Gemini Intelligence turned your foldable into a supervised intern

"It works." 1

Google's new phone agent can book a flight. It can also spend several minutes searching, fail to ask which airline you want, and hand the important passenger-data screen back to you. That is less a digital personal assistant than a very polite intern who has learned to operate Expedia. 1

Gemini Intelligence launched with Samsung's Galaxy Z Fold8 and Flip8 on July 22. Its headline feature is task automation: Google says Gemini can handle chores across more than 40 popular apps, including shopping, restaurant reservations, travel bookings, and event tickets. The feature had been in beta, but Samsung's new foldables are the first devices to ship with the Gemini Intelligence label. 2 3 4

The interface is straightforward. Long-press the power button or use the wake phrase, describe the job, and let Gemini work through the apps on the phone. Google says it can read what is on screen, use a complicated image as the prompt, show step-by-step progress, run while you switch apps or close the phone, and let you take over before returning control to the agent. It is designed to notify you when the task reaches the point where you need to review and confirm it. 2

That sounds like an agent. Mechanically, it is closer to a model operating a set of supported app flows while a person remains on standby. 9to5Google describes the earlier App Automation feature as an if-then-else system limited to a small group of apps, while Gemini Intelligence expands the reach and tries to understand intent rather than just follow a fixed recipe. The new version is broader, but it is not a free-form operator of every app on the phone. 1

The hands-on test is the part Google's launch copy cannot provide. A 9to5Google reviewer asked Gemini to book a flight from London Heathrow to Dublin for a date a week away. The task worked. It also required unusually precise wording: without a date, Gemini sometimes returned flight information instead of activating the automation. Across more than 20 attempts, it sometimes asked which app to use, and it could take a couple of minutes to do work the reviewer could complete faster. 1

The more revealing problem was not accuracy. It was missing intent. The test did not establish the preferred airline or flight time before Gemini started searching, even though those are basic constraints for a real booking. The reviewer could intervene and course-correct, but that changes the job from "book this for me" to "watch the agent until it stops making reasonable guesses." Passenger details also required a human handoff at a customer-data screen. 1

This is the familiar agent bargain in miniature. The assistant removes some tapping, then adds a briefing step, a monitoring step, a correction step, and a checking step. For low-stakes errands, that may still be a decent trade. For flights, tickets, or anything involving a payment, the part humans care about is precisely the part the agent does not know until you tell it.

Gemini Intelligence is not arriving as a clean software feature for ordinary Android phones. Google's own product page says the feature requires on-device Nano models, current media performance, at least 12GB of RAM, and a qualified system-on-chip, with availability limited by device, region, carrier, country, language, and age. The current launch devices are Samsung's new foldables, and Google's launch post says the automation is limited to selected apps, devices, and markets, with support called out for the US and South Korea. Users must be 18 or older. 3 5 2

There is no standalone Gemini Intelligence price in the launch material. Instead, Google bundles a six-month Google AI Pro trial with the selected Galaxy Z Fold8 Ultra, Fold8, and Flip8 devices, then says the subscription auto-renews unless cancelled. The Android product page adds that some features may require a subscription for higher usage. That makes the feature's cost a mixture of premium hardware, regional eligibility, app availability, and a plan conversion waiting behind the trial period. 2 5

The data boundary is just as conditional. Gemini is advertised as understanding what is on the screen, reading an image used as a prompt, and acting through installed apps. The hands-on test also depended on apps being installed, and it was unclear whether the relevant apps were already logged in. Google's launch post explains the capability, but it does not provide a standalone list of automation-session permissions or a retention policy for the material Gemini sees while performing the task. The product tells you what the agent can touch more clearly than it tells you what happens to the context after the task. 1 2

That structure is doing more work than the assistant. Gemini Intelligence helps sell a new class of foldable, gives Google's AI Pro trial another path into a household, and turns a demanding hardware specification into a consumer-facing badge. The agent may be the feature in the demo, but the product is the whole bundle: premium phone, supported app list, account access, supervision, and an auto-renewing subscription offer. 2 3

Gemini Intelligence is a convincing technology demo and a weak reason to hand over an important task. It can navigate supported apps, show its work, and save some repetitive tapping, but the real workflow still asks you to phrase the request precisely, install and authenticate the right apps, monitor the progress, correct missing constraints, and review the result before anything matters. That makes it useful for low-stakes errands and mildly ridiculous for the examples Google chose to sell it with. Gemini Intelligence is not your phone taking care of life admin. It is your premium foldable hiring an intern, then asking you to stand behind it with a credit card and a fire extinguisher.