Skip to content
vibecodershq.
Submit a tool

14 Oct 2026

How to orchestrate Claude Code (and Codex or Cursor) with a Grok Bot

Use a Grok Bot to plan, delegate and check work while Claude Code, Codex or Cursor writes the code. Setup examples, limits, and when it's worth it.

If you want to orchestrate Claude Code, Codex or Cursor without babysitting each session, one setup keeps coming up: a bot acts as the manager and a coding agent acts as the hands. The bot plans the work, hands it off, keeps an eye on it and checks the result. The coding agent edits the repository. A Grok Bot is not a rival to Claude Code. In our view, the two do different jobs and fit together well.

Manager vs hands: what each layer does

The manager (Grok Bot). Each Bot works on a persistent cloud computer with a browser, filesystem and terminal, and work continues while your laptop is closed (Grok Bot docs). Bots can run routines, including routines triggered by events such as a Slack message or a GitHub notification (Work with Grok Bot). Grok Bot can also hand coding tasks to separate Cursor Cloud Agents, under your existing Cloud Agent controls (Grok Bot for teams). The official engineering guide says a Bot can create Cursor cloud agents, read their transcripts, review the proofs attached to PRs, and send follow-ups by queueing a message or interrupting the run (Grok Bot for Engineering).

The hands (coding agents).

  • Claude Code reads your codebase, edits files, runs commands and works with git. It runs in the terminal, IDE extensions, a desktop app and the browser (Claude Code overview). You can run it non-interactively with claude -p, which makes it easy to script (same source).
  • Cursor Cloud Agents run in isolated cloud VMs with full development environments: cloned repos, installed dependencies and secrets (Cursor Cloud Agents).
  • Codex ships with ChatGPT plans and runs on the web, in the CLI, in the IDE extension and on iOS from the Plus plan up (Codex pricing). It also supports scriptable workflows through codex exec and its SDK (same source).

Grok Bot vs Claude Code: rivals or a team?

They're a team. Claude Code is a coding agent you point at a repository. A Grok Bot is a general agent that runs on its own computer and works across your apps. The useful question is not which one to pick. It's which layer does the deciding and which does the editing.

What people are posting

  • @coderchrisdean reports that @thecsguy's Grok Bot "hands coding tasks to Claude Code on his Mac, then reports back when it's done," and calls it "Bot as the manager, coding agent as the hands."
  • @de1lymoon shares a one-prompt "harness" with a lead agent, a web scout, a two-source rule for claims and a red-team pass. In @de1lymoon's words, "you stop prompting models. you sign off on results."

These are each author's own description. We haven't tested either setup.

Example setup A: Grok Bot hands tasks to Claude Code on your Mac

This is an example, not a recipe from any vendor.

  1. Install Claude Code on your Mac and confirm claude --version works (Claude Code overview).
  2. Allow the Bot to run commands on that computer. Local commands run under a local execution policy, which asks for approval on each command by default (Work with Grok Bot).
  3. Give the Bot a brief: "Each morning, pick the top issue labeled bug, run claude -p "fix <issue>, add a test, commit on a new branch" in the repo, then send me the diff summary and test output."
  4. You review the branch and merge it yourself.

In our view, it's worth keeping per-command approval turned on until you trust the prompts.

Example setup B: Grok Bot runs Cursor Cloud Agents

This is an example, not a recipe from any vendor.

  1. Check that your team has Cloud Agent delegation turned on. It's on by default, and admins can turn it off (Grok Bot for teams).
  2. Ask the Bot: "Start a cloud agent on repo to add pagination to /orders. Require a before/after screenshot. Don't report done until the screenshot shows the change."
  3. The Bot watches the agent's transcript and artifacts, and it can queue a message or interrupt the run (Grok Bot for Engineering).
  4. You review the PR.

Codex and Claude Code together

Claude Code can read an existing AGENTS.md, the instruction file that Codex uses, so one file can serve both (Claude Code overview). In our view, a good split is to have one agent write the code and another review it. A Bot can schedule both and send you a single summary.

Guardrails

  • Shared computer. All your Bots share one cloud computer and its logins (Grok Bot docs).
  • Review gate. Let the Bot open PRs, not merge them.
  • Narrow triggers. The docs warn against broad listeners like "every new message" (Work with Grok Bot).
  • Cost. Grok Bot comes with Cursor Premium and up (Cursor pricing); Claude Code needs a paid Claude plan (Claude pricing). Both checked Oct 14, 2026.

When not to bother

If you work on a single repo and are at your desk anyway, running Claude Code or Cursor directly is simpler, in our view. The manager layer pays off when tasks recur, run overnight, or span several tools.

A note on models

@de1lymoon's post says every Bot "now runs on Claude Opus 5.5." Grok Bot's official help page says otherwise: Grok Bot "is not tied to a single model." It picks a backend model for each task and has no model picker (Which AI models power Grok Bot?). We couldn't find official confirmation of the model claim, so we didn't use it.

Sources