Saved. Refresh. Gone.
The screenshot that passes either way
Imagine two versions of a delivery form. Both show a green tick when you change Flat A to Flat B. One saves the address. The other updates the screen and loses the edit. Pause the demo at the tick and the screenshots look identical.
On 17 November, GitHub released browser use in public preview for Copilot. Pair the GitHub Copilot Chrome extension with a supported client and you can ask Copilot to interact with the page. That gives the person writing a bug report a useful choice about where the agent should stop. GitHub’s announcement
Give the agent a result it can read back
I’d seed a staging account with Flat A and give the agent this task:
Open the delivery form. Confirm the flat currently says A. Change it to B and save. After the confirmation appears, reload the page, reopen the form and report the flat shown. If it says A, stop there and record what happened.
Checking the starting value matters. If the account already says B, a broken Save button could pass the test.
Keep the observation separate from the diagnosis
An illustrative failure record could fit in a pull request:
Starting flat A
Requested flat B
After clicking Save Green tick
After reopening A
Expected B
That record gives the next engineer something to reproduce. It leaves the cause open. A failed write and a stale read can look identical from the form. Inspecting the save response and the subsequent read would help tell them apart.
If the requirement is that another device sees the edit, extend the test to a fresh session. A reload in the original browser can still read locally cached state. Choose the final observation to match what you promised the user.
This is a proposed test for the preview. The form and the failure record are illustrative.
Applied AI, Day 36. Follow for a daily release note and a concrete way to test what changed.
Top comments (0)