Transcript
you could describe the interface you want and have it built for you in seconds? That's the idea behind canvases. A canvas is a fully customizable surface, and the best part is you don't build it yourself, you ask the agent and it builds it for you. By the end of this video, that'll feel completely natural. Let's do it live. ♪♪♪ A canvas, sometimes called a canvas extension, is a completely customizable interface, a surface you and the agent share. And when I say customizable, I really mean it. A canvas can be almost anything you can picture. A Kanban board, a triage board, a release checklist, a dashboard, a custom form, even a spreadsheet. It's not a fixed set of panels you're stuck with, it's a UI shaped to your exact workflow. And it's bi-directional, which is the key word. The agent updates the canvas as it works, and you use its own controls, buttons, cards, and filters to change things directly. You're both looking at the same live surface, like a shared whiteboard you can both write on at once. Here's the part that makes canvases so approachable. You don't design or code one by hand, you just ask and the agent builds it for you. You just use the create canvas skill and describe in plain English what you want, what people should be able to do, and what the agent should be able to do. For our Tailspin Toys work, I'll ask for a release notes canvas to include any new feature work, including what we did in our sessions. The agent reads that and builds the whole interface, then opens it in the right side panel, no files to write, no layout to fiddle with. Now we have a custom tool I described in a single sentence sitting right there, ready to use. And you're not locked into that first version, you can shape it into almost anything. Just keep describing what you want, add a column, add a filter, pull in your open pull requests, or turn the whole thing into a checklist for your day. The agent revises the canvas to match. Because the entire UI is generated from your description, there's really no fixed menu of options. If you can describe the workflow, you could probably build a canvas for it. And if you'd rather not start from a blank slate, the community has shared ready-made canvases and collections like Awesome Copilot that you can drop in and then customize the very same way. However it starts, your canvas is saved as an extension in your project to share with the team or in your personal settings just for you. So it's there next time. The whole point of a canvas is people and agents working on the same surface at the same time. When you click a button or move a card, the shared state changes for both of you instantly. There's no send step, no waiting for sync. And it works the other way too. You can ask the agent to use the canvas's own capabilities like adding a card or moving work. So it acts through this very same controls you do. That symmetry is what makes it feel like collaboration rather than command and wait. So that's the idea. A canvas is a fully customizable interface and the agent builds it for you from a plain English description. Almost any workflow you can describe, you can turn into a surface you and the agent share and steer together. If it still feels abstract, that's totally normal. It clicks the moment you make one. So try it. Open a session, run create canvas, and describe a simple board for something you're working on right now. You can also try out some community-made extensions from awesome-copilot.github.com.
