The Agentic Infrastructure Era

Why It Matters

The tools lower friction for AI agents, accelerating cloud automation while preserving governance, which could reshape DevOps workflows industry‑wide.

Key Takeaways

• LLMs now perform ~20% of infra deployments, targeting >50% by year‑end
• Pulumi launches agent accounts and `pulumi do` for frictionless automation
• Neo agent now integrates with CLI, GitHub, Slack, and scheduled tasks
• New CoreWeave and NVIDIA AICR providers simplify AI‑workload provisioning
• InfraBench created to benchmark agentic infrastructure performance across tasks

Pulse Analysis

The rapid maturation of large language models as coders has turned infrastructure‑as‑code into a natural playground for AI agents. Recent benchmarks show coding accuracy climbing from 33% in 2024 to 86% today, with frontier models like Anthropic’s Mythos approaching 94%. Because Pulumi models cloud resources as objects in mainstream languages—Python, TypeScript, Go, and others—agents can apply the same pattern‑recognition skills that power software development to provision, configure, and evolve infrastructure. This alignment creates a feedback loop: improvements in general‑purpose coding models instantly lift agentic infrastructure capabilities without additional engineering effort.

Pulumi’s latest release removes the remaining friction points that kept agents from operating end‑to‑end. Agent accounts let AI services spin up temporary Pulumi Cloud identities without manual onboarding, while the `pulumi do` command offers a single‑line, declarative way to create, update, or delete resources across thousands of providers. Over 30 new CLI commands expose Pulumi Cloud’s governance, drift detection, and secret management directly to scripts, mirroring the convenience of popular developer tools like the GitHub CLI. The Neo agent, now available in the CLI and as GitHub/Slack apps, brings autonomous execution, scheduled tasks, and read‑only safety modes, enabling teams to delegate routine provisioning, cleanup, and compliance checks to trusted bots.

Beyond productivity, these advances raise the bar for operational control. Pulumi’s InfraBench benchmark will quantify how quickly agents improve on real‑world tasks, while built‑in policy enforcement, audit logs, and structured JSON outputs ensure that every automated change remains auditable and secure. By integrating AI‑native providers such as CoreWeave and NVIDIA’s AI Cluster Runtime, Pulumi also streamlines the most demanding GPU‑heavy workloads. As agents take on a larger share of infrastructure work, organizations can expect faster time‑to‑market, reduced human error, and a new DevOps paradigm where humans design and supervise, and AI executes at scale.