Cursor's new skill spawns swarms of AI agents
Cursor just launched /orchestrate, a new skill built on the Cursor SDK that recursively spins up multiple agents to tackle heavy workflows. In internal testing, it autoresearched Cursor's own skills to slash token usage by 20% while cutting backend cold start times by 80%. It's designed to push AI coding assistants far beyond single-agent limits.

When a single coding agent hits a wall on a massive codebase, giving it a bigger prompt rarely solves the problem. Cursor’s answer is to let the AI spawn its own team.
Cursor just unveiled /orchestrate, a new skill built on top of the Cursor SDK that recursively spins up multiple agents to tackle ambitious, complex tasks beyond the reach of single-agent workflows.
Spawning agents to fix the backend
Instead of trying to force one model to keep an entire architectural refactor in its head, /orchestrate coordinates nested sub-agents to divide and conquer demanding workflows.
Cursor put the feature to work on its own infrastructure during internal testing, reporting two notable wins.
First, it auto-researched Cursor’s internal skills, trimming token usage by 20% while actually boosting evaluation scores. Second, it helped slash cold start times on the company's internal backend by 80%.
Why it matters
Most AI code assistants start to fall apart when tasks span multiple services, dependencies, and files. Trying to brute-force those workflows with single prompts usually ends in hallucinated logic and sky-high token bills.
Cursor’s move to recursive orchestration is clever because it shifts the burden from massive context windows to specialized task delegation.
We think this shows where developer tools are clearly headed: moving past chat interfaces into self-managing swarms of agents. The real challenge, of course, will be keeping recursive execution loops from going off the rails once everyday developers start throwing unvetted codebases at it.
Sources (1)
- x.com(opens in a new tab)Originalx.com