DeepSeek unveils an AI multi-agent orchestration design aimed at overcoming the limit of one person supervising only 3–5 work sessions at a time.
Chinese artificial intelligence (AI) company DeepSeek has unveiled a design to make AI coding assistants more autonomous, enabling them to perform more tasks with minimal human intervention. According to a research paper by the DeepSeek Harness team, reported by the South China Morning Post (SCMP) on Sunday, the goal is to reduce the role of humans in coordinating and supervising multiple AI agents at once.
The research team’s core argument is that AI’s bottleneck is shifting. For a single task, the limitation lies in the model’s capabilities; but when multiple tasks run in parallel, the constraint becomes human attention. According to DeepSeek, one person can monitor only three to five active AI agent sessions at a time, and each session beyond that threshold reduces the quality of oversight.
The proposed system would take on tasks traditionally handled by human operators: assigning tasks among agents and coordinating their workflows. People would no longer need to monitor every step of execution, and could instead focus on evaluating results and making final decisions. In essence, this approach shifts people from the role of operator to that of approver—a direction many AI coding-tool developers are pursuing as models become increasingly capable of handling long task sequences.
For businesses, the significance of this approach lies in its potential to scale. If the number of agents an engineer can manage is no longer limited by their own attention, software development teams could see productivity increase exponentially. However, reducing direct oversight also raises questions about reliability, as agents’ errors could accumulate before a person notices them.
The AI agent race between China and the US
According to SCMP, DeepSeek’s proposal comes amid intensifying competition among AI developers, including US companies such as Anthropic and OpenAI. AI agents—particularly autonomous coding assistants—are emerging as one of the main battlegrounds in the technology race between the world’s two largest economies.
DeepSeek has previously attracted attention in digital financial markets. In October 2025, Cointelegraph reported that Chinese AI models outperformed US models in a crypto-asset trading test, based on data from the CoinGlass platform. At the time, the DeepSeek chatbot topped the ranking and was the only model to record a positive unrealized return, at 9.1%. The result surprised traders, since DeepSeek was developed at only a fraction of the cost of its US rivals.