What I like most is that the user and agent are working together towards the same goal. Instead of writing a long prompt explaining what to test, you can just show the agent in the browser.
And since the agent has the same browser controls as the user, it can take over, fill in missing steps, or clean up the messy parts of the recording. People click the wrong things, go back, change inputs, etc. The agent can make sense of all that instead of just blindly replaying every action.
I actually think UIs like this will become pretty common. Instead of having to do everything yourself or explain everything to an agent, you both work on the same thing, with the agent helping you get to the end goal with as few steps as possible.
Happy to answer questions about how we built it!