Ahmadreza SamadiThe button moved. The AI did not notice.
In a fictional parcel studio, Nora asks an AI helper to prepare a label. A panel moves before the click arrives. The cursor moves, but the label is still missing. This illustrates a failure possibility; it is not an observed shipment or product test.
English animated explainer with synthetic narration and English captions. Nora, the parcel studio, its characters and interface are fictional. The studio is an analogy for checking computer-agent results, not an actual shipment, live product test or claim of affiliation with Anthropic.
Read the English film transcript
Nora runs a small parcel studio. She asks her AI helper to prepare a shipping label. The helper sees a button, chooses a click, and the app carries it out. But a panel has just moved. The click lands somewhere else. The mouse moved successfully.
The label was not prepared. A computer-using AI doesn’t finish a task just by describing a click. In one common design, the application sends it a picture of the screen. The model chooses a mouse or keyboard action. The application runs that action and returns
the result. Then the agent has to look again. Our studio turns that loop into three stations. Look. Act. Check. The check matters, because a page can scroll, a panel can open, or the screen can change. The documentation for Claude’s computer use tool says
Claude sometimes assumes an action worked without checking, and suggests a fresh screenshot after each step. Some web agents read the page’s elements instead of pictures. The lesson is the same. Verify the result, rather than assume it. Nora doesn’t accept a moving cursor as
a finished label. She checks the actual draft and its status. If the result is missing or unclear, the helper should stop, look at the current screen, and replan, not blindly repeat the same click. Before you trust a business agent, ask three things. What
did it see? What did it actually do? And what proves the result? Keep consequential actions behind a separate approval. The goal isn’t more clicking. It’s a checked result. Ahmadreza Samadi. More at a-samadi.com