ChatGPT's Chrome extension can read your open tabs and, in Work, act on them

The demo OpenAI posted starts with someone who has three sofa tabs open ^[11]. Engadget spells out what that person wants compared: dimensions, prices, delivery time and return policy ^[12]. Chat does that reading.

Work goes further. Engadget's examples are filling out a new dentist's form and adding the appointment to a calendar ^[13], and weighing several contractors' terms before drafting an email to the best fit ^[14].

The Verge quotes the sidebar's pitch as "the full power of ChatGPT" ^[7]. The example prompts it printed are narrower. One asks for a comparison sheet of vendors "with pricing, key features, and source links" ^[16]. Another reads: "Use the details in this event brief to fill out the registration form, and stop before submitting." ^[15]

Neither report includes usage figures or describes admin controls over which sites the extension can read. These prompts describe a narrow job. The assistant compares what is on the screen and drafts the result, and in the one prompt that mentions submitting, the person keeps the click. We'd treat that as OpenAI's own sketch of intended use. Without usage figures, it can't show what people do.

For Monday, we'd sort each task on two axes: whether the assistant reads or acts, and whether the result stays on the user's screen or lands with someone else. Chat's tab comparisons are read-only and stay on screen ^[5], so they go out first. Grouping tabs and filling a form without submitting are actions that still stay on screen ^[6]. Sending a message is an action that ends in someone else's inbox ^[3], and that cell is where a stop-before-submitting rule belongs. The tradeoff is time: a form the person must still review and submit saves less than one filled and sent.