Use a desktop agent to move files with checks

A user drove a desktop agent through file moves in real apps, and learned that checkpoints matter when the interface lies about success.

Done with Claude CoworkOpenClaw· Intermediate· Apr 2026

What they did
They spent months testing desktop agents like Claude Cowork on real applications. For file handling, the agent read whatever the app exposed in the accessibility tree; when that broke or collapsed into a canvas, it switched to screenshots, OCR, and direct clicks by position. In one 12-file move, it hit an unexpected modal, misunderstood it, and kept claiming the job was done.
How it went
The move task failed completely: after about 20 minutes, the agent said it had moved all 12 files, but none had actually been transferred. The larger takeaway was that stricter instructions and verification steps can reduce mistakes, but they do not solve poor visual understanding.
Tips
Use dry runs and make the agent confirm the destination contents after each risky step. Avoid apps with unreliable accessibility trees, and force a stop for any unexpected dialog before it continues.

Try it yourself with Claude Cowork

You are helping me manage files in [app name] on my desktop. Move the items from [source folder or list] to [destination folder], but do it in small steps and verify each step by checking the destination contents before proceeding. If the app shows any unexpected dialog, warning, or modal, stop immediately and tell me exactly what it says. Before you delete, overwrite, submit, or close anything that could cause data loss, pause and ask for my approval. At the end, report how many items you moved and how many you verified.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on reddit.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.