How Siemens "slices the elephant," advancing agentic workflows for industrial software development

Modernizing large-scale industrial software has been compared to rebuilding a jet while flying it. Siemens faced this reality with software that powers factories, energy grids, and transportation networks worldwide. The company's domain expertise spanning factory automation, process automation, energy infrastructure, and intelligent transportation could not be replicated by off-the-shelf AI solutions, yet legacy code presented an immovable barrier.

The core problem was that hallucinated or unvalidated AI-generated changes are operationally unacceptable in industrial contexts. Unlike consumer software, failures can affect critical infrastructure and public safety. This demanded a fundamentally different approach than existing coding assistants offered.

According to Agata Gołębiowska, technical lead at Google Cloud, the fundamental issue was that code isn't merely text. It has inherent structure and relationships that must be preserved. When flat vector databases flatten code into embeddings, they destroy the semantic and structural information essential for reasoning about industrial systems.

Standard coding assistants lacked the contextual depth required to navigate complex, multi-layered industrial codebases. The gap between what these tools could do and what Siemens needed became the catalyst for developing a specialized solution.

The Knowledge Fabric agent goes beyond keyword matching to understand relationships between assets across the entire software ecosystem. Rather than treating code, documentation, requirements, and historical records as separate silos, it maps connections between domains. This allows agents to link specific code snippets directly to requirements in design documents, trace changes back through Jira histories, and connect current implementation to legacy decision-making.

By ingesting the entire software ecosystem into an intelligent agentic system equipped with custom knowledge graphs, the solution enables autonomous agents to reason across decades of accumulated knowledge. Franz Menzl, senior vice president of product creation excellence at Siemens, described the approach: 'We aren't just helping developers optimize their development time; we are enabling autonomous agents to reason across the past to build the future.' The goal is freeing engineers from repetitive work so they can focus on higher-value problem solving and customer-facing innovation.

The pilot program migrating existing frontiers to web-based interfaces provided proof that the Knowledge Fabric approach could deliver practical value. By automating the most time-consuming aspects of code understanding and navigation, the system allowed skilled engineers to redirect their expertise toward higher-leverage work. This shift from maintenance and investigation to innovation represents the fundamental value proposition of agentic AI in industrial contexts.

The architecture deliberately separates structural querying from semantic understanding. Agents use Graph Query Language for precise navigation of relationships, while embeddings generated for every node enable semantic search and reasoning. This dual approach ensures both accuracy and contextual understanding, which is essential for industrial systems where mistakes are costly.

The solution represents a significant departure from earlier AI-assisted development approaches. Instead of asking what traditional tools could do and then trying to fit the problem to the tool, Siemens and Google Cloud engineered the infrastructure to match the requirements of the problem. The result is an agentic system designed from the ground up for industrial-scale, mission-critical software modernization.

The emphasis on traceability and explainability distinguishes Knowledge Fabric from consumer-oriented AI development tools. Every recommendation an agent makes can be traced back through the knowledge graph to source documents, previous implementations, and requirements. This transparency is not merely a feature but an operational necessity when software controls power grids, factory automation, and transportation systems.

Siemens' approach with Knowledge Fabric suggests a broader pattern for how enterprise AI can evolve. Rather than applying generic large language models to complex industrial problems, effective solutions require specialized knowledge representation, domain-aware agents, and strict frameworks for verification and traceability. This represents a maturation of AI in enterprise software development, moving from narrow automation tasks to comprehensive reasoning across entire software ecosystems.