SAN FRANCISCO — xAI has given developers a new way to hand off serious work to its coding agent, launching /goal in Grok Build, a mode built for long-running autonomous task execution that keeps working until a job is finished and verified.
Announced Monday, /goal turns a single objective into a fully managed workflow. Rather than nudging the agent step by step, a developer states a goal and Grok Build plans an approach, breaks the work into a progress checklist, and executes the items one at a time. Crucially, the agent does not stop at writing code: verification is built in, meaning it reviews its own output, inspects webpages, or runs scripts to confirm the work actually holds up.
Goals as a First-Class Mode
The feature treats goals as a first-class concept inside Grok Build, xAI's terminal-based coding agent. Developers are not locked out while a goal runs, either. They can keep sharing additional instructions mid-task, steering the agent as it works through its checklist. It is a meaningful step beyond single-prompt assistance and into sustained, self-directed execution, and it slots neatly alongside xAI's enterprise push covered in our report on Grok 4.3 going live on Amazon Bedrock.
A Multi-Model Pipeline
Under the hood, /goal natively combines two systems, Composer 2.5 and Grok Build 0.1, assigning different models to different stages of the pipeline. The result is a division of labor, with one model planning, another implementing, and a third verifying, an approach designed to bring higher intelligence to each phase of the task.
