Skip to main content
Back to Community Hub
Workflow

Parallel Sprints with Claude Code — Running Multiple Agents Like Garry Tan

Garry Tan's advanced technique for running 10+ Claude Code sessions simultaneously — using GStack's sprint structure to keep multiple agents coordinated and productive without chaos.

by Peter Kibet24 Mar 2026
productivityparallel-agentsgstackclaude-codegarry-tanadvancedscaling
cat community/parallel-sprints-with-claude-code-running-multiple-agents-like-garry-tan-mn5966ru/README.md

This workflow describes how Garry Tan scales his output by running multiple Claude Code sessions in parallel — and why structure is what makes it work.

The Problem: Running one Claude Code session at a time is productive. But if you want to ship 10,000–20,000 lines per day like Garry Tan reports, you need multiple agents working simultaneously. Without structure, ten agents just create ten sources of chaos.

The Solution — Parallel Sprints: GStack's sprint process (Think → Plan → Build → Review → Test → Ship) is what makes parallelism possible. Each agent knows exactly what to do and when to stop because each skill has a clear scope and defined outputs.

How it works in practice: • Session 1: Running /office-hours on a new feature idea • Session 2: Running /plan-eng-review on yesterday's design doc • Session 3: Implementing a feature from an approved plan • Session 4: Running /review on a completed branch • Session 5: Running /qa on staging • Session 6: Running /ship on a reviewed branch

All at the same time, each in its own isolated workspace.

Key principles: 1. Each agent works in its own Git branch — no conflicts 2. The sprint structure prevents agents from stepping on each other 3. Each skill produces artifacts (design docs, test plans, PR descriptions) that other skills consume 4. You can use GStack's Conductor tool to orchestrate multiple sessions

Why this matters: This is how a single person can match the output of a traditional engineering team. The structure is the multiplier — without it, more agents just means more mess. With it, each additional agent adds predictable value.

This is an advanced workflow best suited for developers who are already comfortable with Claude Code and GStack basics.

Source: Garry Tan's GStack (github.com/garrytan/gstack), MIT licensed.

## Installation

install.sh
1. Install GStack (see the GStack resource for full install instructions)
2. Open multiple terminal windows/tabs
3. In each terminal, navigate to your project and run Claude Code
4. Assign each session a different sprint stage:
   - Terminal 1: /office-hours (new feature ideation)
   - Terminal 2: /plan-eng-review (architecture review)
   - Terminal 3: Build mode (implementing approved plans)
   - Terminal 4: /review (code review on completed branches)
   - Terminal 5: /qa https://staging-url (QA testing)
5. Each session should work on its own Git branch
6. Use /retro to review progress across all sessions

## Comments (0)

No comments yet. Be the first to share your thoughts.