NetEase Zhigi's Emperor Crab upgrades at Wuzhen: launches Agent Team, knowledge engine "Wandian" and debuts smart badge

Smart Things reported on November 4 that today, the Zhejiang Sub-forum of the 2026 World Internet Conference Wuzhen Summit was held. At the forum, Duan Yuzheng, Deputy General Manager of NetEase Zhigi, delivered a keynote speech titled "Agents: Let Good Things Happen at Every Site," and announced three upgrades to NetEase Zhigi's enterprise-level AI Agent platform "King Crab": the debut of a hardware product, the Smart Badge, and the launch of two new capabilities, the knowledge engine "Wandian" and Agent Team.

▲Duan Yuzheng, Deputy General Manager of NetEase Zhigi

Duan Yuzheng said that enterprises' attention to AI is gradually shifting from "how powerful the model is" and "which Agent to use" to "whether AI can truly enter business operations and continuously create value," which means enterprise AI is moving from technical validation to productivity validation.

As model capabilities gradually become widespread, more attention is being paid to the differences in actual business results when the same model enters different enterprises. Duan Yuzheng believes that what will truly set enterprises apart in the future is not just which model they adopt, but whether AI can understand an enterprise's own business, organization, knowledge, experience, processes, rules, and ways of doing things.

He summarized this information, knowledge, rules, processes, and practical experience that have long accumulated in enterprise operations and collaboration as "enterprise context."

Compared with model capabilities, the business logic and practical experience that an enterprise has formed over time are harder to simply replicate. As AI enters business operations more deeply, this enterprise-specific information and ways of doing things will also increasingly directly affect AI's actual work results.

Around how to bring enterprise context into specific tasks, King Crab brought three upgrades this time.

Enterprise business does not only happen in online systems. Scenarios such as customer visits, store inspections, on-site services, and internal communication generate a large amount of frontline information every day, but this content often remains offline, is difficult to bring into enterprises' existing information systems, and is also difficult for AI to further understand and use.

In scenarios such as customer visits, offline communication content can enter King Crab through the Smart Badge, and the platform completes subsequent analysis and processing. The device has a built-in SIM card and can connect directly to the cloud, while also supporting a physical privacy switch, recording status prompts, and recording time period management, making the entry of on-site information visible and manageable.

As a result, the information AI can process is no longer limited to online systems such as documents and CRM, but further extends to real business sites. Customer needs, frontline issues, and on-site feedback can also enter subsequent business analysis and task execution.

For example, a customer need may first appear in a visit recording, then enter CRM, email, or a work order, and further affect proposals, quotations, and delivery.

Second, "Wandian," making enterprise knowledge findable, trustworthy, and usable

After on-site information enters the system, enterprises face another problem: the information needed to complete a task is often scattered across different systems and constantly changing.

Documents, emails, work orders, CRM, databases, third-party knowledge services, as well as customer visits, meetings, and on-site communication may all contain important evidence. When facing complex business tasks, AI not only needs to find relevant content, but also needs to understand the connections between pieces of information and determine which facts are reliable and which have changed.

The knowledge engine "Wandian" aims to enable Agents, when facing complex tasks, to continuously understand and integrate constantly changing information in the enterprise and form a more complete and reliable task basis.

It can dynamically organize information from different sources and at different times, such as documents, emails, work orders, CRM, databases, and on-site communication, around specific tasks, continuously search, supplement evidence, and verify, and identify the logical and temporal relationships between different information and events.

For example, a customer need may first appear in a visit recording, then enter CRM, email, or a work order, and further affect proposals, quotations, and delivery.

"Wandian" not only needs to find relevant information, but also needs to determine which are the latest facts, which can corroborate each other, what evidence is still missing, and form a task basis that Agents can use for subsequent judgment and execution.

When evidence is insufficient, it continues searching; when it cannot be confirmed, it truthfully explains; and relevant conclusions can also be traced back to the original text, specific page numbers, or corresponding time points in recordings.

After obtaining sufficient information, the Agent can continue to call business tools and bring knowledge and analysis results into subsequent task execution; the same set of knowledge and permission systems can also be reused by different Agents.

Third, Agent Team, enabling multiple Agents to collaborate on complex tasks

For complex tasks that span roles and steps, King Crab launched Agent Team.

In actual enterprise business, a task often requires the joint participation of multiple professional roles. Taking contract review as an example, a contract may involve multiple departments such as business, finance, risk control, and legal at the same time, and different roles focus on different issues.

In Agent Team, a coordinating Agent is responsible for breaking down tasks, assigning work, and summarizing results, while Agents in different professional roles handle corresponding tasks respectively and can discuss, question, and verify around the same issue.

For cross-departmental, multi-step processes, different Agents can also communicate and hand off to each other according to task progress. Tasks that can be parallelized are processed simultaneously, while steps with sequential dependencies proceed in order.

Duan Yuzheng said that multi-Agent collaboration, on the one hand, allows different professional perspectives to jointly participate in judgments on complex issues, improving decision quality through mutual discussion and verification; on the other hand, it can also drive long-chain processes that originally relied on manual coordination to continue advancing, improving process automation and autonomous execution capabilities.

At key judgment and important decision points, humans still retain authorization and final decision-making.

IV. Behind the three upgrades is an enterprise's own "context"

From on-site information entering the system, to multi-source enterprise knowledge being understood and used, to multiple Agents collaborating to complete complex tasks, the three upgrades connect the information, knowledge, and execution chains required for AI to enter real business operations, enabling enterprise context to enter the judgment and execution process of specific tasks more completely.

NetEase Zhigi has long served enterprise communication and collaboration scenarios, with its instant messaging and audio/video services handling more than 340 billion messages per month on average and serving 2.35 million enterprise developers.

Its long-term participation in information flows within enterprises and between enterprises and their customers and suppliers has also given it sustained exposure to how enterprise information is generated, flows, and accumulates, as well as how different roles collaborate.

On the path to implementing enterprise AI, Duan Yuzheng believes there is no need to pursue "big and comprehensive" from the outset; a more realistic approach is to start from specific scenarios and tasks, deepen high-value problems, and then gradually extend to upstream and downstream links, connecting them into a complete business chain, so that AI can move from single-point efficiency gains to operational efficiency gains.

He summarizes this process as: scenario implementation, closed-loop connectivity, and operational efficiency gains.

Enterprises can start with specific tasks such as sales, R&D, production, and service, and then extend upstream and downstream to form business closed loops such as "lead to cash," "requirement to delivery," "issue to resolution," and "strategy to execution."

On this basis, multiple Agents can also participate in discussion and verification from different professional perspectives, and AI's involvement in enterprise analysis and judgment will further increase.

In the longer term, King Crab's goal is to further become an enterprise's "decision engine."

Duan Yuzheng believes this does not mean letting AI replace humans in making final decisions, but rather bringing more real business information and professional perspectives into the judgment process to help enterprises make faster, higher-quality decisions, with humans retaining authorization and final decision-making at key stages.