Blast Multiple Agents

Run multiple AI agents in parallel across different tasks for maximum throughput.

Overview

Zenflow’s parallel execution model lets you run multiple agents simultaneously across different tasks. This “blast” approach increases throughput—start the next task while agents finish the current one.

Parallel execution

With Git worktrees isolating each task, there’s no conflict between agents. Run as many parallel tasks as your system resources allow.

How It Works

Isolated Workspaces

Each task runs in its own Git worktree:

.zenflow/
└── tasks/
    ├── task-abc123/    # Agent 1 working here
    ├── task-def456/    # Agent 2 working here
    └── task-ghi789/    # Agent 3 working here

Agents can’t interfere with each other because they’re working in completely separate directories.

Independent Branches

Each task has its own branch:

• zenflow/task-abc123
• zenflow/task-def456
• zenflow/task-ghi789

Merge conflicts only matter when combining finished work, not during execution.

Running Parallel Tasks

1. Create multiple tasks

   Queue up the work you want done. Each task can use a different workflow and agent configuration.

2. Start tasks in sequence

   Click Create and Run on each task. Zenflow handles the parallelism automatically.

3. Monitor from the dashboard

   The Kanban board and task list show real-time status for all running tasks.

4. Review and merge

   As tasks complete, review changes and merge successful work.

Best Practices

Start with independent tasks

Parallel execution works best when tasks don’t depend on each other. If Task B needs Task A’s output, run them sequentially instead.

Monitor resource usage

Each task consumes disk space (worktree) and API credits (agent calls). Watch your system resources and subscription limits.

Stagger complex tasks

For resource-intensive workflows (Full SDD), consider staggering start times to avoid overloading your system or hitting rate limits.

Use different agents

Mix agent types across tasks to leverage different strengths—Claude for design, Codex for implementation, etc.

Merge Strategies

When multiple parallel tasks complete:

Sequential Merge

Merge tasks one at a time, resolving any conflicts as they arise:

1. Merge Task A to main
2. Rebase Task B on updated main, then merge
3. Repeat for remaining tasks

Batch Review

Review all tasks first, then merge in a planned order based on dependencies and conflicts.

Feature Flags

Merge all tasks but gate features behind flags for controlled rollout.

Resource Considerations

Resource	Per Task Impact
Disk Space	Full working copy of repo
Memory	Agent runtime + IDE if open
API Credits	Based on agent turns and model
Network	Git operations + API calls

A 500MB repo with 10 parallel tasks uses ~5GB of disk space. Archive completed tasks regularly to reclaim space.

Next Steps

Tasks Overview

Full guide to task management

Orchestrating Agents

Configure agent behavior