Canvas first, desktop still matters

The Matrix shell has two familiar ideas living side by side: a desktop and a canvas.

The desktop gives people an immediate model. Apps open as windows. Terminals can sit next to previews. The dock and file browser are recognizable.

The canvas is where the product is heading.

Why canvas works for agent supervision

Agent work is spatial. A coding session has a repo, a terminal, a preview, a pull request, a log stream, and notes. A research session has source material, drafts, browser state, and follow-up actions.

Putting those surfaces on a canvas makes the system easier to scan. The user can leave the active context arranged as a map instead of compressing everything into tabs.

That matters when work runs in the background. Returning to a canvas should feel like returning to the state of the project, not reopening a collection of disconnected panels.

Familiar mechanics still matter

Canvas-first does not mean novelty for its own sake. Matrix still needs predictable window behavior, keyboard focus, persistent positions, stable icons, and clean app launch rules.

The product should feel calm under repeated use:

• click to open,
• click the same spot to close,
• dismiss with Escape,
• preserve layout across reloads,
• keep panels overlaying instead of pushing content around.

These rules are not cosmetic. They make the workspace reliable enough to supervise long-running agents.

The shell is a renderer

The deeper architectural point is that neither canvas nor desktop owns the system.

Matrix keeps the core headless. The shell discovers apps and renders state. The same workspace should eventually be controllable from web, mobile, CLI, and messaging channels.

Canvas is the flagship visual expression because it matches how agent work spreads across artifacts. Desktop compatibility stays important because people already understand it.

The best shell is the one that lets the computer keep working while the human stays oriented.