Releases May 2026

Building for agentic infrastructure

As agents take on more of the work of infrastructure, they need tools that meet them where they are. This month, we’ve got new commands, providers, integrations, and more that help humans and agents do infrastructure better.

Agent surfaces — Claude Code, OpenCode, Codex, and Pulumi Neo — around the Pulumi mark

“LLMs are now doing over 20% of infrastructure deployments, up from virtually zero a year ago. We expect this to grow to over 50% this year.”

— Joe Duffy, co-founder and CEO, on the increasing pace at which teams are adopting AI-driven infrastructure

.01

Meeting agents where they are

What do coding agents love more than code? Well-designed CLIs. So we poured some love into the Pulumi CLI to make it even more usable for humans and agents alike.

One-command execution

A new Node.js package enables npx pulumi <anything>-style commands so that agents can discover and run Pulumi commands more easily.

Imperative infrastructure operations

The new pulumi do command enables direct operations like create, read, update, delete, and list across all Pulumi-supported clouds and services.

Agent accounts

Ephemeral Pulumi Cloud accounts allow agents like Claude Code and Codex to spin up Pulumi-managed infrastructure without requiring a human in the loop.

Pulumi Cloud surfaces — stack, env, org, deployment — reachable from the CLI

Pulumi Cloud in the CLI

Dozens of new commands for Pulumi operations previously only available in the browser. Think of it like the gh CLI (which agents love) for Pulumi Cloud.

.02

Neo everywhere you work

With this release, Neo, our infrastructure agent, moves out of the Pulumi Cloud console and into more of the places where the work of managing infrastructure happens.

Neo in the CLI

The new pulumi neo command brings Neo out of Pulumi Cloud and into your terminal, so you can do agentic infrastructure anywhere you can run Pulumi.

Neo GitHub and Slack apps

Now you can @-mention Neo in GitHub issues and pull requests, and in your team’s Slack workspace, to kick off full-context infra tasks wherever you are.

Neo integration catalog

A new integration catalog lets you configure connectors for a growing library of complementary services including Atlassian, Datadog, Honeycomb, Linear, PagerDuty, and Supabase.

Scheduled tasks and read-only sessions

Use Neo to automate recurring infrastructure tasks like keeping providers updated, identifying non-compliant resources, and summarizing infrastructure changes from week to week.

.03

Partnering with the frontier of AI infrastructure

Working closely with companies building the frontier of AI infrastructure, we’re releasing two new partner providers in close collaboration with the teams behind them.

CoreWeave provider

This new provider brings the full power of the CoreWeave platform, the leader in GPU infrastructure, to help teams provision AI workloads. Includes support for all CoreWeave services, including CoreWeave Kubernetes Service (CKS).

NVIDIA AI Cluster Runtime (AICR) provider

This new provider delivers fully functional NVIDIA software stacks atop the underlying cloud provider’s GPU infrastructure. Includes out-of-the-box components like NVIDIA GPU Operator, Kubeflow, NIM Operator, and dozens more.

.04

Educating and measuring agentic infra intelligence

Agent-friendly docs

We now serve all of our documentation, including the complete Pulumi Registry, in easily consumable, agent-friendly Markdown.

Stacked cards representing the new Neo skills

New agent skills

We’ve reviewed and refreshed all of our existing agent skills, and added a new über-skill designed to help agents decide how best to use Pulumi in different scenarios. It’s handy for humans too.

Agent-friendly CLI ergonomics

We’ve made command names, help text, examples, output formats, and exit codes more consistent across the board to help agents navigate and use Pulumi even more efficiently.

InfraBench

A new benchmark for measuring how well agents perform on a wide array of representative infrastructure tasks — like the infrastructure equivalent of SWE-bench.