
OpenAI has been steadily positioning its Codex system as more than just a coding assistant.
In order to move it towards a broader "agent" model that can take on tasks independently and operate across tools, files, and workflows, OpenAI is introducing an AI-generated 'pets' into the Codex app. This particular feature stands out, not because it changes what the system can do, but because it changes how users experience it.
The pets are small, animated companions that sit on top of the interface as an overlay.
They do not write code or affect the outcome of tasks.
Instead, they act as a lightweight communication layer between the user and the agent.
They can display what Codex is currently working on, notify the user when a task is finished, and signal when input is needed, all without requiring the user to switch windows or interrupt their workflow.
Pets. Now in Codex.
Use /pet to wake your pet. pic.twitter.com/aAm4lLP4LW— OpenAI Developers (@OpenAIDevs) May 1, 2026
Functionally, this reflects a practical challenge with agent-based systems.
Codex is designed to run tasks asynchronously, sometimes taking minutes to complete work in the background. That creates a visibility gap.
Users delegate work, but then need a way to monitor progress without constantly checking logs or switching context. The pet interface appears to be a response to that problem, offering a persistent, ambient status indicator rather than a traditional notification system.
To create one of their own, users need to install the hatch-pet skill first, to then use the /hatch command to create the pet.
To create your own pet, install the hatch-pet skill: https://t.co/8TF9SLmV0H pic.twitter.com/TtvUPj9kug
— OpenAI Developers (@OpenAIDevs) May 1, 2026
The feature is optional and can be toggled on or off.
Users can also customize or generate their own pets, suggesting that OpenAI is experimenting with personalization as part of the developer experience.
Early implementations include simple commands like /pet to summon or hide the companion, and variations that allow users to create different visual styles or behaviors.

While the surface presentation is playful, the design choice points to a broader shift in how AI tools are being shaped.
Coding environments have historically emphasized precision, minimalism, and control. Adding animated companions introduces a layer of personality that is more commonly associated with consumer apps or games. Some observers have noted that this borrows from engagement mechanics seen in other software categories, where visual feedback and emotional cues are used to maintain attention and reduce friction.

At the same time, this raises questions about the balance between utility and distraction.
After all, most Codex users are developers, programmers, and software engineers, along with other technically inclined users who are deeply familiar with programming languages and syntax.
For some of those users, a persistent animated element may make background processes more legible and reduce cognitive load. For others, it may feel unnecessary in a professional tool where clarity and efficiency are the priority. The fact that the feature is optional suggests that OpenAI is aware of these tradeoffs and is testing how far interface experimentation can go without undermining the core use case.
The introduction of pets also fits into a larger trend in AI development, where systems are becoming more autonomous and less conversational. As tools like Codex take on longer, multi-step tasks, the interface challenge shifts from prompting to supervision. Users are no longer just asking for outputs, they are overseeing processes.
Learn about Codex pets before adopting your own:https://t.co/rscKfBIErG
— OpenAI Developers (@OpenAIDevs) May 1, 2026
In that setting, status visibility becomes as important as capability, and the way that visibility is delivered can shape how the system is perceived.
In practical terms, the pets do not expand what Codex can do. They only change how its activity is surfaced. That distinction matters to some users because it highlights where much of the current innovation is happening. Not only in model performance, but in the layer between the model and the user, where interaction patterns are still being defined.