Cursor’s Annotate turns on-screen notes into agent instructions

What happened?

Cursor announced Annotate on October 27 as an addition to Design Mode. Its official post describes visual prompts made by editing elements, drawing on the page and reviewing designs at different screen sizes. This is an update to the existing workflow, not a separate canvas product.

The documentation places Annotate in the browser inside the Agents Window, where a running app becomes the surface for instructions. Users can select one or several elements, mark a region, add drawing markup, edit text in place or include CSS changes. Notes can be sent individually or collected into a batch.

Each note supplies element details and a screenshot so the agent can locate the relevant code. The docs specify Cursor 3.24.18 or later for Annotate; earlier versions use a Design Mode pen icon. These are documented features, not results from a hands-on test.

Source: Cursor ↗ · Source: Cursor ↗

Why does it matter?

Our analysis: interface feedback often loses precision when translated into chat. Pointing at an element can make a request less ambiguous than describing its position, while a screenshot preserves the visual relationship that prompted the change.

That improves the instruction channel, not necessarily the implementation. Developers still need to inspect the resulting code and test the interface. A useful evaluation would ask whether visual notes reduce misdirected edits without making it harder to review what the agent changed.