Parallel Execution

Multi executes independent work in parallel. Tool calls that don’t depend on each other run concurrently, and the supervisor can spawn multiple subagents at once to tackle different parts of a task simultaneously.

Multiple subagents working in parallel on the same task

How It Works

When Multi identifies independent units of work, it dispatches them concurrently instead of one after another:

• Parallel tool calls - Reads, searches, and other independent operations execute together in a single step
• Parallel subagents - The supervisor spawns multiple subagents at once, each focused on a separate slice of the task
• Smart sequencing - Steps with dependencies still run in order, so correctness is preserved

When It Helps Most

Parallelization shines on work that would otherwise stack up sequentially:

• Exploring large codebases (many reads and searches at once)
• Refactors that touch unrelated modules
• Multi-area features where each area can progress independently
• Long investigations that branch into several threads

For small, single-file tasks, the gains are minimal - but for ambitious work, parallel execution can cut wall-clock time significantly.

Working With Parallel Subagents

Everything from Monitoring Subagents applies when multiple subagents run at the same time. The subagents bubble shows all active workers, and you can jump between them to inspect progress independently.

The supervisor waits for parallel subagents to report back, then merges their results before continuing.