Agentic Workflows: Human AI Interaction Design for Conversational AI

This podcast is about the paper that investigates agentic workflows for conversational human-AI interaction (CHAI) to address ambiguity and transience challenges. The authors propose a research-through-design approach, creating an AI chat web app with AI agents assisting users. The probe implements a structured workflow consisting of three stages: contextualisation, goal formulation, and prompt articulation, to guide conversations. The study iterates through four design cycles with user feedback driving improvements. Findings suggest agentic workflows help users clarify intentions and designers understand user needs in CHAI, leading to opportunities for enhanced AI design. The interactive probe acts as a 'Needfinding Machine', to reveal both explicit and implicit user needs in CHAI.
To read the paper: arxiv.org/abs/...