Grok Build adds autonomous /goal mode
xAI's Grok Build can now take a one-line goal, plan it, build a checklist, and run until the work is done and verified, with pause and resume controls.
xAI shipped /goal in Grok Build on June 22, a mode that turns its terminal coding agent loose on one instruction and keeps it working until the job is verified. You hand it a single line, like /goal Migrate the auth module to the new API, and the agent plans an approach, breaks it into a progress checklist, executes, and checks its own work, whether that means reviewing code, inspecting webpages, or running scripts. It stops when every checklist item is done.
The difference from a normal Grok Build session is that you are not driving each step. Four commands let you watch and steer the run: /goal status shows a live progress panel, /goal pause halts the work but keeps the goal, /goal resume picks it back up, and /goal clear drops it entirely. You can keep feeding it instructions while it works.
It is available now in the Grok Build CLI, installed with curl -fsSL https://x.ai/cli/install.sh | bash, and access needs a paid SuperGrok or X Premium Plus subscription.
Why this matters
This is the same long-running, hands-off pattern as OpenAI’s recent Codex Record and Replay: hand over a task, step away, come back to a result. That saves time, but an agent that edits code, opens webpages, and runs scripts on its own can also do the wrong thing fast. If you let one run unattended, keep a person checking the output before it ships, and if you are newer to coding agents, start with small, reversible goals on a branch you can throw away.