Grok Build Gets /goal: Full Tasks, No Check-Ins

xAI shipped /goal for Grok Build on June 22 — a mode that takes a single-line coding objective and runs with it. The agent builds a progress checklist, works through it step by step, and verifies the result — running tests, checking that pages render, executing scripts — before it considers the job done. No per-step approval prompts, no hand-holding between actions.

Under the hood, /goal operates as an observe-plan-act loop: examine the current state of the project, plan the next action, execute it, then check whether the overall goal has been reached. Commands are straightforward: /goal status shows where the agent is, /goal pause halts it, /goal resume picks back up, and /goal clear resets. Typical use cases include migrating a module to a new API, refactoring a service with test validation, adding an endpoint and confirming it renders correctly, upgrading dependencies with a build check, or porting configuration across multiple files.

Why the verification step is the important part

Most AI coding tools stop after generating code. /goal doesn't. It runs the code, checks whether pages load, executes tests — and only marks the task finished when it can actually confirm the outcome. The practical difference is meaningful: instead of getting a batch of edits you then have to go test yourself, you get a result the agent already validated. That's not a guarantee everything is perfect, but it catches the most obvious failure modes before they land in your lap.

How to get it

Grok Build is xAI's terminal-based coding CLI. Install it with `curl -fsSL https://x.ai/cli/install.sh | bash`. Access to /goal requires a SuperGrok or X Premium Plus subscription. Because the mode can run unattended for extended periods, keep in mind that it burns tokens continuously — the pause and status commands let you check in and control spend rather than letting an autonomous run go open-ended.

The broader trend worth watching: this is the same autonomous-loop pattern that Claude Code popularized, and now xAI is shipping it too. When the major coding agents converge on the same model — describe goal once, agent runs until verified done — it says something about where the floor for these tools is heading.