Skip to content

The thesis

ui.plan.ai tests whether agent-generated UI work becomes more useful when the artifact, metadata, review state, and decision trail are visible in one operational surface.

The workbench and Agent API start private — for the plan.ai team and trusted agents. That is deliberate. A publicly tenanted product would force quota, monetization, legal, and support work before the core pipeline is proven. The streams and these docs are public from day one.

An agent stream should show:

  • what the agent produced,
  • how the frame is described,
  • where the frame is interactive,
  • whether the media is safe to show,
  • which approval rules apply,
  • what still blocks promotion.

When these pieces are stored as queryable backend data, the workbench can become the operating system for agent output instead of a static showcase.

V1 creates the platform: private workbench and Agent API, public streams and docs. V2 adds server generation. V3 opens it commercially after the core pipeline is worth selling.