Huawei Connect 2026: 360 and Ascend AI Launch Joint Solution to Accelerate AI Agents

The launch follows the two sides' joint demonstration at the 2026 World Artificial Intelligence Conference, according to Leiphone, and represents a further step in bringing their cooperation into deployment.

360 Agent Factory handles agent generation and orchestration, supporting the rapid generation of agents through natural language and the formation of multi-agent collaboration swarms. Its swarm multi-agent collaboration technology supports multi-layer nesting, flexible teaming and shared collaborative memory, enabling autonomous task planning and coordinated execution, according to the report. It reduces a multi-role collaboration task that originally took two hours to 20 minutes, a sixfold efficiency improvement.

As AI Agents move from single-turn question answering to complex task execution, asynchronous parallelism, multi-agent coordination, long-context reasoning and high-frequency tool calls are becoming mainstream application forms. Agent operations require continuous task decomposition, planning decisions, tool calls and result feedback, raising requirements for the stability, throughput and response speed of underlying computing platforms.

Ascend AI has optimized for long-sequence inference and frequent KV Cache reads and writes. Through long-sequence optimization, throughput optimization, runtime optimization and operator optimization, it reduces first-token latency and raises overall throughput, supporting high-concurrency, long-chain Agent tasks.

In typical scenarios such as deep understanding, general understanding and agent execution, Ascend AI has shown good out-of-box performance with mainstream models including DeepSeek, Qwen and GLM, and achieved overall performance improvements after further optimization, according to Leiphone.

With Ascend AI computing support, 360 Agent Factory has in beta testing achieved continuous execution of complex tasks exceeding 1,000 steps without interruption, with a task success rate above 95 percent. It uses a transparent human-machine collaboration mechanism that visualizes the agent's thinking process, task planning path and execution status, allowing users to track progress and results in real time. Millisecond-level response capability underpins that experience.

According to Leiphone, the AI industry is moving from big model competition toward agent competition. The value of agents depends not only on model capability but on collaboration efficiency among models, tools, workflows and computing infrastructure. The 360 and Ascend AI joint solution is another exploration of coordinated development between agent infrastructure and China's domestic AI computing. The two sides plan to continue integrating agent infrastructure with the domestic AI computing ecosystem to provide more efficient and stable support for large-scale AI Agent applications and help more enterprises move faster into the AI era.