Video Director produces video of an application using its own tests, and won't render a line of narration the tests no longer back up.
A product demo is usually recorded once and is wrong within a month, or maybe even 5 minutes. The screen it describes gets redesigned, a button moves, and the video keeps showing a product that no longer exists, to exactly the people you most want to impress.
Each app lists the things it can be made to do, such as creating an event or capturing a contact, and each comes with its own checks. A video is a sequence of those actions with narration, edited on screen without touching the app's code.
Every line of narration is attached to a step the tests check. If a line's step no longer exists, or a step has no line, Video Director refuses to render.
A video is made when a person asks for one, not as a side effect of releasing the app. Rewording a line costs a save and a re-render, not a rebuild of the application.
A full walkthrough, a three-minute overview, a clip showing a single feature: separate videos of the same app, each with its own history.
Each save keeps a version you can compare with others and go back to, so the better wording from two drafts ago isn't lost.
When an application changes, the dashboard tells you which app, which step and which video it affects, before anyone watches a video of a screen that no longer exists.
An end-to-end test is a scripted, checked, maintained walk through what a user actually sees. That's most of a product demo, already written and already kept up to date: if the product changes and the test doesn't, the build fails.
Video Director builds on that. Each action is defined once per application, with its checks attached, so a video made from those actions is still checked. If an action breaks, every video that uses it fails visibly, instead of each video quietly keeping its own out-of-date copy.
Because the sequence of steps and the narration are stored as data rather than code, the same walkthrough is designed to be narrated in different voices and languages without being filmed again.
It's the newest application here, and it's still being built.
Working today: seeing every app on the platform, spotting when an app has changed, putting videos together and writing narration, making videos on request, several videos per app, and version history.
Not built yet: approving a specific version, publishing to YouTube, framing phone footage inside a device, and a guided conversation with an AI assistant for making a video.
The one step that can't be undone will need a person. Once a video is published, people have seen it and linked to it, and unlisting it doesn't change that. So publishing is designed to need a person to approve one specific version, and to check that version against the current tests at the moment it's published, refusing if the narration no longer holds up. Nothing is published automatically in the meantime.
It films applications built on AI Application Manager. It relies on each app's list of actions and its tests, so it can't film just any software.
From an app's tests to a narrated demo, and what happens when the app changes.
How it works