[DeepSeek Releases New Agent Architecture] According to Jinse Finance, on the evening of October 31st, the DeepSeek Harness team published a technical article titled "Exploring Super-agent Design: Towards a More Autonomous Agent System." DeepSeek stated that improved model capabilities have spurred new application forms, with personal agents recently emerging in large numbers. However, existing mainstream agent designs still struggle to support this model on a multi-tasking and team-based scale. Agents still operate on a single session as the basic unit, with different sessions operating independently. Under this design, the upper limit of a single task depends on the model's capabilities, while the upper limit of multiple tasks depends on human attention. Through a two-layer architecture, event-driven architecture, asynchronous multi-channel interaction, storage-computation separation, and external watchdogs, the DeepSeek team has overcome these limitations. The system takes over the cumbersome session management and context synchronization at the lower level, allowing humans to move from specific execution details to higher-level decision-making and acceptance. DeepSeek refers to this agent type and its system architecture as Super-agent. The article mentions that the internal Super-agent instance built based on this idea is already running continuously, participating in daily problem reproduction and fixing, code review responses, and the iteration of DeepSeek Harness itself.