The Canvas Is Not a Feature

I cancelled my ChatGPT subscription last month and moved fully to Claude. No regrets. Genuinely.

But there is one thing I miss. And it's worth being precise about what it is, because it's easy to write off as personal preference.

The canvas.

Not because it's a clever feature. Because it changes how you think.

When a draft lives inside a conversation, every edit is a negotiation. Paste this. Reframe that. Now do it again with the new context. You are always describing what you want. The AI is always responding to that description. The work happens at one remove from you.

When a draft lives on a shared surface, something shifts. You stop directing and start doing. The reasoning, the questioning, the editing all happen on the same piece of work, not around it. The conversation becomes the work rather than a series of instructions about the work.

That is a different cognitive mode. Not a feature preference. A fundamentally different way of thinking alongside a tool.

Why the distinction matters

Most AI interactions are request-response by default. You describe what you want, you get an output, you react to the output. Even with strong context and a long conversation, there is a structural gap between you and the thing being made.

Collaborative canvases close that gap. The artefact becomes the shared object. You are both oriented toward the same thing and working on it directly. That changes the nature of the collaboration from issuing instructions to genuine participation in the work.

It is a small structural difference that turns out to matter quite a lot in practice.

Where this leaves things

Claude is better for me in almost every other way. The thinking is sharper. The context holds. The outputs are closer to what I actually need without as much course correction.

Artefacts help. When Claude produces a document in a dedicated pane, it gets partway there. You can ask for edits, refine the structure, rebuild sections. But the artefact is still, fundamentally, something Claude produced that you are then reacting to. The collaborative surface model, where you are both working on the same thing simultaneously, is a different proposition.

Making Artefacts genuinely collaborative, not just editable, would make them something else entirely.

I switched because the core capability is better. I miss a mode, not a feature. That distinction is exactly the point.