Siemens and Google Cloud automate legacy code modernization

Industrial software modernization has long been hindered by the sheer scale and complexity of legacy codebases. Siemens, a leader in industrial automation and software, partnered with Google Cloud to address this challenge by creating an AI-driven system capable of navigating and refactoring massive, decades-old code repositories. The result, Knowledge Fabric, leverages agentic workflows and knowledge graphs to automate parts of the software development lifecycle, reducing manual effort while preserving system reliability and compliance standards.

How Knowledge Fabric works

Knowledge Fabric is designed to tackle the unique challenges of industrial-grade software, where codebases often exceed hundreds of millions of lines and span multiple decades of development. Traditional AI coding assistants lack the contextual depth required to understand the intricate relationships within such systems. To bridge this gap, the system combines Google Cloud’s Spanner Graph, the Google Agent Development Kit, and the Gemini API to model the structure of the codebase as a knowledge graph. This graph maps connections between code, documentation, and functional requirements, enabling AI agents to traverse dependencies and retrieve precise context for refactoring tasks.

The system employs three core methods to navigate the codebase: Graph Query Language (GQL) for structural queries, vector embeddings for semantic understanding, and full-text search for keyword-based retrieval. By integrating these approaches, Knowledge Fabric can answer complex queries, such as identifying which functions require updates when a specific module is modified. This level of precision is critical for industrial software, where unvalidated changes can introduce operational risks.

The agentic workflow: "Slicing the elephant"

A key innovation in Knowledge Fabric is its agentic workflow, which breaks down large, ambiguous tasks into smaller, manageable subtasks. Dubbed "slicing the elephant," this approach assigns specialized agents to distinct phases of the refactoring process. For example, a search agent explores the code graph to gather context, while an architecture impact agent analyzes proposed changes for potential side effects. Other agents handle user story creation, task breakdown, and coding implementation. Each agent operates with full access to the knowledge graph, ensuring that changes are traceable and verifiable.

Human oversight remains integral to the process. Engineers review and approve each step, ensuring that AI-generated outputs meet Siemens’ strict quality and compliance standards. This hybrid approach allows developers to focus on higher-value problem-solving while reducing the time spent on repetitive tasks.

"By slicing the elephant — breaking complex refactoring jobs into smaller, agent-led tasks — we observed a significant productivity increase. We essentially gave the AI the roadmap it needed to navigate the complexity." — Alexander Lomakin, Project Lead at Siemens

Pilot results and industry implications

In a pilot project migrating legacy control panels to web-based interfaces, Knowledge Fabric reduced overall coding effort while maintaining system integrity. Tasks that previously required senior engineers to spend days analyzing dependencies now take significantly less time. The system’s ability to link code changes to functional requirements and documentation ensures that modernized software adheres to industrial quality standards, a critical factor for systems with lifecycles spanning 15 to 20 years.

The collaboration between Siemens and Google Cloud demonstrates how AI can extend beyond boilerplate code generation to address the complexities of legacy modernization. For industries reliant on large-scale, mission-critical software, such tools could accelerate digital transformation while mitigating the risks associated with manual refactoring.