01What Grok Build Is and How It Works

Grok Build is a new coding agent and CLI from xAI, launched on May 25. It's designed for professional software engineering and complex coding tasks. You talk to it in plain language, and it builds games, websites, 3D scenes, and more.

In this episode, Julian Goldie tested Grok Build extensively. He found it a massive step up from the old version. He says it's the first time Grok has gotten close to or beaten Claude Opus on tests.

Grok Build works by taking a simple prompt, like "create a game" and handling the coding. It uses sub-agents for larger tasks, which work like a team. It also self-checks its work in a loop to improve over time.

02Main Features of Grok Build

Here are the key features Julian highlighted in the episode:

  • Builds games, websites, and 3D scenes from text prompts.
  • CLI (command line interface) that you can add to your own agent operating system.
  • Sub-agents for large tasks. Up to 8 sub-agents can work together on one task.
  • Self-checking loop: It plans, builds, checks its own work, and improves.
  • Available with your Twitter/X subscription (Super Grok or X Premium Plus), so no extra API cost.
  • Integrates with Hermes agent and can generate voice notes, videos, and images.
  • Shared memory when used with an agent OS, so it has context from the start.

03How to Get Started with Grok Build

To start using Grok Build, follow these steps:

  1. Make sure you have a Super Grok or X Premium Plus subscription. If you already pay for Twitter, you're in.
  2. Log in to Grok Build using your Twitter/X account. You can have Grok Build on or off, but for this you need it on.
  3. If you want to use it as a CLI, install the Grok Build CLI on your system.
  4. Add it to your agent operating system alongside Claude, Gemini, or others.
  5. Start with a simple prompt, like "create a game" or "build a website" and see what it does.
  6. Use the preview feature to see what it built and iterate from there.

04How to Use Grok Build with Examples

In the episode, Julian showed many examples of using Grok Build. He started with a game prompt. It created a dark first-person game, though he wasn't sure what it was doing.

He also made a website that looked nice and clean.

For 3D, Grok Build created cool 3D games and an orbit diagram of an inner cellular system. One build, the dragon realm, was smooth and quick to navigate. But the graphics weren't great—the dragon in the sky looked off.

He also made a lava lamp animation and a liquid effect that follows the mouse. You can change colors and reset the ball. These were fun and worked well.

To use it, you just talk to Grok Build like a friend. You don't need to be a coder. It handles the coding for you.

You can also plug it into Hermes agent to generate voice notes, videos, and images.

05Results and Comparisons in the Test

Julian compared Grok Build against other models on the same prompts. For the first time, Grok beat Claude Opus on a test. He said it probably outperformed Kimi K217 and GLM 5.2 too.

For the game prompt, Grok Build made a fun, clean design. Claude Opus 4.8 made something boring in comparison. For a glowing fish animation, Grok Build created hundreds of glowing fish with trails.

But not everything was great. One build was so dark he could barely use it. Kimi Code did better on that one.

For an orbit diagram, GLM 5.2 made a nicer version. Some abstract creations felt disconnected from the prompt.

Overall, Julian says Grok Build genuinely surprised him. It's good for many things, but not everything.

06Tips, Limits, and Mistakes to Avoid

Here are some tips from the episode:

  • Don't expect perfection—some builds are mediocre, like the dark game.
  • Pair it with other models for best results. Use Grok for some tasks, Claude for others.
  • Use your subscription—no need to pay for API if you already have X Premium Plus.
  • Let it self-check—the more you use it, the better it gets.
  • Don't switch everything over—it's not ready to replace all your tools.

Mistakes to avoid: Don't expect it to understand every prompt. Some builds were weird or broken. Also, don't ignore the CLI—it's quite useful if you have an agent OS.

07Who Grok Build Is For and Why It Matters

Grok Build is for anyone who wants to build software without deep coding knowledge. Julian says you don't need to be a coder. You just talk to it, and it gets things done.

It's especially useful if you already have a Twitter/X subscription. You get a powerful coding agent for free, in a way. It also works well inside an agent operating system, where it can share memory with other models.

Why it matters: This is the first time Grok has beaten Claude on tests. It shows that Grok is catching up. For content creators and developers, having more options is always good.

You can build games, websites, and more, then preview them in your workspace.

08How to Integrate Grok Build into Your Agent OS

If you have an agent operating system like Julian's, integrating Grok Build is easy. You add it as a CLI tool alongside Claude, Gemini, and others. You don't use the API—just your subscription.

Once integrated, you can select Grok Build in your workspace. It uses the shared memory system, so it instantly has context even on first use. That means you don't have to teach it about your projects from scratch.

You can preview everything you build, and it gets saved in history. You can come back to it anytime. This setup lets you control Grok Build, Claude, GLM, Kimi, and more from one dashboard.