xAI adds /goal autonomous long-running task mode to Grok Build

xAI has added a /goal mode to Grok Build, its agentic coding CLI, letting developers hand off a single objective and have the agent plan, execute, and verify the work autonomously instead of managing each step through back-and-forth prompts.

What's new

According to xAI: "Today we're introducing /goal in Grok Build. This new mode yields long-running, autonomous execution, helping you hand off larger implementation tasks to the agent."

How it works, per the announcement:

• A developer gives /goal a single-line objective, and Grok Build plans an approach, breaks the work into a progress checklist, and starts executing without further prompting.
• "With /goal, the agent continues until a task is completed and verified, whether that means reviewing code, inspecting webpages, or executing scripts" — replacing the typical pattern where a coding agent stops after each step for human review.
• Developers can keep sending additional instructions while /goal is running, steering the task mid-execution rather than only at the start.
• The mode includes commands for monitoring and steering long-running tasks, and when a goal finishes, its tracking panel flips to a "Complete" state with every checklist item checked off.
• /goal is available now inside Grok Build, installable via a single CLI command with account sign-in.

Context

Grok Build is xAI's agentic coding product, built on a dedicated 256k-context model, that xAI positions against tools like Anthropic's Claude Code, OpenAI's Codex, and Cursor. Most agentic coding tools today still operate in a request-verify loop: the agent proposes a change, a human or a test suite checks it, and the cycle repeats for the next step. /goal pushes further toward autonomy by having the agent own the entire completion-and-verification loop for a stated objective, only surfacing to the developer once its own checklist is satisfied.

This fits a broader pattern across coding-agent vendors in 2026 of extending how long an agent can run unsupervised — competitors have shipped similar long-running or goal-directed modes, and the underlying pressure is the same: agent usefulness increasingly hinges on how much implementation work can be delegated without a developer babysitting every intermediate step.

Why it matters

The practical value of a coding agent scales with how large a task it can be trusted to take on end-to-end; a mode built specifically for "long-running, autonomous execution" with built-in self-verification is xAI's bet that developers want fewer checkpoints, not more, once trust in the agent's judgment is established. Letting developers inject new instructions mid-run — rather than only at the start or only after the agent stops — is a smaller but practical detail: it addresses one of the more common complaints about autonomous agent modes, that steering a task already in flight is awkward or impossible. Whether /goal earns real usage will come down to how well its self-verification actually catches broken or incomplete work before flipping to "Complete," which is the same trust question every agentic coding tool making this pitch has to answer.