OpenAI has introduced Codex, a new cloud-based software engineering agent designed to automate a wide range of development tasks simultaneously—marking a significant step toward AI-powered software automation. Built on a custom model dubbed codex-1, Codex is trained using reinforcement learning on real-world software engineering workflows. The result is an agent that doesn’t just respond to prompts—it completes tasks independently in a sandboxed environment, isolated from other sessions and preloaded with the target codebase.
Unlike many AI dev tools that rely heavily on real-time user prompting or browser plugins, Codex operates more like a self-directed contractor. Developers can assign it jobs such as writing new features, debugging, generating pull requests, or answering codebase-specific questions, and it will carry them out in parallel—typically within 1 to 30 minutes depending on task complexity. It can also produce logs or other proof of completion, giving developers the ability to verify its output. Progress tracking is available, but there’s no real-time feedback loop: once the job starts, Codex runs independently until it finishes.
Codex can be accessed directly from ChatGPT—currently for Pro, Team, and Enterprise users—via a new “Code” option that appears after entering a relevant prompt. Support for Plus and Edu tiers is in the pipeline. For now, OpenAI is offering Codex access as part of a research preview, with no added cost during the rollout. Rate limits and paid access tiers will follow in future updates.
However, Codex comes with caveats. It doesn’t support image inputs, meaning frontend-heavy workflows like UI mockups or visual bug fixes are out of scope. There’s also no in-session correction—once Codex begins a task, users must wait for it to finish or cancel it altogether. Tasks requiring long-term memory or multi-hour execution windows are similarly off-limits, at least in this early phase.
In parallel with Codex, OpenAI also updated Codex CLI, its local open-source terminal tool designed for quicker, lightweight dev tasks. The CLI now runs a trimmed-down variant of codex-1, optimized from the o4-mini model. It’s built for rapid Q&A and real-time code editing at low latency, making it a handy companion for shell-based workflows.
Codex CLI now integrates directly with ChatGPT accounts. Instead of configuring API keys manually, developers can sign in and select their organization from within the CLI. As part of the launch, ChatGPT Plus and Pro users who log in via Codex CLI will receive $5 and $50 in free API credits, respectively, valid through the next month.
OpenAI’s Codex initiative moves beyond code completion into the territory of autonomous task agents—a category that could reshape how development teams manage everyday coding work. It’s early days yet, but with cloud integration, tight ChatGPT coupling, and CLI support, OpenAI is clearly building out an ecosystem where AI engineers don’t just assist—they actually ship.