01What Is Grok Build and How Does It Work?

Grok Build is a coding agent and terminal tool from xAI that lets you build software using simple text commands. It comes with Twitter subscriptions, so if you already pay for X Premium or similar, you can use it without extra cost. In this episode, Julian Goldie shows how Grok Build outperformed many other models on the same prompts.

How it works is simple. You type a command or a description of what you want. Grok Build plans the task, splits it across up to eight sub-agents that work in parallel, and then checks its own work.

If the output is good, it ships it. If not, it tries again. This loop runs until the job is done.

That is why one sentence can turn into a full game or website.

Grok Build is built into the CLI, but you can also plug it into AI agents like Hermes or OpenClaw. That means you can automate images, videos, voice notes, and even live calls with your agents. It is a lot of fun and surprisingly powerful for something that comes with a subscription you might already have.

02Main Features of Grok Build: What You Get

  • Terminal coding agent: You can use it directly in your command line to write and run code.
  • Image and video generation: Create visuals and B-roll automatically from one sentence.
  • Voice notes: Generate audio messages or narrations for your projects.
  • Live agent calls: Talk to your AI agents on a live call, which is a unique feature.
  • Parallel sub-agents: Up to eight sub-agents work at the same time to speed up builds.
  • Self-checking system: The agent reviews its own work and retries until it is good.
  • Xsearch integration: Search Twitter (X) live, something most LLMs cannot do.

These features make Grok Build stand out. It is not just a chatbot. It is a full build system that can produce websites, games, videos, and more.

Julian was impressed because he does not code, yet he built a complete open-world game in a few hours.

03How to Get Started with Grok Build: Step-by-Step

  1. Make sure you have a Twitter subscription. Grok Build is included with it, so no extra payment is needed.
  2. Open your terminal or command line interface. If you use an AI agent platform like Hermes, open that instead.
  3. Select Grok Build as your model. In Hermes, you can switch between Grok and other models like Minimax M3.
  4. Type your command or prompt. For example, you could write "build a voxel racer game" and let Grok handle the rest.
  5. Wait for Grok to plan, spawn sub-agents, and build. You will see the output land in your workspace.
  6. Preview your result. You can test the game, website, or video and give feedback to improve it.

That is all it takes. If you already have a Twitter subscription, there is no setup cost and no need to pay for APIs. Julian noted that he has not hit any token limits, which is a big bonus for heavy users.

04How to Use Grok Build: Examples from the Episode

Julian showed several real uses of Grok Build. First, he built two games: a voxel racer and an open-world RPG. He does not code, but the games looked and played well.

He compared the voxel racer to outputs from other models and said Grok's version was one of the best.

Another use is creating videos. Julian made a six-minute-long, fully AI-generated video. Grok Build created the avatar, generated B-roll, wrote the script, and narrated it with a voice.

All of that came from one sentence. He calls this the Video Director Agent, and it runs inside his Mission Control system.

You can also build websites with Grok Build. Julian showed simple but nice-looking websites that the tool can generate. And because it supports Xsearch, you can search Twitter live.

That is rare for AI models and useful for research or content creation.

Finally, you can use Grok Build inside an AI agent operating system. Julian plugs it into his Agent OS, so when he needs lots of tokens or wants to build quickly, he can use Grok. It works well as a secondary model, especially if your main plan runs out of tokens.

05Grok Build Test Results and Comparisons on GoldieBench

Julian ran Grok Build on his own benchmark, GoldieBench, to see how it compares to top models. On a leaderboard of five models, Grok Build scored about five out of ten. That is not the top, but it is not terrible either.

He compared outputs side by side with Opus 4.8 and GLM 5.2.

Opus 4.8 was still the clear winner in terms of quality. Julian called it "goated" and said he could stare at it all day. GLM 5.2 also produced nice results.

Grok Build's output was decent, but Julian said it is "not bad, but it's not the best." That seems to be a consistent pattern.

However, on the Voxel Racer test, Grok Build actually produced the nicest game compared to other models. So it can win on specific tasks. In terms of performance, Grok Build scored 70.8% on SWE-bench and ran at 100 tokens per second.

The context window is 2K, which is small but okay for many tasks.

Julian also noted that he has not hit any token limits, which is a big plus for a subscription-based tool.

06Tips, Limits, and Mistakes to Avoid with Grok Build

  • Use it as a backup, not a replacement: Julian says it is not as good as Claude 4.8, so keep your main model for top-quality work.
  • Great for token-heavy tasks: If you run out of tokens on Claude, switch to Grok Build to keep going.
  • Small context window: The 2K context window may limit very long prompts, so keep instructions concise.
  • Always review output: Even though Grok self-checks, Julian recommends previewing results and giving feedback.
  • Don't expect the best quality: It scores around 5 on GoldieBench, so it is good but not the top performer.
  • Take advantage of the free inclusion: If you already have a Twitter subscription, you get all this without extra cost.

Avoid the mistake of assuming it is a joke because of past Grok versions. Julian says it has dramatically improved. Also, don't ignore its unique features like Xsearch, which most LLMs lack.

That alone can be a reason to use it.

07Who Should Use Grok Build and Why It Matters

Grok Build is for anyone who already has a Twitter subscription and wants to build software without paying extra. It is especially good for non-coders because you can create games, websites, and videos with simple sentences. Julian, who is not a coder, built an open-world game in a few hours.

It also matters because it gives you access to a powerful CLI and agent integration that was previously only available through paid APIs. If you are on a budget, this is a great way to experiment with AI agents and automation. You can plug it into systems like Hermes, OpenClaw, or an agent operating system.

Julian believes this is the future: models that plan, split work, check their own output, and keep improving. Even if Grok Build is not the absolute best, it is a step in that direction. And because it comes with your existing subscription, there is no reason not to try it.