Organizations increasingly introduce AI agents that do not wait for explicit user requests, but monitor ongoing work in the background and intervene when support may be needed. These proactive or ambient agents are especially promising in high cognitive load interactions, such as customer service calls, sales conversations, technical support, healthcare consultations, emergency coordination, or complex advisory work. In such settings, workers must listen, interpret, decide, document, and respond in real time. AI agents could reduce cognitive burden by surfacing relevant information, detecting risks, or suggesting next steps at the right moment.

However, proactive intervention is difficult to design. If the AI intervenes too often, too early, or in the wrong format, it may distract users and increase cognitive load. If it intervenes too late or remains passive, it may fail to support users when support is most needed. The central challenge is therefore not only whether AI agents can provide useful information, but when, how, and under which conditions they should intervene.

This thesis investigates how proactive and ambient AI agents should be designed for high cognitive load interactions. Based on qualitative interviews with professionals, managers, or AI implementation experts, the study identifies design requirements for AI agents that monitor work in the background and provide timely, context-sensitive support without disrupting human attention.

The thesis is supposed to answer the research question: "What design requirements should guide proactive and ambient AI agents that intervene during high cognitive load interactions?"

If you are interested in this thesis, feel free to contact Marc Grau (marcchristopher.grau@unisg.ch).