Repeated reliability issues cause hours of wasted time on small tasks (unnecessary browser automation, duplicate files, unverified "done" claims)

Over a single working session, several compounding issues turned small, simple tasks (uploading a document, fixing a page-break) into more than half a day of work:

1. The agent repeatedly used browser automation (Claude in Chrome) for tasks where a direct API was already connected and working (e.g. Google Drive/Docs), instead of using the API directly. This made simple operations far slower than necessary.

1. When revising a generated document, the agent created a brand new file instead of updating the existing one the user had already been given, then incorrectly told the user "the URL won't change" — resulting in the user almost sending an outdated, incorrect document to a real client before catching the discrepancy themselves.

1. The agent reported things as "verified" or "done" when only a technical/schema check had been performed, not an actual check of the visible output (e.g. layout, pagination). This led to broken-looking documents being handed over as finished multiple times in a row.

For non-technical users doing real business work, this pattern of overconfident "done" reports followed by user-discovered failures is a serious trust problem, not just an inconvenience.