From Infrastructure as Code (IaC) to AI: Adopting OpenTofu and ChatOps

LY Corporation's Site Reliability Engineering (SRE) team has detailed its journey in adopting Infrastructure as Code (IaC) with OpenTofu and Terragrunt, marking a significant departure from previous disparate and manual infrastructure management practices. The team faced the challenge of consolidating approximately 1,500 infrastructure resources, including virtual machines, load balancers, and DNS records, which were previously managed through various methods like the Verda dashboard, scripts, or wiki procedures. The core of their strategy involved establishing a Git-based operations automation (GitOps) workflow. This ensures that all infrastructure changes are declared as code in Git repositories, providing a single source of truth, version control, and a comprehensive audit trail. This approach has transformed infrastructure management into an engineering artifact, allowing for transparent review processes through pull requests (PRs) and eliminating reliance on individual memory for change history. A key aspect of their IaC adoption was the modularization of frequently used configurations. By creating modules for VMs, LBs, and monitoring alerts, different teams can now provision their infrastructure by simply passing specific values to these standardized modules. This not only promotes consistency but also allows for versioning of modules, giving teams control over when to adopt new versions. Furthermore, the initiative extends beyond traditional IaC by integrating AI agents into the GitOps pipeline. Recognizing that infrastructure declared as code is machine-readable, LY Corporation has developed a system where AI agents can generate OpenTofu code from natural language requests made in Slack. For instance, a request like "add two instances for service A" can trigger the AI to reference existing code patterns, generate the necessary Terragrunt code, and open a PR. This innovation aims to democratize infrastructure changes, allowing teams without deep HCL proficiency to participate in the IaC workflow while maintaining the integrity and traceability provided by GitOps. The migration, though initially challenging, proved successful. The SRE team now acts more as reviewers, empowering development teams to directly manage their infrastructure through code. This shift has not only improved operational efficiency and consistency but also laid the groundwork for future advancements in AI-driven infrastructure automation.