Grok Build’s Goals: Complete Tasks Without Checking In

On June 22nd, xAI released /goal in their Grok Build terminal coding CLI. This feature allows you to set a single line of coding goals and watch as the agent automates the process of completing them. It creates a todo list, completes each item sequentially, and validates its work along the way (running tests, checking page loads, running scripts, etc.) until it declares itself done. There are no interstitial prompts asking if each step was correct, no need to babysit the agent while it does its thing.

/goal is powered by an observe, plan, act cycle where the agent observes the current state of the project, plans what to do next, acts on its plan, and observes if the final goal was achieved. You can easily interact with /goal using simple commands like /goal status (to see where it is), /goal pause (to pause the agent), /goal resume (to continue from where you left off), and /goal clear (to reset). Some examples of when you might want to use /goal include moving a module to a new api, refactoring a service while testing, creating a new endpoint and making sure it renders properly, updating dependencies while building, or transferring configurations between files.

1Why the verification step is the important part

Unlike other AI coding assistants which only generate code, /goal will actually run that code, make sure pages are loading correctly, run tests, etc. until it can verify success. This means you don’t just get a pile of changes that you have to manually verify; you get something that has already been tested by the agent. While this isn’t a guarantee that there won’t be any mistakes, at least you know the most common ones have been filtered out.

2How to get it

To try out /goal in xAI’s Grok Build terminal coding CLI, simply install it by running curl -fsSL https://x.ai/cli/install.sh | bash . Note that to use /goal you must have a SuperGrok or X Premium Plus account. Since /goal can be left running for long stretches of time, remember that it will consume tokens during this period so use the pause and status commands to keep track of how much you’re spending.

This release represents another step towards the autonomous loop style of interaction pioneered by Claude Code. With both major coding agents adopting this approach (where you tell the agent what your goal is, and it continues until it can verify success), we may be seeing the beginning of a minimum viable product for these types of tools.