Skip to lesson
Exit
How Coding Agents Work (and Their Limits)1 / 2

1 min lesson

The autonomy spectrum

Take this situation: "A developer says 'an agent is just a smarter autocomplete that edits files.' What single distinction would you teach to correct that and why does it matter?" Lead with your decision, then add the reason.

Step 1 of 2

The autonomy spectrum

Cursor's surfaces are not three names for the same thing. They sit at different points on how much the agent decides versus how much you steer and teaching developers to pick the right one for the task is pure DX.

Surface
Tab
Autonomy
Lowest - predicts your next edit as you type
Best for
Staying in flow on code you're actively writing
Surface
Agent / 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.
Autonomy
High - plans, edits across files, runs the loop
Best for
Multi-file changes where you describe the goal, not the keystrokes
Surface
Background / Cloud Agent
Autonomy
Highest - runs a task to completion off your machine
Best for
Longer tasks you hand off and review as a diff or PR later

Same underlying ideas, different leashes. Match the surface to how much you want to stay in the loop.

Interview move

If a screen asks you to explain how Cursor works, resist the urge to dive into model internals. Draw the loop - assemble, decide, act, observe, revise - then place Tab, Agent and a cloud agent on an autonomy axis. That shows you can teach the mental model, which is the actual job, not just that you've read a model card.