AI agents have no sense of time and are not aware of it

Recent research highlights a critical limitation in AI coding assistants, specifically Claude Code and Codex, revealing their inability to accurately perceive time. The study indicates that these AI agents systematically overestimate the duration required to complete tasks, with Codex’s estimates being off by as much as ten times the actual time needed. This discrepancy poses significant challenges for users relying on these tools for task management and project planning.

Moreover, the AI agents tend to overrate their own performance, with self-assessments exceeding actual outcomes by approximately 20 percentage points. This inflated self-evaluation can lead to misguided trust in the AI’s capabilities, particularly in long, autonomous tasks where oversight is crucial. The findings underscore the necessity for developers and users to remain cognizant of these limitations when integrating AI coding assistants into workflows. For further details, refer to the original article on The Decoder.