Cursor 3 vs Codex vs T3 Code: agent workflow comparison

This comparison examines Cursor 3's agent window alongside the Codex app and T3 Code. The speaker describes Cursor's older interface as cluttered and difficult to manage across projects. The new window puts projects and chats at the center, with an AI agent handling tasks within each project. Cursor also keeps files, a terminal and a browser available, options the speaker values when checking the work.

T3 Code is the speaker's preferred coding assistant interface in this comparison. They describe its open source interface as a way to access Claude and Codex from one place. That preference concerns the workflow rather than measured model performance. The discussion does not establish support for local inference or offline operation.

The speaker also examines Cursor's agent settings, model selector, plugins, rules and skills, and notes MCP servers and environments in Codex. They miss terminal-style tabs and simultaneous windows, then propose a zoomable canvas for multiple T3 Code sessions as an idea, not an existing feature.

Cursor's design mode remains unresolved: the speaker finds a Shift Command D shortcut, but pressing it does nothing in their setup. They question whether the feature belongs in Cursor 3 and do not demonstrate it working.