Speaker
Description
The recent rise in capabilities from AI systems has begun to change the way that user interfaces are developed, making it significantly easier for domain scientists to create their own applications, or for developers to rapidly construct applications in previously unfamiliar frameworks.
At the Advanced Light Source, the Photon Science Computing (PSC) group is systematically exploring the use of AI with application development. Several strategies are being employed including hands on hackathon training sessions, publishing agent skills, and conducting internal studies to provide recommendations.
This presentation focuses on one particular case study and its conclusions. A beamline scientist used Agentic AI to create a PyQt desktop application for beamline controls and data acquisition with the Bluesky framework. PSC staff then used Agentic AI to recreate this PyQt app in a modern web stack with React and the Bluesky React library Finch.
Throughout this process, limitations in AI coding were observed including inconsistent styling, re-implementation of existing components, and layout instability. Strategies to mitigate these deficiencies have been made which include modifications to existing code bases, creation of new AI skills, and developing guidelines for using Agentic AI.
This work provides insights and recommendations for how to prepare for the continued advancement and adoption of Agentic AI at user facilities for scientific applications.
