OpenAI Codex Thursday update has arrived, with a large amount of information this time. The core is one sentence: Codex is continuing to move from a "code assistant" to a "long-term autonomous engineering agent". This time, there are 5 main updates: 1. Appshots: New Context Input Method Press Command+Command on Mac to directly attach the current application window to the Codex thread. Codex not only receives screenshots, but also the complete text in the window, and even the content that cannot be seen on the screen after scrolling. This feature has been launched for Mac users of all packages, and the enterprise version will soon be available. 2. Graduating in the Goal mode officially The previous/oal is no longer an experimental function. Now Codex can continuously advance around a goal, and both large and small tasks can be dismantled and executed by themselves. The official says it can work continuously for several hours or even days. Supports Codex App, IDE extensions CLI。 You can check progress, adjust direction, and pause tasks at any time. A very practical trick: Open side chats and ask what it has already done and what it plans to do next without interrupting the main task. 3. Advanced browser annotation mode upgrade The browser in the Codex app has become more user-friendly. Add advanced annotation mode, which allows you to directly adjust elements on the page, preview changes in real-time, and then submit a batch of feedback to Codex. This is very useful for front-end and design collaboration. Previously, you had to say 'here to the left, change this button, this spacing is not correct', which was very abstract. Now you can get closer to 'mark it directly on the page'. 4. Team plugin sharing The team can now distribute custom plugins and internal tools across workspaces, and manage permissions uniformly. Business users are already available, and enterprise users can contact OpenAI to apply for early access. This is actually crucial. Because the true competitiveness of Codex in the future is not just whether the model can write code, but whether it can integrate with the team's internal toolchain, permission system, workflow, and private context. 5. Enterprise level Analytics Upgrade The commercial and enterprise versions of Codex have more detailed data analysis, including: Active users, credit point consumption, token usage, number of runs, user ranking, number of generated code lines, plugin usage, etc. The Analytics API has also been updated to facilitate team statistics and monitoring of Codex usage within the organization. This indicates that OpenAI has begun to seriously push Codex towards the direction of enterprise engineering infrastructure.