Run Multiple AI Agents Parallelly Without File Conflicts – Worktrunk Makes Git Worktree as Easy as Branches

Problem: Parallel AI agents clash on the same files

When developers use multiple AI coding agents (e.g., Claude Code, Codex) simultaneously, each agent modifies files in the same repository directory, causing frequent merge conflicts and coordination overhead that outweigh the speed gains.

Background: What Git worktree is

Git normally works with a single working directory tied to the current branch. To work on multiple branches at once you must switch branches repeatedly or clone the repository. Git worktree breaks this limitation by allowing several branches to be checked out simultaneously, each in its own independent directory with separate index and file state.

Core commands of Worktrunk

Worktrunk provides four high‑level commands that cover the entire worktree lifecycle:

wt switch – switch to an existing worktree or create a new branch + worktree in one step.

wt switch feat                # switch to worktree for branch feat
wt switch -c feat             # create branch + worktree and switch
wt switch -c -x claude feat   # create, switch, and launch Claude agent

wt list – list all worktrees with status, ahead/behind counts, and unpushed commits. wt list wt merge – merge a worktree back to main with optional squash, rebase, or fast‑forward, then clean up. wt merge main wt remove – delete a worktree and its branch after merging.

wt remove

Parallel agent workflow

Worktrunk is designed for the "parallel AI agent" scenario. Each agent gets its own worktree, eliminating file‑level interference. Example commands launch three agents in separate worktrees:

wt switch -x claude -c feature-a -- 'Add user authentication'
wt switch -x claude -c feature-b -- 'Fix the pagination bug'
wt switch -x claude -c feature-c -- 'Write tests for the API'

The -x flag runs the specified agent after switching, and the -- delimiter passes the agent’s instruction.

Additional design details for parallel work

Each worktree can be assigned a unique port via a hash_port template, preventing server port collisions.

Shared build caches ( target/, node_modules/) avoid redundant compilation across worktrees.

New worktrees can inherit compiled artifacts from existing ones.

Automatic cleanup removes worktrees and branches after merge.

Full end‑to‑end example

Assume you want to add a "user authentication" feature.

Create and start: wt switch --create feature-auth – creates the branch, worktree, and switches to it.

Develop: work in the isolated directory; changes do not affect main or other worktrees.

Inspect status: wt list shows which worktrees have uncommitted changes and how many commits they are ahead of main.

Commit: wt step commit stages and commits the changes.

Merge or PR: wt merge main performs squash/rebase/fast‑forward and cleans up, or use gh pr create followed by wt remove for a pull‑request flow.

The whole process requires only a handful of short commands instead of the verbose native Git worktree syntax.

Advanced capabilities

Hooks: automatic scripts run at create, pre‑merge, post‑merge, etc., e.g., installing dependencies or starting dev servers.

LLM‑generated commit messages: diff‑based AI creates clear commit messages.

Interactive selector: wt list --full shows real‑time diffs, logs, and CI status, letting you pick a worktree quickly.

PR shortcut: wt switch pr:123 jumps directly to the branch associated with pull request #123.

Comparison with alternatives

Three approaches to managing worktrees are compared:

Native Git commands: fully featured but verbose; creating a worktree requires entering the branch name three times, and cleanup is manual.

Custom shell scripts / aliases: ad‑hoc solutions that lack unified functionality and do not address the parallel‑agent use case.

Worktrunk: purpose‑built for parallel AI agents, chaining creation, agent launch, monitoring, merge, and cleanup, plus extra features (hooks, AI commit messages, CI status) that native Git lacks.

For occasional worktree usage, native commands suffice; for heavy parallel agent workflows, Worktrunk offers a smoother experience.

Who should use it

Heavy users of AI coding agents, teams that need multi‑branch parallel development, and anyone wanting to understand the "agentic coding" standard practice will benefit. Developers who rarely run multiple agents may find the tool unnecessary.

Key takeaways

Optimizing high‑frequency operations (e.g., reducing three‑step worktree creation to one command) yields large time savings.

Isolation of work directories is essential for safe parallel development.

Automation should cover the entire lifecycle, including cleanup.

Designing tools around concrete workflows, not just feature lists, produces higher utility.

Worktrunk turns the most cumbersome part of Git—worktree management—into a branch‑like experience, making parallel AI agent development practical.