This research tackles key limitations in traditional meeting scheduling systems — including weak natural language understanding, inefficient multi-attendee coordination, and imprecise resource matching — by proposing an intelligent scheduling system based on AI agents. Leveraging large language models like DeepSeek-V3 to interpret unstructured time descriptions, the system establishes a multi-workflow collaboration mechanism that enables end-to-end automation from requirement input to meeting confirmation. Experimental results confirm the system’s superior performance over conventional methods in critical areas including time parsing accuracy, coordination efficiency, and meeting room utilization.