“Can you update the website?”
For a school, a simple request might involve a new document, an event change, updated wording, or a photo. The request desk gives those changes a clear path from the school’s owners to a proposed update.
It is a practical AI integration focused on a specific, bounded job.
Connect the tools around the task.
The school submits a request through a private Cloudflare Worker. The request becomes a GitHub issue with the supporting information attached.
A Claude coding agent runs through GitHub Actions, interprets the request, and prepares a pull request. Its permissions and instructions restrict changes to content such as Markdown, photos, and PDFs.
That gives the reviewer something concrete to inspect: a proposed change to the actual website.
Keep approval meaningful.
Corey reviews the proposed changes. A limited autopilot can handle categories of changes only after that kind of request has repeatedly been approved unchanged.
Sensitive decisions still need explicit rules. Student photos, for example, are never published without a release.
The point is a faster path to a reviewable result, with clear boundaries around what the agent can do.
Useful AI is part of a working process.
The integration does not rely on an open-ended chatbot to run the website. It connects a request, a defined task, a proposed change, and an approval step.
That pattern can apply to other business workflows: give the system a specific job, make its work inspectable, and decide deliberately where a human needs to make the call.
