Inherent human limitations pose significant challenges for team planning in a highly dynamic crisis environment. Stress, urgency, resource conflict, and confirmation bias may get in the way of establishing a clear shared mental model of what needs to get done and how the response team will proceed. Our research effort is aimed at developing intelligent agents that effectively particpate in human team planning session for disaster response. We are developing algorithms to detect weak agreements among team members at planning time, and track plan formation directly from the human team's planning conversation. This reseach is performed in collaboration with Lincoln Laboratory, and our long term goal is to integrate this technology into the Next Generation Incident Command System (NICS), a web-based tool for distributed planning and shared situational awareness.