DeepSeek Unveils System to Make AI Coding Assistants More Autonomous

DeepSeek has outlined a design that would let AI coding assistants handle several tasks with far less human supervision. The plan comes from the company’s DeepSeek Harness team, which published a paper on the idea. Its goal is to cut the amount of human effort spent coordinating and watching multiple AI agents at once.

Why Managing Multiple AI Agents Is the Bottleneck

The team argues that the limit on a single task depends on how capable the model is, while the limit on many tasks at once depends on human attention. According to the paper, one person can realistically oversee only three to five active agent sessions at the same time. The proposed system would assign work across agents and coordinate them automatically, so people can focus on reviewing results and making decisions.

Competition in the AI Coding Race

The announcement arrives during fierce rivalry among AI developers, including US firms Anthropic and OpenAI. DeepSeek has also drawn attention in crypto circles. Last October, CoinGlass data showed Chinese AI models beating US rivals in a crypto trading experiment, where DeepSeek was the only model to post a positive unrealized return, at 9.1%. That surprised many traders, given it was built at a fraction of the cost of American competitors.