Atlas: AI inside the Magento 2 admin

Context

Coding agents such as Claude Code and Codex can now make real changes to real websites. For a developer that is a large speed-up. For a store owner or a merchandising team it is tempting and risky: the same tool that can fix a typo can, unsupervised, reach anything else on the server, and an unreviewed change can take checkout down in front of customers.

The challenge

Let the people who run the store change it themselves, in plain language, on development, staging or production, without handing an AI agent the keys to everything and without risking downtime. Every change had to be reviewed, reversible and contained to one site.

What I did

I built Atlas, a platform that puts AI inside the admin of Magento 2 (and other sites), with the same guardrails I used when I owned the release process for 18 storefronts.

• Plain-language changes from the admin. A super admin describes what they want in a chat panel inside their own admin; Claude Code makes the change.
• A second model reviews every change. Codex reviews Claude Code's work before anyone is asked to approve it.
• A person signs off. Nothing goes live on its own: after review, a human uses "push to staging", checks it there, then "push to production".
• Zero-downtime deploys. Hard to do on Magento 2, and Atlas does it for nearly every release; about 1 in 100 uses a short, planned maintenance window instead.
• One sandbox per site. Each site gets its own workspace; a scope guard blocks the agent from anything outside that site's code, and no client can affect another.
• Every change is reversible. Each change is a git commit with a one-click undo, so the history is complete and a bad change is one click from gone.

Outcome

Atlas runs in production on multiple enterprise eCommerce sites, each fully isolated from every other project. Owners and super admins ship their own changes through review, staging and production with zero downtime (about 1 in 100 releases uses a short, planned maintenance window). For a commerce team the lesson carries straight over: I use AI as leverage every day, and I know how to fence it in with review, staging and rollback.