In applications like Claude's Cowork desktop tool or OpenAI's Codex coding platform, users often opt for AI agents to seek confirmation before proceeding with significant actions. This approach promotes security, yet it frequently leaves the agents idle while pending user authorization for subsequent moves.
Observing the Cowork or Codex interfaces directly reveals when an agent requires input. However, during multitasking or when focused elsewhere, individuals might overlook an inactive agent lingering without progress.
OpenAI's recently introduced Codex AI companions address this issue effectively. These animated figures position themselves in the desktop's edge, featuring thought bubbles that indicate the current activities of Codex AI agents.
Beyond basic updates, the companions excel in providing clear visual signals for agent needs. A thought bubble shows a red clock icon when an agent pauses for approval on a command. Completion of a task with readiness for new directives is marked by a green checkmark.
The red clock icon specifically signals the need for user approval to proceed.
These companions remain visible above other windows and can be repositioned freely; many users place theirs in the screen's lower-right area.
The designs are endearing, offering eight options including the standard blue Codex figure, the duck-like Dewey, Rocky for enthusiasts of Project Hail Mary, and even the ominous BSOD variant.
To activate a companion in Codex, users input the /pet command. Selection of alternatives occurs via the Appearance menu. Custom creations involve installing the hatch pet skill using $skill-installer hatch-pet, followed by invoking it with $hatch-pet to generate a personalized version, such as one drawing from recent projects.
Although not revolutionary—Claude Code's terminal features similar buddies—these companions prove practical. They keep users informed about background AI operations, prompting desires for comparable features in additional agent-based desktop software.
Ben, with over two decades in consumer technology journalism, now specializes in AI's impact on everyday life. His work examines cutting-edge large language models and their applications in professional and personal settings to aid adaptation to AI advancements. He emphasizes that AI will transform daily routines faster than anticipated, advocating daily engagement as the optimal preparation strategy. Joining PCWorld in 2014, Ben initially covered topics from laptops to security devices before establishing the publication's AI focus. His contributions also feature in PC Magazine, TIME, Wired, CNET, Men's Fitness, Mobile Magazine, and others. He possesses a master's in English literature.