The Agentic Infrastructure Era

Pulumi CEO Joe Duffy argues that infrastructure as code is the natural substrate for AI agents because it maps cloud resources into programming languages LLMs already understand, enabling a verifiable, self-correcting agentic loop. He reports that LLMs now handle 20% of Pulumi infrastructure deployments, up from near zero a year ago, with a target of 50%+ by year-end. New launches include: agent accounts for ephemeral CLI access from tools like Claude Code and Codex, a `pulumi do` imperative command for single-command CRUD operations, `pulumi neo` CLI for local agentic infrastructure sessions, Neo integrations with GitHub and Slack, scheduled tasks, CoreWeave and NVIDIA AICR provider partnerships, agent-friendly docs and CLI ergonomics, and InfraBench — a new SWE-bench-style benchmark for infrastructure agent performance.