Design challenges for multi-agent, multi-user, and branching collaborative deep research

Determine effective system and interface designs for extending InterDeepResearch-style interactive deep research systems beyond single-agent–single-user settings to support multiple large language model agents, multiple human users, and branching research threads, by identifying and addressing the key design challenges in such complex collaborative scenarios so as to enable higher information processing bandwidth and more comprehensive research coverage.

Background

InterDeepResearch introduces an interactive deep research system that organizes research context across information, actions, and sessions, enabling users to observe and steer an LLM agent’s research process and to navigate evidence provenance. The evaluations focus on a single-agent–single-user setting.

In the conclusion, the authors point to extending the system to more complex collaborative scenarios—supporting multiple agents, multiple users, and branching research threads—as a promising direction for increased capacity and coverage, while noting that the corresponding design challenges remain unresolved.