How China’s Tech Giants Are Mastering AI Agents in Software Development

From Copilot to Full‑Stack Coding Agents

Developers once marveled at GitHub Copilot’s code‑completion, but industry leaders now view agents as proactive task‑finishers. At the 2025 AiDD Digital Development summit, experts from Alibaba, Tencent, Baidu, Kuaishou and others emphasized the shift from a "Copilot era" to a "Coding Agent era," where agents can design, implement, test, and even fix code autonomously.

Agent Impact Across the Software Development Lifecycle

1. Requirements – From Text to Knowledge Graphs

Traditional requirement analysis is often ambiguous. ZTE’s knowledge‑graph‑driven requirement engineering uses agents to simulate multiple roles (product manager, developer, tester, ops) and to accelerate draft creation from three days to half a day, improving quality.

2. Development – Understanding Code, Not Just Writing It

Agents are evolving from IDE plugins to standalone environments. For example, CodeRider’s Vibe Coding offers a AI‑optimized IDE where natural‑language prompts generate project scaffolding, code, and dependency management. JD’s JoyCode IDE 2.0 demonstrates agents that read legacy code, generate unit tests, and perform code reviews for performance and security issues. Kuaishou’s "Intelligent Co‑creation" system positions the agent as a neural hub for large R&D teams, enhancing both individual productivity and team collaboration.

Alibaba’s Mobile‑Agent can interpret UI screenshots, understand mobile UI structures, and automatically generate interface tests, dramatically boosting mobile development efficiency.

3. Testing – From Passive Validation to Active Assurance

ZTE’s AI‑driven unit‑test generation analyzes code logic, creates test cases, identifies coverage gaps, and can even auto‑repair failing code. Tencent’s WeChat Pay testing agent generates API test cases and simulates failure scenarios such as network latency and data anomalies. Ant Group’s multi‑modal GUI agent automates cross‑platform UI element recognition, intent‑driven script generation, and adapts to UI changes without manual effort. Baidu’s front‑end testing pipeline evolves from test‑case generation to self‑healing, end‑to‑end intelligence.

4. Operations – AIOps as Intelligent Engineers

Agents now act as smart ops engineers: real‑time metric monitoring, early fault warnings, rapid root‑cause analysis (reducing diagnosis from hours to minutes), and automated remediation with human approval. Qunar achieved a 75% efficiency gain using AI ops agents. Huawei’s large‑model training clusters leverage agents to correlate hardware metrics, logs, and network topology for swift fault diagnosis.

CI/CD pipelines are also AI‑enhanced: Beike’s AI‑native development loop lets agents assess code change impact, select relevant tests, analyze build failures, suggest fixes, and even optimize pipeline configurations. Xiaohongshu uses agents to interpret business semantics for smarter observability beyond traditional metrics.

5. Data & Knowledge Management – The Agent’s Brain

Knowledge graphs enable agents to truly understand business context. Huawei builds product‑knowledge graphs for rapid business comprehension. Nanjing Kogei’s "Agentic Knowledge Base" lets the knowledge base self‑organize and update based on agent usage. OPPO’s code‑knowledge graph links code, tests, requirements, and bugs, giving agents a holistic view of a project’s DNA.

Multi‑Agent Collaboration

Future trends show stronger multi‑agent cooperation. iFlytek’s system coordinates requirement, development, testing, and deployment agents via shared knowledge graphs and messaging, allowing, for example, a testing agent to notify a development agent of a bug with reproduction steps. Tencent’s cross‑platform agent system uses a unified specification language so agents on PC, mobile, and web can "speak the same language." Internationally, CAMEL‑AI envisions a workforce of specialized agents collaborating with humans.

Practical Deployments by Major Vendors

Alibaba’s Qoder spans IDE, CLI, plugin, and cloud layers, and hosts hackathons for custom AI tools. Tencent tailors agents to vertical scenarios such as WeChat Pay API testing, PCG performance agents, and large‑model evaluation agents. ByteDance focuses on data‑driven efficiency metrics—time saved, bug reduction, cost cuts. Kuaishou promotes organization‑wide AI paradigm shifts, integrating agents into tools, processes, and culture.

Challenges and Open Questions

Experts acknowledge limits: complex business logic remains hard to grasp, multi‑module coordination needs global view, and non‑functional requirements (performance, security, maintainability) require experience. Context window size is a bottleneck for large codebases, prompting research on context engineering (Baidu, JD). Trustworthiness of AI‑generated code—correctness, security, data leakage—remains an open research area, discussed at the "Trusted AI Safety Engineering" forum with contributions from Renmin University, Zhejiang University, and the Chinese Academy of Sciences. Finally, compute cost is a practical concern; cloud providers showcase model training and inference optimization to balance performance and expense.

Future Outlook

Agents will evolve from coding assistants to "digital employees" covering the full software lifecycle, from requirement to deployment. Specialized expert agents will emerge for front‑end, database optimization, and security auditing. The shift from tools to partners will augment human capabilities, similar to the transition from horse‑drawn carriages to automobiles. Industry‑specific agent solutions will grow in finance, automotive, and government sectors, exemplified by NIO’s vehicle data analysis, Puyuan Info’s government AI, and New Hope Financial’s large‑model applications.

Conclusion

AI agents are driving a revolutionary yet evolutionary change in software development. While they dramatically boost efficiency, they are not magical replacements for human creativity and judgment. Embracing agents as collaborative teammates will enable developers to build better software faster.