Build Parallel AI Agent Workflow with Git Worktrees — AI Workflow

Use Worktrunk to run multiple AI coding agents in parallel on separate git worktrees and compress a 2-week sprint into days.

The Problem

Your team has a 2-week sprint backlog with 15 independent tasks. Running one AI coding agent sequentially — task by task — still takes days. You need parallelism, but running multiple agents in the same git repo causes conflicts, stashing headaches, and lost work.

The Solution

Use Worktrunk to create isolated git worktrees, assign one AI agent per worktree, and run them all simultaneously. Each agent works on its own branch in its own directory. No conflicts, no stashing, no branch switching. Hooks automate dependency installation and test validation before merging.

Step-by-Step Walkthrough

Step 1: Configure Worktrunk

bash

cargo install worktrunk
cd ~/projects/saas-app

Create a configuration file:

toml

# .worktrunk.toml
[hooks]
on_create = "npm install"
pre_merge = "npm test && npm run lint"

[create]
copy = ["node_modules", ".next"]

[paths]
template = "../saas-app.{branch}"

Step 2: Create Worktrees and Launch Agents

bash

wt switch -c -x "claude --prompt 'Implement Stripe subscription billing with webhook handlers for plan changes'" feat/billing
wt switch -c -x "claude --prompt 'Add role-based access control with admin, editor, and viewer roles'" feat/rbac
wt switch -c -x "claude --prompt 'Build CSV and JSON export for the analytics dashboard'" feat/export
wt switch -c -x "claude --prompt 'Write integration tests for the user auth flow'" feat/auth-tests
wt switch -c -x "claude --prompt 'Add email notification system with SendGrid for invoice events'" feat/notifications

Each command creates a new worktree, copies node_modules from the main tree, runs npm install, and launches Claude Code with the task prompt.

Step 3: Monitor Progress

bash

wt list
# feat/billing       ../saas-app.feat/billing       [dirty, ahead 4]
# feat/rbac          ../saas-app.feat/rbac           [dirty, ahead 6]
# feat/export        ../saas-app.feat/export         [clean, ahead 2]
# feat/auth-tests    ../saas-app.feat/auth-tests     [clean, ahead 3]
# feat/notifications ../saas-app.feat/notifications  [dirty, ahead 1]

Step 4: Merge Completed Features

bash

wt switch feat/export && wt merge     # pre_merge hook runs tests
wt switch feat/auth-tests && wt merge
wt switch feat/rbac && wt merge
wt switch feat/billing && wt merge
wt switch feat/notifications && wt merge

The pre_merge hook runs npm test && npm run lint before each merge. If tests fail, the merge is blocked and you can inspect the issue.

Step 5: Clean Up

bash

git push origin main

Real-World Example

Elena, an engineering lead at a SaaS company, faces a 15-task sprint backlog. She breaks tasks into 5 independent batches and uses Worktrunk to launch 5 Claude Code agents in parallel. She configures on_create = "npm install" so each worktree is ready to go, and pre_merge = "npm test" to gate merges on passing tests. Within 4 hours, 3 of 5 agents complete their tasks with passing tests. She merges those immediately. By end of day, all 5 are done — feat/export (CSV export with date filtering), feat/rbac (3 roles with middleware guards), feat/billing (Stripe webhooks for plan changes), feat/auth-tests (12 integration tests), and feat/notifications (SendGrid transactional emails). She runs a final integration test suite on main, fixes one conflict between RBAC middleware and the billing routes, and pushes. What would have been a 2-week sprint is shipped in 2 days.

Related Skills

worktrunk — Git worktree management CLI for parallel workflows