HARP platform lets researchers run controlled AI experiments with live agents

A team of researchers from an unnamed institution (paper on arXiv) has introduced HARP, the Human-AI Research Platform, designed to fill critical gaps in studying human-AI interaction. Traditional HCI methods rely on static prototypes or post-hoc transcript analysis, which miss how users formulate, revise, and hesitate over prompts. HARP places participants in controlled mock scenarios with live, configurable AI agents. Researchers can control agent prompts, model parameters (e.g., temperature), response length, and experimental conditions. The platform records detailed behavioral metrics: prompt composition time, deletions, keystroke pauses, and response latency. It also allows triggering surveys at predefined moments during the interaction.

Beyond text, HARP supports planned capabilities for voice, facial expression, gesture, and—where ethically appropriate—emotion analysis. The paper illustrates HARP's utility with a study examining how technical specificity and response length affect retention of LLM output. By combining controllable live agents with behavioral and self-report measures, HARP enables systematic testing of how AI design choices influence user experience. The platform targets researchers, designers, and anyone wanting to answer 'What if AI did this?' in a rigorous, repeatable way.