OpenAI Codex

OpenAI Codex (the CLI coding agent, distinct from the original 2021 Codex model) is an open-source coding agent, written primarily in Rust, that runs locally in a developer's terminal. Given a task described in plain English, it reads an entire repository, writes and edits code across multiple files, runs commands in a configurable sandbox, executes tests to validate its own changes, and can open a pull request directly — a substantially more autonomous workflow than simple inline autocomplete. It supports a genuinely developer-oriented feature set beyond basic code generation: AGENTS.md files for giving the agent persistent, repo-specific instructions, a .codex/skills directory for defining reusable custom capabilities, Model Context Protocol (MCP) server support for connecting to external tools and data sources, and multi-step planning for tasks that require more than one coordinated change. It's available across a wide range of surfaces beyond the CLI itself — a VS Code extension with 9.8 million installs, the CLI itself with more than 88,600 GitHub stars, a web app, an iOS app, and, since June 2026, availability on Amazon Bedrock for enterprise cloud deployment. Who it's for: developers who want an autonomous coding agent that works directly in their existing terminal and repository workflow rather than inside a separate proprietary editor, from individual hobbyists to engineering teams running heavier, parallel agent workloads. Strengths: genuinely open-source, meaning the code itself is inspectable rather than a black box, flexible authentication that lets developers use either an existing ChatGPT Plus/Pro/Business/Enterprise subscription or bring their own OpenAI API key, and real breadth of deployment surfaces (CLI, VS Code, web, iOS, Bedrock) rather than being locked to one environment. The 2026 shift to token-based credit pricing (tied to input tokens, cached input tokens, and output tokens rather than flat per-message counting) gives more granular, usage-proportional billing. Limitations: pricing scales meaningfully with usage — Free covers only first-time exploration and simple local tasks, Plus runs $20/month for part-time use, Pro 5x is $100/month for full-time developers, and Pro 20x reaches $200/month for engineering leads running heavy parallel agent workloads, so serious continuous use carries real ongoing cost. As with any autonomous coding agent operating across a full repository, unsupervised multi-file changes still benefit from careful review before merging, particularly on complex or safety-critical codebases. Real-world use cases: a developer describing a bug fix in plain English and letting Codex CLI locate the relevant files, make the fix, run the test suite, and open a PR automatically, a team defining reusable skills in .codex/skills for repo-specific automated tasks, and an enterprise running Codex through Amazon Bedrock for governed, cloud-hosted agent access. For developers who want an open, terminal-native autonomous coding agent, Codex CLI is one of the most widely adopted options in 2026.