YOLO Studio
DocsHome

Guide

How to run multiple coding agents in parallel

Running one coding agent is easy. Running four on the same repository without them overwriting each other takes some structure: separate checkouts, a queue of well-defined tasks, and a way to review what comes back. This guide sets that up in YOLO Studio.

Get started

The problem with one working copy

Agents edit files and run git commands. Put two in the same directory and one agent’s half-finished change becomes the other’s broken build. Each agent needs its own checkout and its own branch, and the work needs splitting into pieces that do not depend on each other’s unmerged code.

Option 1: a few agents, by hand

  1. 01

    Open a workspace on your repository

    Create a workspace from your GitHub repository, or import a local project with yolostart.

  2. 02

    Start each agent in its own lane

    Add an agent tile and choose Create new lane, or press Quick Lane in the toolbar. Each lane is its own git clone on a lane/<name> branch. Mix agents freely: Claude Code in one lane, Codex in another, Gemini CLI in a third.

  3. 03

    Give each one a separate job

    Keep the tasks independent: one agent on the API, another on the UI, a third on tests. Turn on YOLO Mode so they do not stop for approval.

  4. 04

    Merge the results

    When a lane is done, use the Commit & Push or Merge to Main shortcut on its tile, or open a pull request from the lane’s branch.

Option 2: a backlog the Runner works through

For more than a handful of tasks, let the Background Runner hand them out.

  1. 01

    Describe the work

    Add a Kanban board and say what you want done. It becomes cards, one outcome each, with acceptance criteria spelled out.

  2. 02

    Groom the backlog

    Sharpen vague cards, and link cards that depend on each other so a dependent card waits for the work it needs to be merged.

  3. 03

    Connect the board and start the Runner

    Pick the board’s Ready column as the Runner’s source and choose how many cards run at once (one to four). Each card gets its own isolated lane and agent.

  4. 04

    Choose how results land

    Manual, PR + review, or Auto merge into the branches you allow. Turn on Agent review to have a different model check each result before anything relies on it.

  5. 05

    Review what comes back

    Each finished card records its result and, for code, its commit or pull request. Cards that need you are held for review instead of guessed at.

Tips for cards that parallelize well

One outcome per card

A card that says exactly what done looks like is one an agent can finish without asking.

Split by area of the code

Cards that touch different files merge cleanly; cards that rewrite the same module conflict.

Make dependencies explicit

If B needs A’s code, link them. The Runner starts B only once A has landed.

Keep reading

Guide: from a Kanban backlog to merged pull requestsRun Claude Code in the cloudRun Codex CLI in the cloudRun Gemini CLI in the cloud

Try it on your own repository

Sign in with GitHub, open a workspace on your repository, and put your first agent to work in a few minutes.

Get started
YOLO Labs

© 2026 Yolo Labs, LLC

Home
Security
Docs