WEB-03 / WEB AGENT

An AI website agent.
Changes you can review.

A website specialist that turns an approved task into a tested, reviewable change in your repository.

THE MISSION

Web Agent.
A defined role.

The planned Web Agent closes the gap between a recommendation and a working website update. Every task should have a small scope and an obvious way back.

01

Edit within the boundary

Use existing components and content schemas. Avoid unrelated design changes or new dependencies.

02

Validate the result

Build the site, check the links, and inspect the pages at the sizes people will actually use.

03

Package the change

Describe the diff and leave it in the approval queue. Deployment is a separate permission.

INPUTS

What it needs.

  • An approved task and allowed file paths
  • Repository access scoped to the job
  • The existing tests and deployment rules
OUTPUTS

What comes back.

  • A branch and focused change set
  • A validation report and preview
  • A revision that can be rolled back

THE BOUNDARY

Permission is
part of the brief.

No changes to secrets, billing, DNS, authentication, or legal content without a separate explicit authorization. Passing a build is not proof that the copy is accurate.

Read the control model →

START SMALL. BUILD FROM THERE.

Put a specialist
on the right problem.

Share the first job you would give this agent. The initial scope should be small enough to review and useful enough to matter.

Explore early access

BEFORE THE FIRST WORKFLOW

Questions worth
asking first.

Which websites are you designing the Web Agent for first?

The initial focus is a Git-managed website where a change can be inspected as a diff and tested before deployment. Other platforms remain subject to the integration roadmap.

Can website changes be reversed?

The proposed workflow keeps the previous revision, a diff, and a deployment record. Rollback is a requirement of the design; the demo on this site does not deploy or revert a real website.