OpenAI's Codex goes GA with Slack integration and a dev SDK

OpenAI is taking the training wheels off. Codex, its cloud-based coding agent, entered general availability this week, a move that comes with a new Slack integration, a public SDK, and a suite of admin tools designed to make the agent a fixture of enterprise engineering workflows. It’s a clear signal that the research preview phase is over and the push for real revenue has begun. The numbers back up the ambition. The company says daily Codex usage has surged more than 10x since early August, and the freshly launched GPT-5-Codex model has already churned through over 40 trillion tokens in just three weeks, making it one of OpenAI’s fastest-growing models ever. Internally, the tool is no longer optional—nearly all OpenAI engineers now use it, up from just over half in July, and they’re reportedly merging 70% more pull requests each week.

The headline feature for many teams will be the Slack integration. You can now tag @Codex directly in a channel or thread, treating it like a colleague you can delegate grunt work to. The agent automatically parses the conversation for context, spins up the right environment, and replies with a link to a completed task in the Codex cloud. From there, you can merge, iterate, or pull the task locally. It’s an attempt to weave the agent into the social fabric of software development, not just the IDE.

For developers who want to bake this into their own toolchain, the new Codex SDK is the more significant release. It lets you embed the same agent that powers the Codex CLI into custom workflows and apps, with structured outputs and built-in context management for resuming sessions. The company also tuned the agent’s prompt and tool definitions specifically for GPT-5-Codex to improve speed and accuracy. The SDK is available in TypeScript to start, and there’s a new GitHub Action for CI/CD pipelines. The early returns from customers like Instacart show the agent automatically cleaning up tech debt and dead code, while Cisco reports slashing pull request review times by up to 50%.

On the admin side, ChatGPT workspace managers now get environment controls to delete sensitive data, enforce safer defaults for local usage, and monitor everything through analytics dashboards. The offering is available to Plus, Pro, Business, Edu, and Enterprise plans starting today. But the honeymoon period for unlimited experimentation is winding down. Starting October 20, Codex cloud tasks will begin counting toward your usage limits. The era of the AI coding copilot is quickly becoming a line item on a monthly bill. It’s a maturation that feels inevitable, even if the long-term productivity gains still need to prove themselves out in a way that transcends the impressive, but often cherry-picked, case studies.