📝Parallel Agents is the practice of running several AI agents at the same time on independent pieces of work, shortening total time and keeping each agent's context focused.
A single agent works sequentially. Parallel agents split a job into tasks that do not depend on one another, such as separate features, files, or research questions, and run them at once. Each agent has its own context, so none carries the full weight of the whole job.
Isolation is the main technical requirement. Agents editing the same repository need separate working copies, commonly provided by git 📝Worktree checkouts, so their changes do not collide. Results then return for review and merging, by a person or by an orchestrating agent.
Parallelism trades cost and coordination for speed. Token spend scales with the number of agents, and poorly separated tasks produce merge conflicts or duplicated work. In 📝Claude Code the pattern appears as agent teams and 📝Agent Swarms in Claude Code, and it underlies the orchestrator-worker pattern.