AgentExtension: A Three-Modality Framework for Human–Agent Web Interaction

In the past few years, there has been a growth of interest from both developers and users in agentic systems. The main issues with these systems are the lack of transparency, the psychological problems related with trust and the necessity to provide more control to the end user. For this reason this thesis aim is to design and develop a web agent which operates with three different modalities- Learn, which enables the user to learn new tasks; Teach, which allows users to teach tasks to the agent; Collaborate, which can assist users in automating the steps of their tasks. The developed agent was evaluated and validated through a user study, using the guidelines from current paperwork about agents design and development, which have then been put together with heuristics from Human-Agent Interaction.