[Digital Today Yoo Seung-a, intern reporter] Chinese artificial intelligence (AI) company DeepSeek has unveiled a blueprint that can run multiple AI coding agents simultaneously while minimizing human involvement.

On Nov. 1 (local time), blockchain media outlet Cointelegraph reported that DeepSeek set out ways to boost the autonomy of AI coding agents in a technical paper published by its DeepSeek Harness team.

The key aim is to reduce the burden of having people adjust each agent's work and check progress one by one. DeepSeek's research team assessed that the bottleneck in operating multiple agents lies not only in model performance but also in human attention. The team explained, "The limit of a single task depends on the model's capabilities, but when performing multiple tasks at the same time, human attention acts as the limiting factor."

According to the paper, the number of active AI agent sessions that one person can supervise at the same time is about 3 to 5. The system DeepSeek proposed aims to ease that constraint by allocating work across multiple agents and coordinating their progress. Instead of managing each task individually, people can focus on reviewing the results produced by agents and making decisions.

The blueprint comes as competition among AI developers over agent technology intensifies. Major U.S. companies such as Anthropic and OpenAI are also accelerating development of AI coding assistance technology. DeepSeek also joined the competition by presenting an approach in which multiple agents autonomously carry out complex tasks.

DeepSeek also drew attention last year in a cryptocurrency trading experiment. At the time, Cointelegraph cited data from blockchain analytics platform Coinglass in reporting that a Chinese AI model outperformed U.S. models in a cryptocurrency trading test. Among the AI chatbots that took part in the experiment, only DeepSeek posted an unrealised return of 9.1 percent, making it the only one to generate positive returns. Interest also grew because it was said to have been built at a much lower development cost than U.S. competitors.

The proposal shows that the focus of AI competition is expanding beyond the performance of individual models to how efficiently multiple agents can be operated. The details released this time are a blueprint for autonomous agent operations, and actual performance results or whether it will be commercialised have not been confirmed.