Grok Build adds /tutorial: CLI coding agents now compete on the "first ten minutes after first launch"
At 08:25 this morning, Elon Musk forwarded a very short update: download Grok Build, then type /tutorial. Just one line. On its face, this is not a model breakthrough, and it hardly looks like an official release.
Yet it reveals a product signal. The battlefield of AI coding is shifting from "can it modify code" to "will an ordinary developer figure it out within ten minutes of opening it for the first time."
Grok Build is no longer a bare CLI. The official product page lists a packed feature set: Plan mode, Skills, Plugins, Hooks, MCP, AGENTS.md, Memory, parallel subagents, standalone worktrees, headless mode, ACP integration, code review, sandboxing, and background tasks. With the model switched to Grok 4.5, xAI is selling a full terminal-based development workbench—code completion is just one cell in the grid. [1][2]
/tutorial fills the least glamorous but most discouraging layer: onboarding. As of July 22, the latest public version in the official changelog is still v0.2.111, which contains no /tutorial entry, so today's announcement looks like a heads-up before the documentation catches up. v0.2.111 already turned Workflows on by default and allowed grok doctor fix to run inside the TUI; when the permission classifier times out or fails, it no longer silently rejects, but falls back to the normal permission prompt. A few versions earlier, /resume was progressively extended to recover sessions across directories, across machines, and from other coding tools. [3]
Piecing this together, xAI has recently been tackling three kinds of real-world friction:
1. Don't know how to use it: /tutorial leads the way;
2. Environment broken but no idea where: /doctor diagnoses it;
3. Task interrupted, subagent failed, session moved to another machine: Workflows and /resume pick things back up.
That looks more like product engineering than stuffing in another ten benchmark numbers. Where long-running agents die most often is permission dialogs, environment variables, session recovery, background-task state, and failure retry—the functions themselves usually get written. Whoever polishes these edges gets the chance to move from "tried it once" to "resident in my daily workflow."
Still, don't mistake /tutorial for the trust problem being solved. Grok Build's Rust CLI/TUI and agent runtime are on GitHub under Apache 2.0, but the repo is a mirror synced periodically from an internal SpaceXAI monorepo, and external contributions are not currently accepted. Open-sourcing makes the local harness auditable; the hosted model, account services, telemetry, and data retention remain a separate chain. The repo's SECURITY.md only gives a vulnerability reporting entry point, without a full explanation of runtime data policy. [4][5]
My read is simple: today's update is small but very "product-shaped." The next round of competition for terminal agents will hinge not only on who is smarter, but on who can teach the user, diagnose itself, explain its permissions, show progress, and keep going after failure. /tutorial is a small button, but behind it, CLI coding agents have started building a consumer-grade entry point.
Sources
1. Elon Musk: download Grok Build and type /tutorial
2. Grok Build official product page and docs · Docs
3. Grok Build official changelog
4. xAI Grok Build open-source repository
5. Repository security policy