Run agents in parallel

Parallel agents are useful when work can be split into independent parts or when you want to compare approaches. Each agent receives its own git worktree.

Choose the right split

Good parallel work has clear boundaries. Examples include:

• Implementation and test coverage.
• Backend and frontend work with an agreed interface.
• Two independent investigations of the same bug.
• Separate features that touch different files.

Avoid splitting tightly coupled edits that are likely to modify the same code. The merge cost can outweigh the time saved.

Start parallel work

Open the task drawer and select Split into parallel agents. Define a focused outcome for each agent, or use Suggest with AI to draft the split.

Before starting, check that every agent has:

• A distinct responsibility.
• The shared constraints or interface it must respect.
• A clear definition of done.

Monitor and combine results

Use the Workbench to watch all active sessions, or let the Queue surface only agents that need you. The session rail lets you switch between agents working on the same task.

Review each result independently. Save only the work you want to keep, then continue through the normal review and shipping flow.

TIP

Parallelism is best for reducing elapsed time, not for making an unclear task clear. Plan the boundaries first when several agents will share a task.