Firefox’s new AI tools can fill forms and watch pages. The browser must stay open

Mozilla is giving Firefox’s Smart Window two jobs that are easier to evaluate than an open-ended conversation: help complete a web form and watch selected pages for changes. The 27 October release describes an opt-in, English-language desktop beta in the United States and Canada. It is not a general announcement of availability on every device or in every market.[1]

The most consequential details are in the workflow boundaries. Form filling still ends with a person reviewing and submitting the information. Page watching depends on the browser remaining open. Neither capability should be understood as an independent worker that can be assigned a job and then forgotten.[2][3]

Filling a form is not the same as ordinary autofill

Smart Form Fill can use material from tabs selected by the user to populate supported text fields. Mozilla says those selected page contents are sent to Gemini 3.1 Flash Lite, irrespective of the model selected for chat. By contrast, the browser’s conventional saved-autofill information remains on the device. The two routes should not be treated as having identical data handling.[2]

The documented process leaves the generated entries available for review and editing before submission. It does not automatically submit the form, and the generated information is not currently saved as reusable autofill data. The feature is aimed at supported web text fields, not PDFs or every kind of form control.[2]

For a hypothetical event application, that could mean drawing a speaker biography from one chosen tab and a project description from another. The relevant review would then be concrete: did a past job become a current one, did a word limit cut off an important qualification, and did an answer address the question actually asked? These are suggested checks, not observations from our own product test.

A page watcher needs an operating schedule

The task interface accepts /watch requests and can monitor up to five pages per task, with up to 15 tasks. Checks can run daily or weekly, but Firefox must remain open. Mozilla says the monitoring uses Qwen regardless of the chat model. Tasks pause after 60 days without a match and stop after 90 days.[3]

Those limits make a page watcher different from a guaranteed alert service. A user waiting for a deadline-critical change needs to know when a check last ran, not simply that a task was created. Closing the browser is an operational interruption, even if the original instruction was perfectly clear.

A useful request would identify both the pages and the condition that matters: for example, whether an application page changes from closed to open. That is more discriminating than asking to be told about any change, which could include an unrelated edit. This is a proposed way to specify a task, not a claim about measured alert accuracy.

The model selector does not describe the whole product

Together, the features reveal a subtle interface issue. Someone can choose a chat model and reasonably assume that choice governs the surrounding AI functions. Mozilla’s documentation says otherwise for these two workflows. The appropriate privacy and capability question is therefore about the action being performed, not only the name displayed in the conversation.[2][3]

The appeal is modest but tangible: less copying between tabs and less repetitive checking. The release is most useful when read as a set of documented, bounded tools. Keeping the final submission with the user, and making the browser-open requirement explicit, gives people a more accurate picture of what they are delegating.

Report an error or suggest a correction