2. Introduction¶

2.1 About the Working Group¶

The Data Director Blueprint has been produced by the Data Director Agentic AI Working Group,¹ established following a global community consultation on agentic AI in research² conducted by the Research Data Alliance (RDA)³ in collaboration with Microsoft⁴ in November 2025. That consultation built on prior engagement Microsoft undertook with around 50 research institutions across the UK, which had identified data preparation and FAIR-aligned data sharing as a significant and widely felt pain point across the research lifecycle. The consultation explored current use of agentic AI by researchers and sought perspectives on its value across the research lifecycle, with agentic AI defined as 'artificial intelligence systems capable of autonomous operation with minimal human oversight'.

Three tools emerged as clear community priorities: the Literature Librarian, the Data Director, and the Funding Finder. Of these, the Data Director, designed to support research data preparation and sharing in alignment with the FAIR principles,⁵ was identified as the tool most closely aligned with the RDA's mission, vision, and guiding principles,⁶ placing data sharing, interoperability, and FAIR-aligned data management practices at its core. It was therefore selected as the focus for this Blueprint.

The Working Group operated over eight weeks through a community-driven process comprising two facilitated sessions per week, covering global time zones and open to all. The facilitation was supported by Microsoft within the framework of the RDA public-private partnerships.⁷

2.2 Purpose and Scope¶

This Blueprint is a written functional and non-functional specification that defines what a Data Director agentic AI tool should do and how it should work. The development of any deployable implementation or prototype solution is out of scope for this phase of work and may be explored by the community in future. The Blueprint constitutes the Working Group's formal output under the RDA Recommendations and Outputs process.⁸

The Blueprint covers the following areas:

The background and context for research data publication in alignment with the FAIR principles
The problem definition and case for a Data Director agentic AI tool
The architectural and design principles that should govern the tool
A full functional specification, including a process flow and a high-level reference architecture diagram
Evaluation criteria and implementation guidance
Blueprint governance and open questions

The Blueprint is intended to be sufficiently detailed for a range of diverse implementers: research institutions, national research infrastructures, and commercial or non-profit vendors, to use as the basis for implementation. Existing tools and platforms may overlap with some capabilities described here. The Blueprint does not seek to replace or endorse any specific solution, but to provide a common, community-agreed specification against which tools can be developed or evaluated.

2.3 What is an Open, Technology-Agnostic Blueprint?¶

This Blueprint is a community-developed, openly governed specification, not a product or piece of software. It is designed to be implemented by any organisation regardless of their technology stack, and is defined by three principles:

Open: Freely accessible, transparently developed, and community-governed in line with RDA's Guiding Principles and general principles of open science.
Technology-agnostic: Defines what the tool should do and how it should behave, without prescribing which AI model, cloud provider, infrastructure, or data repository to use.
Living and versioned: A specification designed to evolve through community review as needs, AI capabilities, and the broader research infrastructure landscape evolve. Version 1.0 represents the first community-endorsed version.

2.4 The Data Director: Tool Definition¶

The Data Director is a proposed, technology-agnostic, human-supervised agentic AI tool that supports research data documentation, publication, sharing, and reuse. It does so by helping researchers create ethically and scientifically responsible metadata and data documentation, identify appropriate repositories, metadata standards and persistent identifier options, and meet relevant funder, institutional, disciplinary, and governance requirements.

The original consultation (November 2025) defined agentic AI as operating with 'minimal' human oversight. Over eight weeks of community-driven development, the Working Group's model of human involvement, continuous approval for all consequential actions (Section 6, P4), moved well beyond that starting point, reflecting the considered judgement of the Working Group rather than an inconsistency, and this Blueprint's human-supervised model should be read as superseding the earlier framing.

As described in Section 2.1, the Data Director was identified as a community priority through the RDA-Microsoft global consultation and selected as the focus for this Blueprint based on its close alignment with the RDA's mission and guiding principles.

Community engagement identified three core capabilities as the starting point for the tool, listed here in the order in which they arise across the data publication workflow:

Recommending appropriate data repositories in line with funder, journal and institutional requirements
Automatically suggesting metadata in alignment with the FAIR principles
Supporting the assignment of persistent identifiers (PIDs) for datasets

These capabilities represent the community-validated foundation from which the Working Group's programme of work began. The Working Group developed them into the detailed functional specification set out in this Blueprint, expanding beyond the initial three capabilities to encompass a broader range of uses, functional requirements, and goals identified through the programme.

2.5 Intended Users¶

The Data Director is designed primarily for researchers, from early-career researchers to principal investigators, who need to prepare and share their data in alignment with the FAIR principles across the research lifecycle, but may lack the time, training, or institutional support to do so.

Data support professionals are a secondary but important user group, both as direct users of the tool and as those who may configure, deploy, or provide support for it on behalf of their institution.

While institutions, journals, and funding organisations are not direct users of the tool, they are key stakeholders whose requirements, particularly around data repository selection and policy requirements, shape what the tool recommends.