The Importance of Context in AI x Infrastructure: Your Infrastructure is a Graph
TalkAI in production
2025-12-10 | 09:00 AM - 09:00 AM | Auditorium Niv-2
Information
AI tools are getting better at generating code, including for Infrastructure-as-Code. But when it comes to managing real infrastructure, they often miss critical details. Seemingly correct suggestions can cause errors, ignore dependencies, or lead to risky changes. The reason is simple: AI works without context. Infrastructure isn't just code. It's a system made up of resources, environments, services, and the complex relationships between them. Most LLMs have no visibility into these connections. As a result, their help remains superficial and doesn't hold up in production. In this session, we'll share our experience: what happened when we tried to use AI for real infrastructure tasks, why it failed initially, and how we succeeded by providing it with the context it lacked. The key shift was to think of our infrastructure as a graph. By mapping the dependencies between services, resources, configurations, and environments, we built a model that AI could query and understand with much greater precision. We'll explain how this graph was built from existing sources like AWS and Terraform. Once the data was connected, we integrated a natural language interface via Slack, allowing engineers to ask real questions like, "Who owns this resource?" or "What were the latest changes to RDS in production?" This presentation will include a live demo of the system in action, as well as practical tips for building something similar. We'll detail which data sources proved most useful, how we structured the relationships between components, and how we ensured the reliability of the graph without creating too much operational overhead. Attendees will walk away with a deep understanding of why context is the missing element in AI-assisted infrastructure tools, and how modeling infrastructure as a graph can transform what AI can do.
