[DeepSeek releases new Agent architecture] On October 31st, the DeepSeek Harness team published a technical article "Super-agent Design Exploration: Towards a More Autonomous Agent System" on the official public account on the evening of October 31st. DeepSeek said that the improvement of model capabilities has given rise to new application forms, and personal assistants (Personal Agents) have begun to emerge. However, existing mainstream Agent designs are still difficult to support this model on a scale of multi-tasking and team collaboration. The upper limit of multi-tasking depends on human attention. Through two-layer architecture, event-driven, asynchronous multi-channel interaction, storage and calculation separation, and external Watchdog, the DeepSeek team broke through these limitations. The system took over the tedious session management and context synchronization at the bottom, allowing humans to move from specific execution details to higher-level decision-making and acceptance. DeepSeek calls this Agent form and its system architecture Super-agent. The article mentioned that at present, the internal Super-agent instance built based on this set of ideas has been continuously running, participating in daily problem recurrence and repair, code review response, and DeepSeek Harness's own iteration. When team members report software problems in a group chat, even if no one takes the initiative to @AI, it can identify potential flaws, try to reproduce the problem in the background, and find fixes. (China Business News)