it findable, accessible, interoperable, and reusable. By defining an open, community-governed specification for an agentic AI tool, the Blueprint creates a foundation for agentic AI tool implementations that could narrow this gap across institutions, disciplines, and geographies, supporting the long-term goals of open science and making high-quality research data a realistic outcome for any researcher, regardless of their institution or expertise.

Contribution to United Nations SDGs:

Industry, Innovation and Infrastructure: The Blueprint provides an open, community-governed specification for agentic AI tool implementations that strengthen the infrastructure through which research data is prepared, deposited, and made reusable across institutions and disciplines.

Reduced Inequality: By addressing systemic gaps in researcher capacity, metadata standards, access to data support, and the complexity of navigating funder, journal, and institutional requirements, the Blueprint aims to support equitable access to high-quality data support tools regardless of institution, discipline, or geography.

Responsible Consumption and Production: The Blueprint promotes the responsible stewardship of research data through a functional specification covering metadata quality, provenance, persistent identification, and the maximisation of value from publicly funded research outputs.

Language: English (change if necessary)\ License: Attribution 4.0 International (CC BY 4.0) (change if necessary)

RDA webpage: https://www.rd-alliance.org/groups/data-director-agentic-ai-blueprint/outputs/data-director-agentic-ai-tool-blueprint/

Data Director Agentic AI Tool Blueprint, Version 1.0

Initiative: RDA-Microsoft Global Agentic AI Initiative

Status: Final Version for RDA Council Endorsement

Version: 1.0

Date: August 2026

Programme Lead: Connie Clare (RDA)

Programme Facilitators: Connie Clare (RDA), George Earl (Microsoft), Trish Radotic (RDA), Benjamin Wright-Jones (Microsoft)

Blueprint Authors: See Appendix C: Acknowledgements

This document is a community-developed technical specification for the Data Director as an open, technology-agnostic agentic AI tool. It is not a finished product or piece of software. It is freely accessible, transparently developed and governed in line with RDA Guiding Principles.1 It defines what the Data Director should do and how it should behave, without prescribing any specific AI model, cloud provider or infrastructure stack.

Document Control

Version History

Version Date Change Summary
0.1 29 June 2026 Initial draft for community review
1.0 5 August 2026 New section Appendix G 'Community Review Commentary Log’ added for provenance to keep track of the community review suggestions made and by whom, so that these can be easily discussed and implemented should v2.0 be developed. Final version 1.0 for RDA Council endorsement.

Contents

  1. https://www.rd-alliance.org/about/ ↩