Built-in subagents

Cursor ships three built-in subagents, chosen by analyzing where context limits got hit:

Subagent	Job	Why isolate it
Explore	Search and analyze the codebase	Generates large intermediate output; uses a faster model to run many parallel searches
Bash	Run a series of shell commands	Command output is verbose; isolating it keeps the parent focused on decisions, not logs
Browser	Drive a browser via MCP tools	Produces noisy DOM snapshots and screenshots; the subagent filters to relevant results

Notice the model-flexibility trick: the Explore subagent uses a faster, cheaper model by default, so it can run ten parallel searches in the time one main-agent search would take. That ties back to pillar 5, tiered execution: use a cheap model for the grunt work and reserve the expensive one for the hard reasoning.