Record video prompts for AI agents
Video prompts are a specialized recording mode in Loom that turns your screen recordings into high-fidelity instructions for AI agents. By combining your voice instructions with on-screen interactions, visited links, and technical metadata, Loom creates a structured "Action Plan" that agents like Rovo or Claude can use to execute tasks.
For example, you can record a video explaining a UI change while navigating between a Figma design and a live website. Loom captures the visual context and your verbal intent, then generates an action plan that an agent uses to draft code or automatically create work items in Jira for agents to act on. This eliminates the need to manually write long prompts or copy-paste multiple screenshots to give an agent context.
Video prompts are currently in open beta.
This mode is available to all Loom users on Business + AI and Enterprise plans. However, the Jira work item creation flow is exclusively available to users with Atlassian-managed accounts.
Video prompts are only available on the Loom Chrome Extension.
Create a video prompt
You can use video prompt mode to capture complex instructions and automatically generate an action log for AI agents to follow.
To create a video prompt:
Open the Loom Chrome extension.
Select the Generate tab from the recorder menu.
Select Video Prompt from the list of enhanced recording modes.
Select Start recording.
Verbally describe the actions you want the agent to take while navigating to the relevant screens, buttons, or links.
Select the Stop button when you finish your instructions.
View the generated Action Plan in the Generate tab on the Loom share page.
If you have an Atlassian-managed account, select Create in Jira to automatically turn specific actions into work items for agents to act on.
Alternatively, copy the prompt to paste it directly into a third-party agent such as Claude.
Video prompt settings and capabilities
This section describes the specific data captured and the limitations of the video prompt recording mode during its open beta phase.
Data capture
When you record in video prompt mode, Loom captures enhanced metadata to ensure AI agents have full context. This includes:
Transcript: A full text record of your spoken instructions.
Keyframes: Visual snapshots of the screens you shared.
Interaction events: Data on where you clicked or hovered during the recording.
Visited links: A log of the URLs you navigated to while recording.
Recording limits
Video prompts have specific constraints to optimize them for AI processing:
Time limit: Recordings are limited to a maximum of five minutes. A countdown timer appears during the recording to help you track your time.
Camera: The camera is automatically turned off in video prompt mode to focus the agent's attention on the screen content.
Action Plan
The video prompt summarizes your video into an Action Plan of discrete executable tasks. Each action in the log typically includes:
A concise title and bulleted description.
A transcript quote for verbal context.
A visual reference or keyframe from the video.
Timestamps indicating when the action was discussed.
Loom runs the video file through an LLM to generate the Action Plan.
この内容はお役に立ちましたか?