An information agent is a loop rather than a query. Criteria get stored once, something fetches candidate material on a cadence, a comparison step decides whether those criteria are now met, and an update or an action fires. Google's description names the ends of that loop and leaves the middle unspecified. Blogs, news and social content are the monitored surfaces [3], and the account does not say whether the fetch hits Google's index or your origin [10]. That distinction decides whether your published facts get read on a crawl cadence you can observe in logs, or on a schedule set per user.
If the fetch is live, anything that only exists after client-side render is invisible to the comparison step. So is availability that exists only inside a booking widget, and a price that exists only after a form submission. None of that is new advice for crawlers, but now the reader arrives with a task, a set of constraints and a next step, so ambiguity costs you the comparison rather than a ranking place [13].
Then the loop reaches the action step. Google described booking local experiences and services, including calls placed to businesses on a user's behalf [4], and the dev.to preparation list puts "review how enquiry, booking and customer-service processes handle AI-assisted interactions" next to the usual data hygiene [9]. At that point the interface stops being HTML. It is your phone line. The front desk becomes an endpoint with no schema and no rate limit.
The evidence itself comes with a caveat worth flagging. This is one write-up of Google's announcement, and it is dev.to that characterises the plan as a confirmed product direction and rollout plan rather than a prediction [1]. And no ranking formula accompanies the announcement [8], so the only work you can test now is whether a reader with no context can extract what you offer, where you operate, what conditions apply and how to proceed [9]. That test you can run today, with a colleague and a stopwatch.
The schedule is tighter than "summer 2026" sounds. The subscriber launch and the wider United States availability sit in the same season [5], which leaves at most about three months between the paid cohort and a much larger one [11]. Google also says it intends to expand agentic capabilities so actions can be completed through providers, with shopping on that path [12], which is where a comparison becomes a transaction rather than a report. The same announcement also names agentic coding features that dynamically create custom interfaces and mini-apps [6]. That means the surface a user finally acts on could be one Google assembles on the fly, separate from anything you served.
