About the workshop
From acting agents to a shared research agenda
Agentic AI systems—autonomous agents capable of planning, reasoning, and acting on behalf of users—are rapidly moving from prototypes into the world. That shift raises urgent questions of reliability, autonomy, accountability, and governance, especially when agents operate over personal and enterprise data.
Personal Knowledge Graphs (PKGs) offer a promising foundation: explicit, machine-readable representations of context, policies, and provenance. Building on a successful Dagstuhl Seminar and prior TAPAI/TAAPAAI editions at ISWC and ESWC, this Barstuhl workshop convenes the community to consolidate architectures, decentralized ecosystems, and governance models for agentic systems grounded in semantic technologies.
The format is discussion-first. The primary contribution is not a proceedings volume, but a co-created research agenda, a map of open challenges, and a common conceptual foundation that places the Semantic Web community at the forefront of PKG-based agentic AI.
Motivation
Five challenges that need a community
Agentic AI is moving from systems that generate outputs to systems that act in complex environments. Centralized infrastructures risk undermining data sovereignty. The field still lacks shared architectures, conceptual frameworks, and interoperability standards for PKG-based agentic AI. These challenges are socio-technical—and best tackled in a Dagstuhl-style room.
Architectures for PKG-based agentic systems
How should a personal knowledge graph sit inside an agent loop—planning, memory, tool use, and action? We need reference architectures that make context, policies, and provenance first-class, and that scale from a single personal agent to families of collaborating systems.
Themes
Explore the topic space
A selective set of themes at the intersection of PKGs, agentic AI, and governance. Filter a cluster, then open any topic for a short framing.
Selected theme
Workshop format
A Barstuhl afternoon
Named for Bari’s turn as host, Barstuhl borrows the Dagstuhl seminar spirit: interaction over presentations, collective sense-making over a packed talk list, and artifacts the community can take home.
Arrive with a point of view
We open with framing from the organizers, four short invited provocations, and lightning talks from the room. Submit a one-slide proposal in PPTX or PDF if you would like to speak.
Think together, then rotate
Thematic breakouts and a rotating discussion groups let mixed groups of Semantic Web, AI, policy, and industry voices work the same questions from different angles, then reconvene.
- Thematic breakout sessions
- Rotating discussions
- Plenary consolidation with live maps and clusters
Leave with a shared artifact
The afternoon closes by turning discussion into a draft research agenda: challenges, terminology, standardization priorities, and a vision the community can keep building after Bari.
- Roadmapping and consensus-building
- Documented on this website and in follow-up writing
- Seeds for collaborations and joint proposals
Schedule
Half-day Program
Introduction and framing
Organizers set the Barstuhl brief: why PKG-based agentic AI, and what we intend to leave with.
Invited provocations
Four 15-minute talks.
So how come we can’t do this??
Jim Hendler
25 years ago, I was one of the authors on a paper that started with an agent-based scenario. All the technology we referred to, and much more is now available — yet problems like the one we stated are still not doable. My provocative question is “why not”???
An Agent in the Classroom: Deploying an Agentic TA for Ontology Engineering
Deborah L. McGuinness
While LLM-based tools are changing the way students learn, agentic systems present new possibilities for changing the way educators teach. In this talk, I will share observations from deploying an agentic teaching assistant in an ontology engineering course. I will examine how the agent handled grading and feedback on student-designed ontologies and ontology-related artifacts, where it failed, and what surprised us. I will use our experience as a concrete case study to consider opportunities and open research questions at the intersection of agentic AI, personal knowledge modeling, and human-agent interactions in educational settings.
Towards the Cognitive University: What Happens When the Whole University Becomes Agentic?
John Domingue
AI agents are beginning to move beyond individual applications to become participants in the processes through which universities teach, support learners, create knowledge and operate. What happens when we follow this trajectory to its logical conclusion?
Drawing on decades of work at the intersection of AI, Knowledge Graphs and education, together with recent experience developing AI assistants for students and academic staff, I will introduce the idea of the Cognitive University: a university understood as an ecosystem of people, knowledge and collaborating AI agents.
I will argue that the interesting challenge is not simply to make universities more automated. It is to create institutions in which knowledge is machine-understandable, agents can act over it safely and interoperably, and students and staff retain the agency, skills and judgement required to flourish alongside them. This raises a set of questions that should be of particular concern to the Semantic Web community: What knowledge infrastructure do such institutions require? How should humans, agents and institutional knowledge interact? Where should autonomy reside? And what does it mean for a university itself to become cognitive?
Agent Spaces - last chance for a decentralised Web (... already gone?)
Axel Polleres
Talking about "personal agents" while relying on centralized, API-based frontier models is quite a paradox, isn't it? So, when we build such agents, we also need to define their scope: what knowledge they share, with whom, within which boundaries, and what tools they may use, but also which models they use. While we see local open-weight models progressing alongside large frontier models, which could enable such decentralized "Agent Spaces", we seem to again fall for the same trap as ever: are network effects and (perceived?) comfort stronger than the wish for decentralization and sovereignty?
Lightning talks
Short talks from participants. Submit a one-slide proposal (PPTX or PDF) via the Google Form.
Coffee break
Breakout session 1
Working groups on key themes: architectures, accountability, autonomy, coordination, and interoperability.
Breakout session 2
Rotations so every group inherits (and pressure-tests) another table’s notes.
Plenary synthesis
Expected results
What we will make together
Challenge map
A map of key challenges and future research directions for PKG-based agentic AI.
Shared language
A conceptual framework and terminology alignment the community can actually use.
Interop priorities
Identification of standardization and interoperability work that should come next.
Vision / manifesto
A draft vision document for the field, continuing the TAAPAAI series.
Collaborations
New collaborations, follow-up initiatives, and seeds for joint proposals.
Call for participation
Submit Your Ideas!
Barstuhl is open to ISWC attendees. We especially welcome Semantic Web and Knowledge Graph researchers, AI and agentic-systems researchers, experts in decentralization, data governance and policy, industry practitioners building agent-based systems, and early-career researchers.
Anyone interested in participating should submit a talk proposal as a single slide, in PPTX or PDF, via the Google Form.
Submit a one-slide talk proposal-
Semantic Web & Knowledge Graphs Researchers shaping PKG models, reasoning, and provenance.
-
AI and agentic systems People building agents that plan, remember, and act.
-
Decentralization, governance & policy Solid, sovereignty, GDPR, and the EU AI Act in practice.
-
Industry practitioners Teams shipping agent-based products who need governable context.
-
Early-career researchers & PhD students A focused room in which to help set the next agenda.
Chairs
