The AI chat panel and Human in the Loop

Beside the live browser in the recorder is a chat panel. Type what you want to happen in ordinary language and Testsnag works out which element you mean and adds the right step. It is at its most useful on the interactions that are awkward to record by hand: a custom dropdown that closes when you look away, a date picker several clicks deep, a menu that only appears on hover.

1234

1. 1Use AI: describe the interaction in plain English and let Testsnag find the element.
2. 2Use Browser is Human in the Loop: you drive, Testsnag records.
3. 3Add Step authors one by hand when you already know the selector.
4. 4Steps the AI generates are ordinary steps. Review them before saving.

Writing a good instruction

Be concrete about the target and the intent. "Click the Continue button at the bottom of the address form" works better than "continue", because the page probably contains several things that could plausibly be a continue action. If a command produces the wrong step, the fix is usually to name the element more specifically rather than to rephrase the whole sentence.

On mobile

The chat panel is the main way to author an iOS or Android test. It works the same way and does the same thing, reading the screen, resolving what you described, and adding a step, with one difference worth stating plainly: the action is carried out on the device in front of you as the step is created, so you watch it happen rather than finding out on the next replay. Tapping the device directly in its floating window still records as normal whenever showing is faster than describing.

Human in the Loop

The Use AI and Use Browser switch in the recorder header decides who is driving the live browser during authoring. Choose Use Browser and you control the session directly with your own cursor and keyboard while Testsnag records what you do. Choose Use AI and AI browser control takes over again. You can flip between them as often as you like during a session. It is not a pause step that happens during later runs: it only affects the recording session in front of you.

Reach for it when a flow is genuinely easier to show than to describe, when you need to get past something fiddly and want precise control, or when the AI keeps landing on the wrong element and demonstrating once is faster than arguing.