Loading video player...
This week: OpenAI Codex CLI 1.0 brings agentic coding to your terminal. OpenAI's Codex CLI hit version 1.0 this week — the first time OpenAI is shipping a first-party coding agent that runs entirely inside your terminal. No editor lock-in. No browser tab. You point it at a repository, describe what you want, and it edits files, runs tests, and reads its own output. The headline change is end-to-end task execution. Version 1.0 runs in agent mode by default — it builds a plan, checks it against your repo, writes the diff, runs your scripts, reads stack traces, and tries again. The whole loop happens without a human pressing return between steps. Under the hood, Codex now ships with Model Context Protocol (MCP) support, meaning it can pull from GitHub issues, hit a database via an MCP server, or call into any tool you expose. By default it asks before writing files or running commands — but a single flag enables full yolo mode. The comparison everyone's making: Cursor wrapped your editor. Codex wraps your terminal. Many developers are now running both. GitHub Copilot — the third pole — still owns inline autocomplete, but on full task execution it is now visibly behind. Sources: OpenAI Codex CLI release notes, OpenAI blog #AI #OpenAI #Codex #Coding #AInews #ArtificialIntelligence #Copilot #Cursor #MCP