Google’s Gemini can now handle some multi-step jobs inside Android apps, which sounds like the sort of AI promise that usually collapses the second you try it in real life. This time, the more interesting answer is that it does seem to work — just not in a polished, magical, everyday way yet.
According to a hands-on report from The Verge, Gemini can now book rides and build food orders inside a small set of Android apps by tapping and scrolling through them on your behalf. Google’s own support page says the beta feature can also help with grocery orders, and that it runs inside an on-device virtual phone environment rather than taking over your whole handset.
That is the important shift. For years, voice assistants have mostly been good at simple commands: set a timer, play music, tell me the weather. This is closer to asking your phone to do a task rather than just answer a question. Earlier this month, we looked at how phone AI is moving beyond clever writing tricks. Gemini’s new automation feature is one of the clearest examples so far.
What Gemini can actually do right now
In its current form, Gemini’s app automation is narrow. Google says it works only on certain phones, including the Pixel 10 range and Samsung’s Galaxy S26 series, and only for adults using a personal Google account in the US or Korea. It is also limited to select apps and English for now.
The supported tasks are ordinary: booking a ride, reordering coffee, putting together a takeaway order, or adding groceries to a basket. Instead of needing exactly the right command, you can use natural language. In Google’s examples, you can say things like “Book a ride to the airport” or “Order pizza for delivery”, and Gemini works through the app steps for you.
That matters because normal people do not speak to technology in neat menu labels. They describe what they want in messy everyday language. If AI assistants are ever going to feel genuinely useful, that is the hurdle they have to clear.
Why this is impressive even though it is still awkward
The Verge’s testing makes the trade-off clear. Gemini is slower than a human using the app directly, and sometimes noticeably so. One dinner order reportedly took around nine minutes while the assistant made a few wrong turns and hunted around the menu. That is hardly a great advert for saving time.
But speed is not really the point yet. The bigger deal is that the assistant can reason its way through a task that was designed for human eyes and fingers. It can spot that two half portions equal one full serving, pause for confirmation before the final step, and keep working in the background while you do something else.
That is why this feels more significant than another round of AI summaries or chatbot polish. It is an early glimpse of what phone assistants may become once they are less brittle, faster, and connected to more apps.
There is also a useful mental model here. As we argued recently, AI tends to work best as a helper rather than a substitute. Gemini’s automation makes most sense when you treat it that way: handy for low-stakes admin, but not something to trust blindly with anything important.
Why most UK readers should not get carried away
If you are reading headlines about this in the UK, the first practical point is simple: this is not a normal feature on normal Android phones here. Google says it is currently limited by device, account type, app support, language and region. So even if the idea sounds appealing, most people in Britain cannot just pick up their current handset tonight and start ordering dinner through Gemini.
The second point is that Google itself is being fairly cautious. Its help page says the feature is still learning, can make mistakes, and should not be used for emergencies or tasks involving sensitive information. Users are told not to type payment details, passwords or other private information into the Gemini chat, and to take control for steps like signing in or finalising transactions.
That caution is sensible rather than disappointing. If an AI assistant gets a side dish wrong, that is mildly annoying. If it books the wrong pickup point, sends a message you did not mean to send, or misreads something more sensitive, the problem grows quickly. Google also warns about prompt injection, where hidden instructions inside websites or app content could try to mislead an AI agent into doing something unintended.
So what should ordinary people take from this?
The calm answer is that this is worth watching, but not worshipping. Gemini’s new automation feature looks like a real step forward for AI on phones because it is doing an actual job inside real apps, not just producing tidy demo text. At the same time, it is still a beta feature with sharp limits and a lot of supervision built in.
For UK users, the practical takeaway is less “I need this now” and more “this is where phone AI is heading”. Over the next year or two, the useful versions of these tools will probably be the ones that quietly remove friction from boring digital chores: booking, reordering, filling forms, comparing options, and teeing things up for a human final check.
If that future arrives carefully, it could make smartphones feel less fiddly and less full of repetitive taps. If it arrives carelessly, it could just add a new layer of confusion between you and a task you could have done yourself in half the time. Right now, Gemini’s automation is both a promising preview and a reminder that “can do it” is not the same as “should do it unsupervised”.
Sources:
The Verge — Gemini task automation is slow, clunky, and super impressive
Google Gemini Apps Help — Ask Gemini to handle your multi-step tasks in select Android apps
TechCrunch — Gemini can now automate some multi-step tasks on Android
