DeepSeek AI Agents Could Operate With Less Human Input

Chinese artificial intelligence developer DeepSeek is designing a system that could allow AI coding assistants to complete multiple tasks with less human supervision. The proposal comes from the company’s DeepSeek Harness team and was reported by the South China Morning Post.

DeepSeek’s approach would distribute work among multiple AI agents, coordinate their activities and let people focus on reviewing results and making final decisions. The team estimates that a person can currently monitor only three to five active AI-agent sessions at once, highlighting human attention as a key constraint when running several agents simultaneously.

The development comes amid strong competition between DeepSeek, OpenAI and Anthropic to build more capable autonomous AI systems. DeepSeek previously gained attention after its chatbot led a cryptocurrency-trading experiment with a 9.1% unrealized return, making it the only AI model in the CoinGlass-backed test to achieve a positive result.