Enterprise Knowledge Search · IT Operations Automation
ShineWing needed to move beyond isolated assistants and build a secure enterprise AI layer for document-heavy professional services. It had to coordinate multiple agents, unify AI assets, connect internal systems safely, and control cost and auditability across a growing multi-model estate.
Reported impact
ShineWing established a controlled enterprise AI foundation that gives employees a common AI service entrance, reduces the friction of finding internal information and invoking workflows, and lets the firm extend AI from information lookup into multi-step task execution. The project also proved the platform could be integrated and launched in a one-week final sprint.
ShineWing built an internal multi-agent AI operating layer on Alibaba Cloud so employees could move from simple information lookup to governed task execution across audit, tax, consulting, and administrative workflows. AgentTeams acts as the orchestration plane: it exposes a unified entry point, recognizes user intent, routes requests to specialized agents for policy Q&A, OA integration, information retrieval, and document processing, and can decompose more complex tasks across several agents. Around that runtime, the firm centralized management of models, MCP servers, skills, and agent templates so permissions and reuse are handled consistently instead of in isolated projects. AI Gateway fronts the model layer with authentication, traffic control, failover between models, token accounting, and content-safety inspection, while end-to-end observability records agent runs, model usage, and abnormal behavior for audit and governance. The result is a secure enterprise AI base rather than a collection of disconnected assistants.
Employee requests enter a unified AI interface, where AgentTeams classifies the task and routes it to the right specialized agent or coordinates several agents together. Those agents call internal systems through MCP Server and use models protected by AI Gateway, while usage, exceptions, and token consumption are logged for governance and cost analysis.
In the final stage, both teams completed system integration, testing, tuning, and launch within one week. (original: 在项目最终关键阶段仅用一周时间完成系统联调、测试优化并正式上线。)
ShineWing said enterprise AI was moving from a single assistant to complex scenarios that required agent collaboration, unified asset management, and secure integration with OA and knowledge systems. (original: 多智能体协同难... AI 资产管理散... 需要将 AI 能力与内部 OA、知识库等核心业务系统安全连接。)
The company built multiple specialized agents for policy Q&A, OA integration, information lookup, and document processing, and AgentTeams can dispatch them or split complex tasks across several agents. (original: 构建了制度问答、OA 集成、信息查询、文档处理等多类专业 Agent。通过平台统一入口识别员工需求，自动调度相应的专业 Agent，或拆解复杂任务实现多 Agent 协同工作。)
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
ServiceNow Sample Modified · Jul 30, 2026 Deloitte Canada Deloitte Canada's rapid growth left employees relying on calls and emails to IT, finance, HR, and risk teams, which created millions in lost productivity and forced staff to navigate too many legacy tools for routine… 18 legacy IT tools eliminated across IT, finance, HR, and risk management AI Agents · AI Search · IT Service Management 
OpenAI Sample Jul 22, 2026 NTT DATA Group NTT DATA Group needed to raise productivity across technical and nontechnical roles, automate repetitive analysis work, and make agentic AI safe enough for broad enterprise use rather than isolated engineering… -99.3% incident-analysis time change from 3 days to 30 minutes ChatGPT Enterprise · Codex · OpenAI Center of Excellence 
Google Cloud Sample Archived · May 10, 2026 Cogedis Cogedis needed to replace unreliable on-premise infrastructure and outage-prone API management, lower infrastructure spend, and create enough operational flexibility to pursue AI services for its accounting and advisory… 66.6% reduction in infrastructure costs Apigee · BigQuery · Cloud Run 
