# Run Claude Code and Codex side by side without losing the work

> A concrete two-agent workflow with separate checkouts, clear ownership, shared project context, and one final review.

Canonical HTML: https://canopyide.dev/use-cases/parallel-claude-code-codex-agents
Article date: 2026-09-27

Two agents can make progress at once when the jobs do not depend on the same edits. The operational challenge is keeping their files, decisions, and final review connected.

## Example: implementation and independent review

Suppose Claude Code is building a settings screen while Codex reviews the authentication flow. Write a separate outcome for each: the first must produce a running screen and a diff; the second must produce findings tied to specific files and behavior. If the reviewer needs the new settings code, wait for a stable branch or PR before asking it to review that change.

## Give code-changing work a separate checkout

Launch each CLI from its own Git worktree and branch. Git worktrees are separate working directories attached to one repository, so an agent's uncommitted edits do not overwrite the other's working files. Shared dependencies, API contracts, and eventual merge conflicts still need coordination. Canopy keeps branch, files, commits, and PR beside the agent session.

## Use shared context for handoffs

In the same Canopy project, inspect recent prompts, file activity, claims, and agent messages. Tell the reviewer which branch or PR represents the implementation. Keep the acceptance criteria in the handoff: what should work, what should be tested, and what concerns deserve a second opinion. Shared context does not merge conversations or provider accounts into one model plan.

## Choose the model in each CLI

A quick investigation may suit a lighter model; a difficult change or critical review may justify a stronger one. Select the model through Claude Code or Codex as supported by that CLI. Canopy brings the sessions into one workspace and summarizes usage signals exposed by the tools. Displayed cost is an estimate, not a combined bill.

## Finish at the diff and running app

Open the implementation branch's changed files and run the project. Compare the preview with the request, check tests and CI, and ask the review session to challenge assumptions. Resolve findings in the implementation branch, inspect its new diff, then merge through the normal PR process. Count accepted changes, not active agent tabs.

## Frequently asked questions

### Do Claude Code and Codex share a paid subscription in Canopy?

No. Each CLI uses its own account, model settings, and billing. Canopy organizes their local sessions and project context.

### Can both agents edit the same file?

They can, but separate worktrees do not resolve competing decisions or merge conflicts. Assign clear ownership and review the combined result.

## Sources and further reading

- [Canopy app README: sessions and worktrees](https://github.com/FluidWorksApp/canopy-ide/blob/main/README.md)
- [Git worktree documentation](https://git-scm.com/docs/git-worktree)

## Related Canopy pages

- [Git worktrees for parallel coding agents: commands and cleanup](https://canopyide.dev/guides/git-worktrees-for-parallel-coding-agents.md)
- [Keep coding-agent context useful between sessions](https://canopyide.dev/guides/keep-agent-context-between-sessions.md)
- [AI-generated pull request review checklist and reviewer prompt](https://canopyide.dev/guides/ai-generated-pr-review-checklist.md)
- [Agent session, terminal tab, branch, or worktree: what is the difference?](https://canopyide.dev/guides/agent-session-vs-tab-vs-worktree.md)
- [Canopy vs Claude Code Desktop: which agent workspace fits?](https://canopyide.dev/use-cases/canopy-vs-claude-code-desktop.md)

Canopy runs installed coding CLIs; CLI accounts, model selection, and provider billing remain separate. Check the installed release before relying on version-specific behavior.