How to Use the Projects Feature in ChatGPT

Why Read This from an HCI Perspective

This article helps you see AI not as a ‘smart feature,’ but as something users work with—an interactive partner. For HCI/UX practitioners and researchers, it suggests that interaction elements such as trust, user control and intervention, and failure recovery may matter more than raw model performance. It’s especially meaningful because it prompts you to consider how responsibilities are divided and how user workflows change once this capability is built into a product.

CIT's Commentary

AI agents and automation features may look convenient on the surface, but in practice the key is the interface that tells users when to trust and when to stop. Instead of reducing the number of buttons, if the system state becomes unclear, users end up taking on greater risk. So it’s not enough to focus only on performance improvements—you also need to design, together, where users can intervene when something fails and what signals the system uses to communicate its state. The interesting part is that solving these problems well requires sophisticated UX measurement tools and evaluation methods. Research rigor shouldn’t loosen just because you’re using LLMs; if anything, more transparent validation approaches are required. In a domestic service environment where teams add features quickly, this balance becomes even more important.

Questions to Consider While Reading

• Q.What is the minimum set of state signals that enables users to predict the AI’s next action?
• Q.When automation fails, how far should the intervention points that the system hands over to users go?
• Q.When building UX measurement tools using LLMs, what criteria are needed to preserve research rigor?