Staying in the Loop: A Biocurator's Guide to Agentic AI Developments

Staying in the Loop: A Biocurator’s Guide to Agentic AI Developments

Large language models have rapidly become part of everyday scientific workflows, yet most biocurators still interact with AI primarily through web-based chat interfaces. While useful, these tools are inherently limited: they lack deep context, struggle with complex multi-step tasks, and cannot reliably manage evolving datasets or ontology workflows. A new paradigm—agentic AI—is beginning to change this.

Agentic systems move beyond one-off prompts. They operate in iterative loops: calling tools, searching databases and literature, editing files, reasoning over results, managing state, and adapting plans. For biocuration, this opens the door to end-to-end support for tasks such as ontology refactoring, large-scale annotation review, literature-driven disease mechanism modeling, and guideline-based pathway revision. Increasingly, ontology communities such as CL, GO, Mondo, Uberon, and EFO are beginning to experiment with these approaches.

However, agents do not eliminate the need for expertise—they amplify it. Effective use of agentic AI requires skilled “drivers” who can supervise autonomy, choose appropriate modes (from tightly supervised to fully autonomous), and integrate these systems into real-world curation pipelines. The role of the curator is evolving from information extractor to workflow designer, systems thinker, and AI manager.

This workshop provides a practical, forward-looking guide for biocurators navigating this shift. Rather than offering a simple prompting recipe, it outlines how to build the necessary skills, choose appropriate tools (including deep research modes and code-based AI environments), secure institutional support, and cultivate communities of practice

Recording (whole conference)
https://www.youtube.com/watch?v=jXHmHHLp9q0