On October 26, 2026, NetEase Intelligent Enterprise officially launched CodeWave SDD “AI Software Delivery Team,” marking a shift of AI Coding from a tool that assists individual developers to an organized, engineering‑driven end-to-end delivery system. The product comprises collaborative teams of specialized Agents—such as requirements analysts, architects, development engineers, and test engineers—covering the entire software delivery pipeline, from interpreting raw requirement documents and generating structured specifications, to designing technical solutions, implementing code, and performing automated testing and deployment.
This evolutionary context stems from real‑world industry bottlenecks: Faros AI’s analysis of data from over 10,000 developers and 1,255 engineering teams reveals that, while AI adoption has significantly boosted the number of PR merges (+98%), code review time has simultaneously increased by 91%, and there is no significant positive correlation between AI adoption rates and enterprise‑level delivery efficiency metrics such as lead time and defect rate. This indicates that simply accelerating the coding phase cannot overcome systemic challenges like unclear requirements, inefficient cross‑functional collaboration, and insufficient understanding of legacy systems.
The key technical design of CodeWave SDD comprises two aspects: first, Spec-Driven Development, in which the Requirements Analyst Agent proactively identifies multimodal information such as text, tables, and flowcharts, initiates human clarification for ambiguous points, and employs methods like EARS to generate structured specification documents (Specs), ensuring strict alignment across subsequent stages; second, an enterprise‑level Harness engineering environment—where approximately 70% of R&D resources are allocated—providing unified runtime constraints, task timeout monitoring, Agent communication assurance, and consistency validation of stage deliverables, thereby supporting AI teams in executing complex, multi‑hour tasks in production environments with stability and traceability.
The system has been validated through real-world projects both within NetEase and at over 100 enterprises, claiming to deliver a roughly 3–5x increase in productivity. Approximately 80% of its customers are software development service providers, serving scenarios such as management systems, mobile applications, and data analytics. The platform supports on-premises deployment and can integrate with enterprise knowledge bases, proprietary models, and security‑checking mechanisms. For upgrading existing systems, it leverages AST syntax trees and LSP semantic services to construct code maps, ensuring the safety of iterative changes. According to disclosures, the first‑pass pass rate for P0‑level test cases is about 90%, while for P1/P2 levels it stands at around 80%.