AI Agents Surge in Infrastructure Management
Just one year ago, infrastructure executives dismissed the idea of AI agents managing cloud operations. Today, those agents account for 20% of all actions on Pulumi's platform, according to CEO Joe Duffy. He told The New Stack that figure could eventually reach 100%.
“A VP of infrastructure a year ago wouldn't have put ‘agentic AI’ and ‘infrastructure’ in the same sentence,” Duffy said. “Now we see agents driving real workloads, and we're building for that future.”
New Capabilities Tailored for Agent Workflows
On Tuesday, Pulumi is releasing a suite of features designed specifically for agentic infrastructure. The updates aim to let AI agents operate autonomously without human intervention.
A key addition is ephemeral Pulumi Cloud accounts that agents can spin up without manual sign-up. These accounts expire after 72 hours but can be claimed by a human to make permanent. The company also launched an npm package that turns the CLI into a one-shot command: npx pulumi <anything> skips installation steps.
Another new command, pulumi do, lets agents provision single cloud resources — like an Amazon EKS cluster — without needing a full project structure. “The operations are stateful and subject to policy,” Duffy explained, “but we removed the scaffolding that required the agent to infer files and versions.”
CLI Enhanced with Cloud-Feature Parity
More than 30 features from Pulumi's web console are now available in the CLI, including change history, drift detection, audit logs, secrets management, and policy enforcement. The CLI also outputs JSON and structured errors for easy parsing by agents.
Duffy compared the result to GitHub's gh CLI, which agents naturally gravitate toward. “We're giving agents all the tools they need to step through their workflow without a human in the loop,” he said.
Why General-Purpose Languages Give Pulumi an Edge
Duffy noted that large language models are highly fluent in Python, TypeScript, and Go because those languages dominate production code. “But they're less fluent in HCL — Terraform's DSL — because real HCL code is rarely public,” he said.
Pulumi's decade-old bet on general-purpose programming languages is now paying off. “We built for what humans wanted, which turned out to be exactly what agents need,” Duffy added.
Background
Pulumi is an infrastructure-as-code (IaC) platform that lets developers manage cloud resources using standard programming languages. Founded in 2017, the company competes with HashiCorp's Terraform, which uses a domain-specific language (HCL). Until recently, AI agent integration in infrastructure was seen as experimental.
Over the past year, agentic AI has moved from concept to production in many areas, including infrastructure. Pulumi's latest release is one of the first major IaC platforms to embed agent-first design deeply into its tools.
What This Means
The 20% adoption figure signals a tipping point: agents are no longer a fringe experiment but a core part of cloud operations. For DevOps teams, this could mean fewer manual tasks, faster deployments, and fewer human errors.
However, it also raises questions about governance and security when agents act autonomously. Pulumi's policy enforcement and audit logs provide guardrails, but enterprises will need to adapt their workflows. The shift also puts pressure on competitors like Terraform, whose HCL language may be less AI-friendly.
“The next decade of infrastructure belongs to agents,” Duffy concluded. “We're making sure they have the right tools to run it.”