Token Efficiency vs Cognitive Efficiency: Choosing IaC for AI Agents

In a comparison of infrastructure code generation by AI agents using Claude Opus 4.6 and GPT-5.2-Codex, the study explores the trade-offs between using HCL and Pulumi TypeScript. HCL, being more compact, offers token efficiency, reducing output tokens by 21-33% in simple resource generation, thus lowering costs. However, its advantage diminishes in scenarios requiring code refactoring and deployability, where Pulumi TypeScript shows better performance. Pulumi consistently produced deployable code with fewer refactoring repairs, particularly when used with Claude Opus, which resulted in a 41% lower total pipeline cost compared to HCL. This success is attributed to Pulumi's use of familiar TypeScript refactoring patterns and tooling that AI models are well-versed in, contrasting with the domain-specific constructs required by HCL. As the industry shifts towards more iterative and maintainable infrastructure processes, the deployability and refactoring capabilities of Pulumi make it a more effective choice for AI-driven infrastructure tasks.