Back to Flash News

Codex Pets Popup Can Reply Directly

After the update of OpenAI's Codex Pets desktop version, users can reply to specified content directly in the pet popup window without opening the main interface, supporting both voice and text. The pets are optional pixel-style floating layers that still display the current thread, running status, waiting for input, or review status when the main window is minimized, with bubble prompts for progress; clicking sends the reply back to the agent. There are about eight built-in characters, and the command “/pet” can wake or hide them, also accessible via the command panel and settings.

This feature will be launched with the desktop application in May 2026, designed as an environmental indicator for long-running agents. Users can use “/hatch” to hatch local images into custom pets, stored in the Codex main directory. The community subsequently created usage rings, nest widgets, and reading plugins. Issues on Windows included misalignment of clickable hot zones and characters, and inability to drag.

When the agent modifies the repository in the background, the user is in another window. The popup keeps the approval authority at the top of the desktop, reducing the number of application switches. Voice and text are parallel, integrating earphone scenarios and keyboard scenarios into the same light channel.

Visibility of status does not equal visibility of process. The smaller the window, the easier it is to only reply with “continue” or “stop”.

The market mechanism is the duration of stay on the desktop. The buy side consists of subscription users who need to handle multiple tasks in parallel while fearing idle agents; the sell side is OpenAI, which aims to make the coding agent a permanent layer. Funds are already in the paid tier of ChatGPT, and pets are retention accessories. The beneficiaries are individual developers who reduce window switching; the pressured side is code review that must open the complete diff to be responsible. The event-driven aspect is an interactive redesign, not a new foundational model.

The pets notify first, then receive commands. The main window becomes an exception, not the default.

Source: Public Information

ABAB AI Insight

OpenAI addresses the biggest experiential gap in agent products with a pixel pet: tasks are running, but the user has left the window. The bubble indicates status, and clicking to reply signifies control. Voice further shifts control from keyboard to mouth, suitable for when wearing headphones and suddenly being asked while writing something else. By May, the value will already be environmental awareness, and this time it transforms one-way prompts into a two-way light channel. Custom hatching and third-party nest skins make the ecosystem easier to spread than the functionality itself.

The capital path is subscription retention. The coding agent charges based on usage and seats, while pets reduce the friction of "opening the application," thereby increasing the number of delegated tasks. The hot zone defects indicate that the floating layer still has engineering debts under multi-monitor and scaling conditions. When review is compressed into a single reply, the boundary of responsibility becomes blurred: the continue button replaces viewing the diff. If companies enforce complete interface approval, pets will only serve individual seats.

Analogies include instant messaging floating window replies and system tray resident assistants. The development tool phase is about controlling the presence of asynchronous agents. Whoever can obtain a user's command without interrupting foreground work can run the agent in the background overnight. The cute character is a shell for obtaining permissions, while permissions are the product.

The structural change is the interactive layer in technological substitution. The mechanism markets the attention of the IDE main thread, partially giving way to the micro-channel at the top of the desktop. Pricing power remains with the model and quotas, while experience power lies with whoever occupies the screen corners first. If the process only exists in the main window, the popup will only accumulate "continue" replies.

ABAB News · Cognitive Laws

  1. The biggest lack of background agents is not capability, but a command that doesn’t require switching windows.

  2. The smaller the window, the easier it is to only reply with continue.

  3. The cute shell applies for the permission to remain on the screen.