1 min lesson
Inside Agent: modes are a toolset, models are separate
Tell someone how to act on this idea: "Modes are a toolset, models are separate."
Step 1 of 3
Inside Agent: modes are a toolset, models are separatethe detail that proves you've used it
The single sharpest signal of real usage is knowing that the Agent dropdown has two independent axes. The mode changes the toolset the same underlying model gets; the model is picked separately. Conflate them and you sound like you read the changelog.
Learn more
Advanced table
Reference table
- Mode
- Agent (default)
- Toolset
- Everything - writes code, runs the terminal, makes plans and to-dos
- When it's the right one
- Normal build-and-change work where you trust it to act
- Mode
- Plan
- Toolset
- Read-only research; outputs an editable Markdown plan, no edits until you switch
- When it's the right one
- Anything non-trivial - spend tokens reasoning before touching code
- Mode
- Debug
- Toolset
- Reproduces the symptom, adds instrumentation, watches terminal/debug logs
- When it's the right one
- “My search bar isn't working” - diagnosis, distinct from PR review
- Mode
- Ask
- Toolset
- Read-only Q&A over the codebase, output stays in chat
- When it's the right one
- Day-one onboarding and non-technical users exploring a legacy repo
| Mode | Toolset | When it's the right one |
|---|---|---|
| Agent (default) | Everything - writes code, runs the terminal, makes plans and to-dos | Normal build-and-change work where you trust it to act |
| Plan | Read-only research; outputs an editable Markdown plan, no edits until you switch | Anything non-trivial - spend tokens reasoning before touching code |
| Debug | Reproduces the symptom, adds instrumentation, watches terminal/debug logs | “My search bar isn't working” - diagnosis, distinct from PR review |
| Ask | Read-only Q&A over the codebase, output stays in chat | Day-one onboarding and non-technical users exploring a legacy repo |
Changing the mode changes the toolset available to the model, not the model itself.
Plan modeA mode that makes no edits: it researches the codebase and produces an editable plan you review before any code changes. Press Enter for the full definition. is the most-recommended habit across Cursor's own workshops. It researches the whole indexed codebase, follows existing paradigms, asks clarifying questions and emits a hyperlinked Markdown plan - scope, implementation, tests, concrete to-dos. Editing that plan costs zero tokens because you're editing a Markdown file, and a teammate can sign off before any code is written. The flagship loop is plan with a frontier/thinking modelA reasoning model (shown with a brain icon in Cursor's picker) that spends extra compute before answering; reach for it on complex, nuanced work and a standard model for fast, simple tasks. Press Enter for the full definition. (it only reasons, not outputs), then execute the to-dos with ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. - Cursor's own coding model at roughly a tenth the cost. Letting an expensive model write every line burns the budget fast.
If asked “walk me through Cursor,” don't recite features top to bottom. Sequence them as adoption funnels: Tab is acquisition, Chat is onboarding, Agent is the value ceiling and Rules plus MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. are the moat that encodes a team into the tool. The sequence is what tells the room you've run the product, not just opened it. Land one concrete detail - “the mode is a toolset, not a model, and Plan modeA mode that makes no edits: it researches the codebase and produces an editable plan you review before any code changes. Press Enter for the full definition. is free to edit because it's just Markdown” - and you've separated yourself from candidates reciting the marketing page.
QAn engineer says “Cursor's answers about our codebase are garbage.” Before touching the model, what is the most likely cause and where would you look first?