An agent run for updating the pricing page hero, with its three-step plan written out, nothing ticked off, and an Approve plan button waiting under it AGENT RUN 3 STEPS Update the pricing page hero Draft the headline and subhead Swap in brand-kit/hero-2026.jpg Queue the page for review AWAITING APPROVAL Edit plan Approve plan

Your stack. Your rules. Everyone moves faster.

What the agent can do

  • Eleven named skills

    Four write, three propose a change set, four are read only. Draft an entry, backfill SEO metadata, rename across content, audit, and the rest.

  • A plan card first

    The goal, the steps and the boundaries, each boundary with a shield. Plus a usage estimate, or the name of your own key.

  • Review every document

    One card per document with a real checkbox and a change count. Confirm all, discard, or apply only the ones you ticked.

  • An undo that is honest

    It removes only what is still a draft. Then it tells you how many changes it kept, and that they were published or edited.

  • Stop it mid run

    A Stop control halts a run that is already going. The note it leaves reads stopped before anything was changed.

  • Three speed tiers

    Lite, Balanced and Max. You never see a model name. A request above the workspace ceiling is clamped down to it, not rejected.

Agent run questions.

What teams ask before they let an agent near their content.

  • No. An agent writes drafts. The highest thing it can do is request a publish, and a person decides.

  • The goal, the steps the agent intends to take, the boundaries it has to stay inside, and the cost of the run.

  • Nothing runs. No content is written and no draft is created, because the run never starts.

  • Yes. A run writes drafts, you review them together, and you can confirm all of them, discard all of them, or apply only the ones you ticked.

Watch an agent plan a change

Content OS for modern websites.