The limits, stated plainly
I'm not selling you magic. AI is excellent at executing. You still need to be good at deciding.
Don't record screens with sensitive data, credentials or confidential information. Everything that ends up in the Clip ends up, transcribed and in frames, in the link you pass to the agent.
The workflow, ready to replicate
Here's the method, in 6 moves.
- 1
Sign up on Clips, connect storage (Builder.io's free tier is enough), then install the official Chrome extension: pin it, sign in, and grant microphone and camera permissions.
- 2
Plan the recording: keep the goal in mind, show only the necessary screens.
- 3
Record max 45-60 seconds: the reference to replicate + the target style, describing the problem out loud.
- 4
Generate the "for agent" link (not the one for a person).
- 5
Open it in Claude Design: new project, paste prompt + link into the chat. Even better with a design system already set up.
- 6
Download or publish (export, or connect to a platform like Vercel), then check and refine: the human eye remains the final check.
And the prompt to paste into Claude Design, ready to use:
Here's a video recorded with Clips: [PASTE THE AGENT LINK HERE].
In the video I show a reference linktree and the style of the platform
[DESCRIPTION OF YOUR PLATFORM OR SITE'S STYLE]. Replicate that linktree
using the design system I gave you. Keep the colors, typography, and
spacing from the design system. Don't add sections you don't see in the video.
Show me the result ready to download. Replace the parts in [brackets] with your own details. Clips solves a real problem: giving agents the eyes and ears they were missing. Claude Design turns those 45 seconds into a real page. The rest — the judgment, the style, the "this yes, this no" — stays your job. And that's how it should be.