xAI launches Grok Build, its AI coding agent

xAI has launched the initial beta of Grok Build, an AI agent designed to help with professional software projects and complex programming tasks. It is now available to subscribers of SuperGrok and X Premium Plus.

Grok Build installs with a single command and runs directly inside your repository. The tool can analyze project conventions and use files such as AGENTS.md, along with plugins, hooks, skills and MCP servers that are already configured.

More control before changing code

For larger jobs, Grok Build includes a planning mode. The agent proposes the steps before executing anything, and you can approve the plan, comment on each part or rewrite it entirely.

After approval, every modification appears as a clean diff, meaning a view that shows exactly which lines were added, removed or changed. This lets you review the work before incorporating it into the project.

The goal is to reduce one of the common risks of coding agents: making many changes at once without making it clear why they made them.

Multiple agents working in parallel

For larger tasks, Grok Build can delegate parts of the work to specialized subagents. These processes can run in parallel, for example to analyze different areas of an application or prepare independent changes at the same time.

It also supports worktree integrations, a Git feature that creates separate workspaces so each subagent can work without interfering with the main codebase. The team can then review and combine the changes that make sense.

It also includes a headless mode, activated with -p, to run the agent from scripts and automations. xAI also offers full ACP compatibility, allowing you to create your own bots and applications to coordinate agents.

What this means for you

If you develop software and already pay for SuperGrok or X Premium Plus, you can use Grok Build inside your repositories for tasks such as:

• Planning and executing changes across multiple parts of an application.
• Reviewing modifications through diffs before accepting them.
• Distributing tasks among agents working in parallel.
• Integrating the agent into scripts, bots and automated workflows.

This is not a final release. xAI describes Grok Build as an initial beta, so its performance, compatibility and behavior may still change. The company asks users to submit bugs, requests and feedback with the /feedback command inside the tool.

Grok Build adds to the competition among coding assistants that do more than suggest lines of code. They also plan tasks, modify repositories and coordinate processes. The key question now is how much control developers retain when the work stops being a one-off suggestion and starts running across multiple stages.