ChatGPT sites now handled by Codex instead of Git and CI, adds team invites

OpenAI's developer account unveiled a feature on ChatGPT sites that lets multiple people build and deploy the same project together. When teammates are invited as editors, each person can push changes to the project, while Codex handles code version control (Git) and continuous integration (CI) behind the scenes. According to screenshots that were shared, the Codex sidebar already includes tabs beyond chat, such as Sites, Scheduled, and Plugins, along with a list of team-based projects and a model selection window at the bottom of the screen.

ChatGPT sites appear to be a project-based workspace where users write code within the Codex interface and carry it through to deployment. On August 11, OpenAI introduced syncing of projects, conversations, skills, and plugins between the ChatGPT desktop app and Codex, and on August 13 it released a "computer history" feature that picks up context from recent work and suggests skills for repetitive tasks. This latest announcement reads as an extension of that trajectory — an attempt to expand Codex from a personal work assistant into a team collaboration tool. Competitors are moving in the same direction around the same time. Anthropic has opened up a collaboration feature to early-access organizations that attaches a repository to Claude Code as persistent context and allows multiple threads to be split within a single project, a move that ties in with bringing the Claude tagging feature from Slack over to desktop. Both companies are effectively expanding coding agents from personal chatbots into work platforms that teams use together.

What changes in practice with this update is the workflow itself. Even when multiple people work on the same project, there's less need to worry about who should manage which branch or who should run tests before deployment. With Codex handling Git conflicts and build verification, teammates with limited development knowledge can now join as editors and help build and deploy sites directly. However, the announcement did not include details such as which pricing plans support the feature or any limits on team size.