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

A new study reveals that coding assistants like Claude Code and Codex significantly overestimate task durations. Codex can be off by as much as ten times the actual time required. Additionally, both systems rate their own performance about 20 percentage points too high. The findings indicate a lack of temporal awareness in these systems, leading to real challenges in overseeing long, autonomous tasks. This inaccuracy can impact the effectiveness of using these tools in larger projects, highlighting the need for improvements in their performance.