Agentic AI
New System Lets AI Agents Build and Share Scientific Workflows From Start to Finish
CURATE uses AI agents to automate the entire lifecycle of computational work, from coding through deployment and team sharing.
Researchers have introduced CURATE, a system that hands off multiple stages of workflow development to AI agents, addressing a long-standing gap in how computational tasks get built, organized, and shared among teams.
The Gap in Current AI Coding Tools#
Today's AI coding agents excel at one task: generating code on demand. They solve the immediate problem in front of them but leave users to handle everything else—stitching components together, configuring deployment, and managing how others on the team access and reuse the work. That leaves developers doing repetitive, error-prone manual labor outside the AI's purview.
How CURATE Works#
CURATE is a human-in-the-loop multi-agent architecture that treats workflow development as an end-to-end problem. Rather than stopping at code generation, the system coordinates:
- Composition: AI agents design and assemble modular components into complete workflows
- Cataloging: A central repository stores reusable modules and workflow segments, making them findable and ready to apply to new tasks
- Deployment: The system handles the technical setup needed to run workflows in practice
- Automated execution: Agents manage ongoing task execution without constant human intervention
The catalog sits at the heart of the design. By making modules discoverable and reusable, CURATE lays groundwork to support FAIR principles—standards that emphasize making research data and tools findable, accessible, interoperable, and reusable.
What the Research Shows#
The team tested the approach with a working prototype and six experiments that demonstrate the system's potential to streamline workflow development from conception through deployment. Rather than each developer reinventing the wheel, teams can now draw on a shared library of pre-built, proven modules.
