While Anthropic pushes the agentic workflow forward with Claude Code and Claude Tag, xAI is catching up. With /goal, Musk’s company has introduced a new mode for Grok Build — one that takes the developer out of the loop for the duration of a task.
Plan, execute, verify
The idea is clear: instead of hanging in the usual ping-pong of prompt and answer, you hand the agent a larger implementation task and let it run. /goal plans an approach, breaks the work into a progress checklist, and starts executing. The key part comes at the end: the agent keeps going until the task isn’t just done but verified. Verification can mean reviewing code, inspecting webpages, or running scripts.
You’re not locked out, though. Status, pause, resume, and clear give you control, and you can keep feeding the agent extra instructions while it works. So it’s not fire-and-forget — it’s a steerable long-runner.
Three models, one pipeline
The plumbing is the interesting part. According to xAI, /goal natively uses two models — Composer 2.5 and Grok Build 0.1 — and combines them across the pipeline. The logic: each stage gets the model best suited to it. Plan with one, implement with another, verify with a third. /goal is available now inside Grok Build; the CLI installs with a single command and requires a SuperGrok or X Premium Plus subscription.
My take
If you know Claude Code, you read this feature list with a sense of déjà vu: autonomous execution, planning in steps, built-in verification, control via status and pause. That’s not a knock — it’s the pattern the whole industry is converging on. The agent that doesn’t just answer but works independently to a verified result is becoming the standard.
What I find interesting is the multi-model approach. Different models for planning, building, and checking — that’s a bet that specialization beats what a single large model can do. Whether that delivers better results in practice or just more complexity remains to be seen. But for me the real headline is something else: the race for the best autonomous coding agent is in full swing. And competition has never been bad for the people who actually use the tools.
Sources: xAI: Introducing /goal