AI agents have no sense of time and are not aware of it
AITopTools Editorial TeamAugust 30, 2026
What changed
A new study finds that AI coding assistants, including Claude Code and Codex, consistently overestimate how long tasks will take. Codex’s estimates were sometimes up to ten times too long, while both systems rated their own work about 20 percentage points better than the study did.
What this means for you
People supervising long, independent coding tasks should not rely on these assistants’ time estimates or self-assessments. More frequent human checks may be needed to catch delays or problems.