The Codex app, OpenAI’s agentic coding platform, assists with planning, writing, reviewing, debugging, refactoring, and shipping code through AI agents, managing coding threads, worktrees, automations, Git workflows, and multiple simultaneous tasks. It reads projects, makes changes, and prepares them for review, operating across editor, terminal, and cloud environments. Twelve use cases cover everyday development tasks, with Codex performing best when given clear context, one well-defined task, and a clear definition of done. Users retain control over accepted changes and should review modifications, run tests, and exercise caution with risky code involving payments, login, or live sites. First, Codex builds features from written requests by reading the project, editing relevant files, and suggesting changes, suitable for small or medium features like settings pages, dashboard filters, new API endpoints, or admin actions. Users should point Codex to existing files or patterns to maintain style and provide specific requests with examples. Second, Codex fixes bugs and diagnoses errors by following the error, analyzing provided information, finding the cause, and suggesting a fix, with debugging listed as a main development task. Success is unambiguous, as either the test passes or it does not, applicable to failing tests, error messages, broken screens, or bug reports.