Inside the New Trustworthy AI Agent Testbed 1.0: Standardizing Multi‑Agent Collaboration

Recently, the 2025 Nanjing Artificial Intelligence Industry Development matchmaking event was held at Hongfeng Technology Park in the Nanjing Economic Development Zone. At the event, the China Academy of Information and Communications Technology together with eight companies including 360, Tencent and Huawei released the "Trustworthy AI Agent Testbed 1.0", aiming to build a standardized environment covering multi‑agent collaborative testing and to comprehensively verify and optimize agents' understanding, planning, communication, and task execution capabilities, promoting the transition of agent technology from the lab to large‑scale industrial applications.

As a key application modality of large models, agents can transform model capabilities into practical problem‑solving abilities, showing great potential in cross‑scenario collaboration and task execution. However, agents face many challenges in real‑world deployment, such as complex workflows, unclear collaboration logic, doubtful identity trustworthiness, and inconsistent application reliability.

According to the organizers, the Trustworthy AI Agent Testbed features three core capability layers—resource allocation, environment construction, and task testing. Its core mission is to drive standardization and industry collaboration through three testing scenarios, helping enterprises reduce testing costs and improve collaboration efficiency.

From 2025 onward, large models are entering a new stage of empowering industry, making the conversion of technical capabilities into productive power a central industry question. In this process, agents, as the primary vehicle for large‑model deployment, are reshaping industry application patterns by linking intelligent decision‑making with business processes, allowing AI to truly integrate into industry and unleash huge value. The market size of agents is expected to reach ten times that of traditional SaaS models, further pushing the transformation from "software development" to "agent development".

Taking 360's recently launched Nano AI Super‑Search Agent as an example, the Nano AI Super‑Search Agent can understand complex user needs, automatically plan processes, decompose tasks, autonomously invoke tools, execute the entire workflow, and deliver final results, achieving several‑fold efficiency gains over traditional manual collaboration.

At the conference, 360 and the China Academy jointly released the "Agent Industry Map", outlining and promoting agent technology innovation and industry application practices, accelerating deep integration of agent technology with industry, and providing a compass for industry insight into technology trends and core talent selection. 360 AI Research Institute expert Feng Meng shared experiences on building agents.