DeepSeek unveils design for AI coding agents requiring less human input
Key Takeaways
- •DeepSeek has unveiled a design intended to make AI coding assistants more autonomous by enabling them to complete multiple tasks with less human intervention.
- •A paper by DeepSeek's Harness team states that a person can supervise only three to five active AI agent sessions simultaneously, making human attention the bottleneck for multitasking.
- •The proposed system would distribute tasks across agents and coordinate their work, leaving humans free to review results and make decisions.
- •The announcement comes amid intense competition with US companies OpenAI and Anthropic, with coding agents emerging as one of the most contested fronts in the AI race.
- •The news follows an October report in which DeepSeek's chatbot was the only AI model to post a positive unrealized return, at 9.1%, in a cryptocurrency trading experiment.
Chinese artificial intelligence firm DeepSeek has unveiled a design intended to make AI coding assistants more autonomous, enabling them to carry out multiple tasks with less human intervention.
The project aims to reduce the human effort needed to coordinate and monitor multiple AI agents, according to a paper published by the company's DeepSeek Harness team and reported by the South China Morning Post on Sunday.
The team said managing multiple agents currently demands substantial human attention, noting that "single-task limits depend on model capabilities, while multitask limits depend on human attention."
Citing the paper, DeepSeek said a person can oversee only three to five active AI agent sessions simultaneously. Its proposed system would distribute tasks among agents and coordinate their work, leaving humans free to review results and make decisions. That framing underscores why agent orchestration has become a priority: as developers run more AI agents in parallel, the supervision workload itself can become the bottleneck, regardless of how capable each individual model is.
The proposal comes amid intense competition among AI developers, including US companies Anthropic and OpenAI, according to the SCMP. Coding agents have emerged as one of the most contested fronts in that race, with labs working to automate larger portions of the software development workflow and to cut the human oversight each deployed agent requires.
The development follows a Cointelegraph report last October that Chinese artificial intelligence models were outperforming US counterparts in cryptocurrency trading, based on data from blockchain analytics platform CoinGlass. DeepSeek's AI chatbot led that crypto trading experiment, standing as the only AI model to generate a positive unrealized return, at 9.1%. The results surprised crypto traders, given that DeepSeek was developed at a fraction of the cost of its US rivals.
Whether DeepSeek's design moves beyond a published paper into a tool developers can actually use — and how it stacks up against agent offerings from rivals like Anthropic and OpenAI — remains to be seen.
Source: Cointelegraph
