Introducing the Claude Agent step
As agentic coding changes the way teams work, there is a growing need to bring AI agents into the deployment process. Whether for investigating failed deployments or running QA checks, teams currently face a difficult choice: run agents outside of Octopus with heavy manual coordination, or run them as unmanaged scripts that are difficult to audit and secure.
We're now introducing the Claude Agent step (alpha) to provide a better way.
This lets you run a Claude agent as a step inside a runbook or deployment process. By using the credentials and deployment context Octopus already manages, you can run agentic workflows within a sandboxed, scoped, and fully audited environment.
Speakers
John Bristowe
John Bristowe helps engineering teams deliver reliably as a developer advocate at Octopus Deploy. He leads community engagement, creates educational content, and appears regularly on Octopus streams and webinars.
Kit Dergilev
Nikita Dergilev is a visionary product leader and enthusiast with a talent for cultivating a solid product culture within teams and across entire companies. He firmly advocates using data and a customer-driven approach to guide product development.
Robert Erez
Robert Erez is a principal engineer at Octopus Deploy, and a longtime expert in CI/CD, deployment systems, and software delivery.
Additional resources
Claude Agent walkthrough
Hands-on walkthrough: build a sample project where the Claude Agent step rates commit risk and sends high-risk changes to a person for approval.
How the Claude Agent step works
What the step does, why it runs in the deployment, how it uses your own Anthropic API key, and the limitations to plan around before you build on it.
Getting started with the Claude Agent step
How to add the step to a deployment process or runbook: store your API key, install the Claude Code CLI on the worker, and run your first prompt.