I Audited My AI's To-Do List. A Quarter of It Was Already Done.
My coding agent has a to-do list. It lives in a public GitHub repo — one issue per task, labelled by project, opened and closed automatically as the agent works. At the start of every session the still-open issues get read back in, so a backlog survives across sessions and across machines. The agent picks up exactly where the last instance left off. This morning I asked it to reconcile that…
An audit revealed that a quarter of the open tasks for an AI agent were already completed, despite the system reporting them as "open." This discrepancy arose because the AI system creates tasks automatically when work begins, but closing tasks requires manual bookkeeping after the work is finished. As a result, tasks would remain in the backlog even after completion, leading to inaccurate reporting.
The audit confirmed the reality of completed tasks by checking external sources like email, calendar, and git tree, ensuring that only verified completion would be considered "done." This illustrates how systems where the completion signal is separate from the creation signal can produce unreliable self-reporting, and highlights the importance of verifying reality against these reports to maintain accurate task management.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
