The Terminal Is the New IDE. Every Major Lab Agrees.

xAI launched Grok Build on May 14. With that, every major AI lab now ships a coding agent that lives in your terminal. The competition isn't "can we build one" anymore. That question is settled.

The lineup: Anthropic has Claude Code. OpenAI has Codex CLI. Google has Gemini CLI plus Antigravity. xAI has Grok Build ($99/mo introductory, $300/mo regular for SuperGrok Heavy subscribers). All four share the same architecture. CLI entry point. Parallel subagent execution. Plan-then-execute workflows. MCP and plugin support. Convention-file reading (CLAUDE.md, AGENTS.md). The convergence is remarkable. Four independent engineering teams arrived at the same design.

I use Claude Code in my personal projects every day. When I looked at Gemini CLI and Codex CLI's architecture, the resemblance was uncanny. Same terminal UI patterns. Same file-editing primitives. Same approach to project context through convention files. The differentiation now is cost, parallelism, and model quality. How many subagents can run at once? How many tokens per task? Which model writes better code for your specific stack?

The IDE isn't dead, but the center of gravity shifted. Cursor 3.5 shipped multi-repo automations this week. Windsurf 2.0 bundled Devin Local directly into the editor. The IDE vendors see the same thing: the terminal agent is becoming the primary interface, and the IDE is becoming the visualization layer that wraps around it.

Meanwhile, the tools are racing toward persistent autonomy. Claude Code's new /goal command lets the agent work across session breaks without re-prompting. OpenAI's Codex CLI stabilized Goal Mode with workflows that self-schedule across days. Cursor 3's /best-of-n runs the same task across multiple models in isolated git worktrees and lets you pick the best output. We're past the point of "type a prompt, get code back." These are autonomous systems that run for hours.

What to do: try at least two CLI agents on the same task. The differences are material. If you're on Claude Code, keep an eye on Gemini CLI's free tier for scaffolding work. Running a free agent for boilerplate and a paid agent for complex logic is a pattern that makes economic sense as token costs become real.