DeepSeek Unveils New Agent Architecture

On the evening of October 31, the DeepSeek Harness team published a technical article titled "Super-agent Design Exploration: Toward More Autonomous Agent Systems" on its official WeChat account. DeepSeek said that improvements in model capabilities have given rise to new application forms, and personal agents have begun to emerge en masse. However, existing mainstream agent designs struggle to support this model at the scale of multitasking and team collaboration, as the ceiling for multitasking is limited by human attention. Through approaches including a two-layer architecture, event-driven design, asynchronous multi-channel interaction, separation of storage and computation, and an external Watchdog, the DeepSeek team has overcome these limitations. The system takes over tedious session management and context synchronization at the underlying level, allowing humans to move further up from specific execution details to higher-level decision-making and acceptance. DeepSeek calls this agent form and its system architecture Super-agent. The article notes that internal Super-agent instances built on this approach are now running continuously, routinely participating in issue reproduction and fixing, code review responses, and iterations of DeepSeek Harness itself. When team members report software problems in group chats, it can identify potential defects even without anyone tagging the AI, and proactively attempt to reproduce the issues and find fixes in the background. (First Financial)