AI coding agents struggle with sense of time
ai coding assistants like claude code and codex can't predict task duration or track elapsed time, causing issues with long-running jobs.
a new study by two independent ai researchers found that popular coding assistants, anthropic's claude code and openai's codex, have no sense of time. the agents were tested on 218 tasks from programbench and the researchers' own benchmarks, and consistently overestimated task duration. claude code was off by three times on average, while codex was off by six to ten times. the study also found that the agents' performance varied greatly depending on their software setup. the researchers emphasize that an agent's ability to self-assess and track time is crucial for working reliably on long tasks.