DeepSeek unveils AI ‘super agent’ in coding race with OpenAI, Anthropic

Chinese artificial intelligence start-up DeepSeek has previewed a plan for a more autonomous AI agent that can handle multiple tasks with minimal human intervention, as it steps up competition with US rivals OpenAI and Anthropic in developing coding assistants.

The proposed system, dubbed super agent, aimed to reduce the need for humans to coordinate multiple AI agents and monitor their progress, addressing a key bottleneck in managing today’s increasingly complex workflows, according to a technical article published by the company’s DeepSeek Harness team on Saturday.

DeepSeek said that advances in AI models had fuelled the emergence of agents that could automate a growing range of tasks, from managing calendars and drafting emails to preparing work reports and writing code.

However, managing multiple agents simultaneously still demanded considerable human attention, the team said.

“Single-task limits depend on model capabilities, while multitask limits depend on human attention,” the paper said.

The DeepSeek team estimated that a person could typically oversee only three to five active AI agent sessions simultaneously, limiting the productivity gains from deploying multiple assistants.

DeepSeek’s proposed system seeks to overcome this by automatically distributing tasks among agents and coordinating their work, leaving humans to focus on decision-making and reviewing results.

The initiative comes as competition intensifies among AI developers to build more autonomous coding assistants, with Anthropic’s Claude Code and OpenAI’s Codex increasingly capable of handling complex software development tasks.

DeepSeek’s proposal seeks to take agent automation further by allowing multiple agents to work simultaneously and proactively respond to events such as messages and code updates without requiring explicit instructions for every task.

01:58

China’s biggest chilli sauce maker Lao Gan Ma denies online rumours of mass lay-offs triggered by AI

Under the proposed architecture, a central coordinating agent would distribute assignments among multiple independent agents, which could execute tasks in parallel and report their progress to the central agent, reducing the need for human oversight.

In one example, the team said an agent could detect a software bug reported in a group chat and independently attempt to reproduce and fix it, even without being directly instructed to do so.

If it identified a viable solution, the agent could submit a proposed code change for human review. Otherwise, it would remain silent to avoid unnecessary interruptions, the team said in the paper.

In experimental tests, the approach enabled completion of one task in just under 16 minutes, compared with more than 29 minutes for a conventional approach, according to the team’s findings.

Some industry experts acknowledged that DeepSeek was making progress in the right direction, but questioned whether the proposal offered meaningful advances over existing systems.

“The challenge in AI software development today is ensuring reliability … rather than completing tasks faster,” said Liu Ming, a Beijing-based AI entrepreneur, on China’s Quora-like platform Zhihu.

DeepSeek itself acknowledged potential challenges around security and privacy, including the risk that malicious instructions could compromise the agents connected to other services and systems.

DeepSeek released an open-source research prototype of the super agent on code hosting platform GitHub, while stressing that the initiative remained an architectural exploration rather than a mature product.