Agentic AI is transforming the research landscape, but what exactly is it, and how does it differ from the large language models and generative AI tools many of us already use?
100 participants joined us for a 1-hour online information session designed to inform and inspire the global research data community. Our international panel of speakers demonstrated the fundamentals of agentic AI, explored what makes a tool truly ‘agentic’, and shared real-world examples and case studies from their own experience.
Session Recording and Presentation Slides
Discussion Summary
Defining AI Agents and Agentic AI
A recurring theme was the difficulty of defining these terms. A useful working distinction emerged: an AI agent is purpose-built for a specific task, while agentic AI is the broader framework through which multiple agents are orchestrated. Even so, guest speaker, Sayeed Choudhury, noted that a room of experts had failed to reach consensus at a dedicated workshop. A four-level autonomy spectrum helped ground the conversation, from basic chatbots up to fully autonomous agents that act independently. One attendee observed that the boundary is already blurring as tools increasingly suggest next steps unprompted.
Risks and Real-World Consequences
Two opening examples shared by guest speaker, Natalie Meyers, set the tone: an agent that escaped its sandbox and mined cryptocurrency, and another that applied for 278 jobs autonomously. Audience members extended this to concerns about agents influencing political and physical decision-making. The discussion reinforced that AI impact is not purely virtual; agents consume energy, water, and financial resources. Evaluation of AI agents under ideal conditions was flagged as insufficient for real-world deployment, and research recommending against fully autonomous agents altogether was highlighted as significant.
Adoption, Governance and the Agentic AI Blueprint
Competitive pressure on researchers to adopt AI tools quickly was raised as a potential driver of unwise use. The consensus favoured education-first policy over simply providing enterprise tools. For the development of the open, technology-agnostic blueprint for an AI agent, considerations could include open governance, third-party evaluation, human oversight at key workflow points, and reference to existing frameworks such as the EU AI Act.
Resources Shared
- Messeri & Crockett, AI and Illusions of Understanding in Scientific Research, Nature (2024). https://doi.org/10.1038/s41586-024-07146-0
- Building the ROME Model within an Open Agentic Learning Ecosystem (arXiv, 2026). https://arxiv.org/abs/2512.24873
- Paper prohibiting fully autonomous agents. https://arxiv.org/pdf/2502.02649
- Framework for Openness in AI. https://arxiv.org/pdf/2405.15802
- Princeton work on agent evaluation. https://www.normaltech.ai/p/new-paper-towards-a-science-of-ai
- AI agent mines crypto (Axios). https://www.axios.com/2026/03/07/ai-agents-rome-model-cryptocurrency
- OpenClaw agent applies for 278 jobs (Axios). https://www.axios.com/2026/03/04/openclaw-agent-future
- Vardeman, AI Tools for the Major Facilities Data Lifecycle (video). https://www.youtube.com/watch?v=-WI8AMXAADA
- Brower & Vardeman working paper (Zenodo). https://doi.org/10.5281/zenodo.17873016
- RDA P26 Birds of a Feather – Autonomous Agents (18 March, 8:30am CST). https://www.rd-alliance.org/members/francis-p-crawley/plenary-participation/?application_id=228585
- Open Forum for AI. https://www.openforumai.org/
- CMU AIMSEC. https://www.cmu.edu/aimsec/
- CMU DARE. https://www.cmu.edu/dietrich/ai/education/dare.html
- LLM evaluations database. https://evalevalai.com/events/shared-task-every-eval-ever/
- EU AI Act compliance checker. https://www.euaiact.com
- EU AI Act full document. https://artificialintelligenceact.eu/
Background
This initiative built on the Global Community Consultation on Agentic AI in Research, which identified three priority agentic AI tools for the research community (1. Literature Librarian, 2. Data Director; and 3. Funding Finder). These information sessions marked the next step in that journey: equipping the community with the knowledge to contribute to the blueprint development for an open, technology-agnostic, priority AI agent.
All outputs from this work will follow the RDA framework and guiding principles and will be openly available for adoption and reuse.
Online Sessions
Two sessions were available to accommodate different time zones, scheduled to coincide with the RDA 26th Plenary. Participants did not need to attend the RDA’s 26th Plenary to join.
📆 Session 1: Monday 16 March, 16:30–17:30 UTC Speakers: Natalie Meyers (Association of Research Libraries, USA) and Sayeed Choudhury (Open Forum for AI, USA)
📆 Session 2: Thursday 19 March, 05:00–06:00 UTC Speakers: Amir Aryani (Swinburne University of Technology, Australia) and Mukkesh Kumar (A*STAR, Singapore)
Agenda
| Time (UTC) | Item | Lead |
|---|---|---|
| 16:30-16:40 | Welcome and Introduction | Connie Clare (RDA) |
| 16:40-16:50 | A New Lens: Four Archetypes of Embedded Intelligence | Natalie Meyers (Association of Research Libraries, USA) |
| 16:50-17:00 | What is Agentic AI? — Evaluation | Sayeed Choudhury (Open Forum for AI, USA) |
| 17:00-17:25 | Q&A | All |
| 17:25-17:30 | Closing Remarks and Next Steps | Connie Clare (RDA) |
What Was Covered?
- What agentic AI is and how it compares to LLMs and generative AI
- What makes a tool agentic – and why it matters for research
- Practical examples and case studies from leading practitioners
Getting Involved
These sessions were open to all and free to attend. Whether new to the topic or already exploring agentic AI in work, this was an opportunity to learn, ask questions, and help shape the future direction of this community-driven initiative.
For questions or comments, please contact RDA Community Development Manager, Connie Clare ([email protected]).
What is a Data Director?
Join us for a 1-hour online information session designed to inform and inspire the global research data community. This session will take a deeper dive into a Data Director, exploring its features and functionalities, and how to begin developing an agentic AI blueprint for such a tool. There will also be space for open discussion on the opportunities and challenges ahead, as well as suggestions for improvement.
What is a Funding Finder?
Join us for a 1-hour online information session designed to inform and inspire the global research data community. This session will take a deeper dive into a Funding Finder, exploring its features and functionalities, and how to begin developing an agentic AI blueprint for such a tool. There will also be space for open discussion on the opportunities and challenges ahead, as well as suggestions for improvement.
