OpenAI Codex goes GA with Slack bot and new SDK

Delegating a code change usually meant firing up a dedicated editor, pulling the repo, and writing a wall of prompts. OpenAI now wants you to ping an AI teammate right where your team already chats.

OpenAI announced the general availability of Codex, moving its flagship coding agent out of research preview. Alongside the GA rollout, the company introduced three major upgrades: a native Slack integration, an official Codex SDK, and a suite of workspace admin controls.

OpenAI also shared some staggering adoption metrics. Daily usage of Codex has jumped more than 10x since early August, and its underlying GPT-5-Codex model served over 40 trillion tokens in just three weeks post-launch. Inside OpenAI, virtually all engineers now use Codex on a daily basis, up from barely half in July. Those engineers merge 70% more pull requests every week, with Codex auto-reviewing nearly every single PR before production.

Delegating code in Slack and pipelines

The most noticeable change sits inside Slack.

Engineers can now simply tag @Codex in a channel or thread to hand off an engineering task. The agent reads the conversation for necessary context, spins up the right cloud environment, and drops a link to the completed task. From there, developers can review the diff, keep iterating, or pull the branch down locally via the terminal.

Beyond chat, OpenAI is embedding Codex deeper into the developer stack. The new Codex SDK, which debuts with TypeScript support, allows engineering teams to embed the exact same agent powering the Codex CLI into their internal apps with a few lines of code. It features built-in context persistence and structured outputs. A new GitHub Action and CLI execution command also make it straightforward to run Codex directly inside CI/CD pipelines.

Customers are already leaning heavily into these workflows. Cisco reported cutting complex pull request review times by up to 50% using Codex as a dual checker for both human and machine code. At Instacart, the engineering team integrated the Codex SDK into Olive, their internal background coding platform, allowing developers to trigger end-to-end bug fixes and clear technical debt like expired experiments with a single click.

To keep enterprise IT happy, ChatGPT Business, Edu, and Enterprise admins gain new dashboards to track usage across CLI, IDE, and web interfaces. Admins can also enforce security overrides for local developer environments and wipe unused cloud environments to prevent credential leaks.

Slack and SDK access roll out immediately to ChatGPT Plus, Pro, Business, Edu, and Enterprise subscribers, while admin tools are restricted to organization tiers. Starting October 20, cloud tasks will officially count against plan quotas.

Why it matters

Codex hitting GA signals that AI coding is finally moving from individual auto-complete to asynchronous team collaboration.

For engineers, this shift is genuinely useful. Instead of staying tethered to an IDE prompt box, you can assign tedious chores—like pruning dead code or running initial PR triage—directly from a Slack thread or a CI hook. If your team already runs on GitHub and Slack, this is worth setting up today, especially for catching regressions before human reviewers step in.

For the AI landscape, OpenAI is clearly aiming beyond Cursor and GitHub Copilot's editor strongholds. By owning the agent loop, the SDK, and the collaboration interface in Slack, OpenAI is trying to position Codex as the operating layer for software development rather than just a smart typing assistant.

That said, handing the keys of your codebase to an agent running in cloud environments brings clear risks. While the new admin controls address basic data leakage concerns, letting an autonomous agent review code that another agent wrote can create a dangerous illusion of security. Pumping PR volume by 70% is only a win if that code does not quietly accumulate subtle technical debt down the road.