DeepSeek’s New AI Agent System Needs Less Human Input

Chinese AI developer DeepSeek’s Harness team has outlined a system designed to make coding agents more autonomous by coordinating multiple tasks with less human intervention. The proposal, detailed in a company paper and reported by the South China Morning Post, aims to reduce the human effort required to assign, monitor, and synchronize the work of AI agents.

DeepSeek said a person can currently oversee only three to five active AI agent sessions at once. Its proposed system would distribute tasks among agents, coordinate their activities, and allow humans to focus on reviewing results and making higher-level decisions. The team argued that “single-task limits depend on model capabilities, while multitask limits depend on human attention.”

The initiative comes amid strong competition among AI developers, including Anthropic and OpenAI, as companies seek to build coding systems capable of handling increasingly complex workflows. DeepSeek’s approach focuses on orchestration and supervision, potentially allowing a smaller team to manage more agents without requiring continuous manual input.

The development also has implications for AI-driven crypto trading. Cointelegraph previously reported that, according to CoinGlass data, DeepSeek was the only model to achieve a positive unrealized return of 9.1% in a cryptocurrency trading experiment, a result that surprised traders given DeepSeek’s comparatively low development costs.