Grok Build: The Code Agent, from Beta to Open Source

Update November 2, 2026. Since August 19, 2026, app creation in Grok is open to everyone on web and mobile, with publishing to a grok.me address (lesson 10).

Seven weeks: that’s how long it took xAI to turn Grok Build into a complete code agent ecosystem, from the confidential beta on May 27 to the open source release on July 15. To understand how this tool changes your workflow, you first need to understand what a code agent is: not an assistant that suggests lines in your editor, but a program that receives a goal, reads your project, writes and executes code, fixes errors, and reports back. Here is the timeline, followed by the capabilities that matter most for your practice.

The Timeline

The pace speaks for itself: a major capability almost every week. The trajectory matters just as much—the tool started reserved for SuperGrok and X Premium Plus subscribers, opened to the API five days later, and ended up fully published as open source. In seven weeks, Grok Build went from a closed product to a community asset.

Key Capabilities

Among all these releases, three mark the tool’s real shift in nature. The first is the /goal command, released on June 22: it gives the agent a long-term objective that it pursues autonomously. You no longer converse turn-by-turn with an assistant—you set a destination, and the agent moves forward on its own, over time. This is xAI’s equivalent of the general 2026 trend: agents that work for hours, not minutes.

The second is the Agent Dashboard, available since June 15, and it’s the logical complement to the first. Once each session can work autonomously for a long time, nothing stops you from running several at once—provided you can steer them. The dashboard does exactly that: see what each session is doing, respond to those waiting for a decision, dispatch new work. The developer’s role changes: you no longer write every line, you become a supervisor of a fleet of agents, and your time shifts from typing to arbitration.

The third dimension is extensibility. The Plugin Marketplace, integrated on June 11, lets you extend the agent with third-party capabilities; Workflows, added on July 23, structure reusable sequences—the building block that turns improvised sessions into procedures your team can replay.

Open Source, and Why It Matters

Since July 15, the Grok Build harness and its terminal interface are open source. For a team, this isn’t a licensing detail: it answers the three questions that usually block agent adoption in enterprises. What exactly does it do with my code? You can audit its behavior, line by line. Can it adapt to my internal environment? You can modify it. And how is an industrial code agent actually built? The source code answers better than any article—it’s a learning resource as much as a tool.

This openness isn’t isolated: it fits into a broader strategy. The Grok subscription works in third-party tools—Cursor on all plans since the Grok 4.5 launch, but also Warp, OpenCode, Kilo Code—rather than locking users into the built-in terminal alone. If your team already works in one of these environments, accessing Grok doesn’t require switching.

One last point to keep in mind: since July 16, Grok 4.5 powers the default agent. The grok-build-0.1 model, accessible directly via the API since June 1, remains available for those who want to build their own tools on the fast, economical code model seen in the previous lesson.

Official source: x.ai/news — announcements from May to July 2026

Key Takeaways

• Grok Build went from beta (May 27) to open source (July 15) in seven weeks
• /goal enables long-term autonomous execution; the Dashboard manages multiple sessions
• grok-build-0.1 is accessible directly via the API since June 1
• Grok 4.5 has been the agent’s default model since July 16