Platform-aware AI agents: Teaching LLMs to respect your infrastructure boundaries

The speaker tackles one of the fastest-growing pain points in enterprise engineering: AI coding agents that generate Terraform, Kubernetes manifests, and CI pipelines with no knowledge of organizational platform standards. The result is a flood of non-compliant pull requests that platform teams must review, reject, and remediate. The speaker presents a practical architecture for making AI agents platform-aware by feeding them organizational context through retrieval-augmented generation, tool-use integrations with internal developer portals, and policy-as-code validation loops. Key takeaways include:

• Why AI-generated infrastructure code is creating a new category of platform drift
• Patterns for injecting platform context into AI agent workflows via MCP servers and RAG
• Building automated compliance gates that catch AI-generated violations before merge
• Real results from enterprises that reduced non-compliant AI PRs by over 70%