A Terminal-First AI Coding Agent for Complex Projects
xAI has unveiled Grok Build, an AI coding agent that runs directly from the terminal and is aimed squarely at professional software engineering, application development, and workflow automation. Branded as an “agentic CLI,” Grok Build is designed to manage complex coding workflows via natural-language prompts, from generating implementation plans and editing files to executing shell commands and handling dependencies. A dedicated plan mode tackles a frequent pain point with coding assistant tools: loss of control. Before any code changes occur, the agent presents a full execution plan that developers can approve, comment on, or rewrite step by step. Once a plan is locked in, Grok Build surfaces every subsequent change as a clean diff, giving developers a clear before-and-after view of modifications. The tool is currently in early beta and is available exclusively to SuperGrok Heavy subscribers through a single-command installation tied to their accounts.
Parallel Subagents and Arena Mode: xAI’s Architectural Bet
The headline feature of Grok Build is its ability to spawn up to eight specialized subagents that work in parallel. Instead of a single AI agent marching through tasks sequentially, Grok Build splits large, multi-file problems into smaller units that can be tackled simultaneously. Subagents can operate inside separate Git worktrees or branches, keeping lines of work isolated so they do not overwrite each other while experimenting. This architecture underpins a forthcoming Arena Mode, where multiple agents will compete on the same problem and rank their own outputs before presenting results to developers. While Arena Mode is not yet live in the early beta, xAI has confirmed it is in the roadmap, and code traces surfaced it earlier this year. Together, parallel subagents and Arena Mode signal xAI’s focus on speed, redundancy, and self-evaluation for advanced software engineering tasks.
Local-First Design, MCP Support, and Workflow Integration
Grok Build is built with a local-first philosophy: code remains on the developer’s machine during sessions, and the tool is compatible with air-gapped environments once it has been set up. That matters for teams working with proprietary codebases or under strict compliance rules. The AI coding agent is also designed to slide into existing toolchains rather than replace them. When launched inside a project folder, Grok Build automatically recognizes local conventions, AGENTS.md instruction files, plugins, hooks, skills, and Model Context Protocol (MCP) servers. MCP support is particularly notable as it allows Grok Build to connect to external services and aligns with broader adoption of the open standard in modern IDEs. Beyond the CLI, Grok Build integrates with VS Code for those who want a graphical front end and offers a headless mode for embedding the agent in scripts, CI pipelines, and broader automation workflows.
Pricing, Access, and the Grok-Code-Fast-1 Model
Access to Grok Build is currently restricted to SuperGrok Heavy subscribers, with xAI pricing that tier at USD 299 (approx. RM1,390) per month and promoting an introductory offer of USD 99 (approx. RM460) per month for the first six months. Under the hood, Grok Build is powered by grok-code-fast-1, a coding-focused model trained on programming content and real-world pull requests. xAI reports that the model scores 70.8% on SWE-Bench Verified using its internal evaluation setup and supports a 256,000-token context window, allowing it to keep large codebases in memory throughout a session. The model is priced at USD 0.20 (approx. RM0.93) per million input tokens, positioning it competitively among AI coding assistant tools. xAI is using the beta period to gather direct feedback from developers, who can submit bug reports and feature requests without leaving the CLI.
Competing with Claude Code and Codex in the Developer Tooling Race
Grok Build marks xAI’s first serious push into the professional coding market, where incumbents like Anthropic’s Claude Code and OpenAI’s Codex already command significant traction. Codex has surpassed three million weekly active users, while Claude Code has become a primary growth engine for Anthropic, contributing to annual recurring revenue cited in the tens of billions. Grok Build enters this landscape as a terminal-native alternative focused on agentic workflows, parallel subagents, and strong local control. Features such as plan mode, clean diffs, Git worktree integration, and support for MCP servers and ACP-based bot orchestration are aimed at developers who want both automation and reviewability. While Grok Build lacks the adoption history of its rivals, xAI is betting that professional developers will value its combination of local-first design, high-context coding model, and flexible CLI integration enough to trial it as a primary AI coding agent.