Modes: Agent, Ask, Plan, and Debug

An agent that can edit any file and run any command is exactly the right tool for some work and exactly the wrong one for the rest. Cursor’s control for that is the mode: not a different model and not a different chat, but the same Cursor reading the same project at a different posture - how far it’s allowed to go before you look. On a one-line fix the default is fine. The moment a change reaches into code a dozen call sites depend on, the posture is what decides whether you review an approach or a mess.

The surface

Cursor’s built-in modes are Agent, Ask, Plan, and Debug (as of September 2026 - Plan, Debug, and the Agent surface each have a page under cursor.com/docs; Ask has no page there, and its documentation is a Help Center article, ask mode):

• Agent - full execution, and the default posture for most work. It plans, edits across the files it judges relevant, runs commands, and self-corrects until the task is done. (Agent overview)
• Ask - read-only Q&A. It traces and explains your code; it does not change your files.
• Plan - asks clarifying questions, researches read-only, then writes the approach as an editable markdown plan you approve before anything is built. (Plan mode docs)
• Debug - a reproduce-instrument-fix loop for bugs that resist a straight Agent pass. (Debug mode docs)

Older tutorials also lean on Custom Modes - user-defined modes configured in settings. That surface was removed during Cursor 2.1’s infrastructure overhaul and has not returned in that form (forum discussion). The current prompting docs describe a newer mechanism in the Agents Window and the CLI: a skill (a packaged procedure) promoted to run as a persistent mode under the same Custom Modes name, which is a different thing from the old settings surface (prompting docs). Mode lists are exactly the sort of thing that shifts between Cursor releases; when a tutorial disagrees with the linked docs, trust the docs.

The mode picker

Switch with the mode picker next to the chat input, press Cmd/Ctrl+. to open the mode menu, or tap Shift+Tab to cycle (shortcut reference). The reason this is a chord rather than a setting buried in preferences is that you’re meant to switch often - sometimes several times inside one piece of work. The natural rhythm: Ask to understand unfamiliar ground, Plan to agree the shape of a consequential change, Agent to do the building. That’s one task widening in latitude as your confidence grows, not three separate conversations.

The moves of this chapter

• Ask & Agent - the read-only posture for when you can’t yet name the change, then full execution on a contained change where reviewing a finished diff beats narrating every step - Ask and Agent: read-only Q&A, then full execution
• Plan - clarifying questions, read-only research, and an editable markdown artifact before a line of code is written - Plan: clarify, research, then a plan you can edit
• Debug - the reproduce-instrument-fix loop for a bug a straight Agent pass can’t crack - Debug: reproduce, instrument, then fix

A mode is a control decision in this course’s sense - control being what guides execution and enforces boundaries - and the boundary it sets here is how much happens before you look. The skill this chapter builds isn’t memorizing a mode list; it’s reading the task in front of you and knowing which posture it’s asking for before you type. Start with the two you’ll use all day. Next: Ask and Agent.