Grok Build opens the coding-agent harness layer

xAI open-sourced Grok Build, including the coding-agent harness and terminal UI. The signal is that agent runtimes are becoming inspectable product surfaces, not just hosted chat experiences. Why it matters: Builders can now study and compare the harness layer: planning flow, CLI/TUI ergonomics, subagent shape, extension points, privacy defaults, and local-first options. That separates agent-loop design from model choice.

Try this: Inspect the repo before installing anything. Look for the agent loop, privacy controls, extension model, and how the terminal UI handles planning and review. Treat this as harness evidence, not an open-weights model release.