Siemens' 'Salami Slicing': Advancing Agentic Workflows for Industrial Software Development

Siemens has struggled with the limits of existing AI tools alone for software modernization due to its vast legacy code spanning hundreds of millions of lines and fragmented documentation. Standard RAG (Retrieval-Augmented Generation) approaches had the problem of failing to sufficiently reflect the hierarchical structure and relationships of code.

To address this, Siemens and Google Cloud built an AI system called Knowledge Fabric. This system uses Spanner Graph to model the structure of the codebase and directly connects code snippets with requirements from design documents.

The key features of Knowledge Fabric are as follows:

• Modeling relationships between code and documents through Spanner Graph
• Agent-based workflows leveraging the Gemini API and Google Agent Development Kit
• Performing structural queries using GQL (Graph Query Language)
• Domain-aware understanding that goes beyond simple keyword matching

Through this system, Siemens is enabling engineers to move away from repetitive tasks and focus on high-value problem solving, automating the development lifecycle of complex industrial software.