I originally thought that the Samsung Galaxy S26 series had already been exposed, and the press conference would just go by the wayside. Unexpectedly, Samsung and Google are hiding something. The two companies jointly demonstrated the new Gemini intelligent capabilities of the S26: with a verbal command, Gemini can help you hail a taxi from Uber or order takeout from DoorDash.
Source: Android Central
This feature is currently in early preview and is only available in the United States and South Korea.
You can understand that Google and Samsung have joined forces to create a global version of "Doubao Mobile" (called Doubao Mobile Assistant to be precise).
The Galaxy S26 series is just the beginning, and these capabilities will be pushed to Google Pixel 10 phones and more Android 17 devices later.
After seeing and using many mobile phone/computer system-level AI agents, and also using the "Beanbag Phone" in depth, looking at the Gemini agent this time, I feel that the discussion about it should not stop at a "new function".
Admittedly, this is not the first time that the underlying framework of the Android operating system has been deeply customized to accommodate smartphones - many manufacturers, including OPPO, Honor, Huawei, etc., have already made quite a few early attempts.
But this is Google, the absolute owner of the Android operating system.
If ByteDance, as an "outsider", makes an attempt that is "disrespectful" to national-level apps - if Google comes to do this, the meaning will be completely different.
But don’t worry, let’s take a look at what’s going on with the “bean bag phone” made by Google and Samsung this time.
Samsung’s “Bean Bag Phone”, how is it used?
The "Gemini Automated Task" capability demonstrated by Samsung and Google this time can imitate humans operating mobile phones to automate tasks. The implementation idea behind it is the dual path of AI screen reading + system bottom layer/application layer API.
It should be noted that the "Doubao Phone" jointly developed by Byte and Nubia heavily uses the ability of system-level permissions and screen reading, not APIs. You can understand that Doubao Mobile’s main approach is to “fail to greet application developers” (at least the mainstream national-level apps do not). The “forced” implementation idea also leaves a handle for national-level apps to block and resist it.
The Gemini smartphone developed by Samsung and Google this time on the Galaxy S26 series can be said to have both. According to information disclosed by Samsung, the top 200 apps in its app store can all support it (but only the effects of specific apps can be guaranteed, as detailed later) - indicating that Samsung and Google have at least generally said hello to these app developers.
Let's take a look at the experience of "Wired" magazine: call Gemini directly and tell it that you are going to the airport. The Gemini application itself will open a "virtual window" to open Uber and start executing this action in the background. The user can click to enter at any time to view the execution process of Gemini.
Since there are several different airports in the local area, Gemini quickly reminds users to choose the appropriate destination; when placing an order, Gemini will also push the interface to the user to facilitate the user to select the appropriate vehicle and pay.
Gemini's "virtual window" can be understood as a sandboxed "virtual machine", which is Google's consideration for user privacy protection.
In the past, Gemini ran on the Android system, but this time when the new Gemini agent operates applications, it only works within this sandbox and does not touch other parts of the device.
One more mention: If you have used Manus, Kimi computer, AutoGLM, etc., intelligent products with cloud computer/cloud phone capabilities, you should easily understand the logic of this Gemini virtual machine.
Image source: 9To5Google
This is a fairly simple task, and many domestic AI mobile assistants have already overcome this scenario a year ago.
The more killer capabilities of Gemini are combined with the screen reading and information grabbing features that have been laid out for a long time.
For example, when a user talks to a friend about ordering pizza for a party, the user can directly call Gemini and say "clarify the order", and Gemini can directly capture the pizza shop mentioned in the chat, and even the specific pizza type, and sort out everyone's needs.
Afterwards, users can directly ask Gemini to order takeout on the takeout platform Grubhub, and AI will automatically add all the food to the shopping cart in the background according to the order requirements that have just been sorted out, and deliver it to the user for confirmation and order placement.
Sometimes, the food ordering situation does not go so smoothly, and Gemini will try to solve the unexpected situation first and provide users with solutions. Once, when the pizzeria limited orders for large pizzas during busy hours, Gemini would ask if she could order two mediums instead.
Another example: a Google Keep note listing the attendance list for a barbecue party and noting vegetarians. Gemini can first calculate how many hot dogs and buns are needed for the entire party, and then ask it to purchase ingredients. In a few minutes, all the items will be placed in the shopping cart on the DoorDash platform.
Sammer Samat, president of Google's Android ecosystem, revealed that Gemini does not "remember" the steps and routes for operating these platforms in advance, but is really using reasoning capabilities to imitate humans to view the screen and perform the next operation, which means Gemini can exert its potential in more scenarios in the future.
Here you can see that Gemini’s first batch focuses on food ordering and taxi-hailing scenarios, which is more like what Qianwen did before the Spring Festival.
Source: Wired
Another “beanbag phone” from Android official
Compared with the truly "all-round" Beanbao mobile assistant that can even help you find WeChat collections (at least before it was boycotted), Gemini's current capabilities are still quite limited, focusing on daily scenarios such as taxi hailing, takeout, and groceries. Although the underlying technical capabilities are stronger, the user's actual use effect is not much different from domestic mobile phone AI assistants such as Hongmeng's Xiaoyi and Honor's YOYO.
However, as mentioned at the beginning of the article, Google holds the entire Android ecosystem and has absolute appeal and control.
With the release of Gemini's automation capabilities, Google has also disclosed in detail the underlying layout and future plans of the Android system behind it - there are two directions. To put it simply,It is both "apple" and "bean bag".
First, Google released a framework called "AppFunctions" last year, which allows developers to expose application-specific functions and feature entrances for AI assistants to call.
Google compares AppFunctions to Android's "Model Context Protocol" (MCP), which can be simply understood as a conversation standard to help third-party App applications and AI models connect.
This framework is similar to Apple’s App Intents. In Apple's concept, users can use Siri to operate various apps to implement functions, and the underlying implementation method is through App Intents. Under the premise that the new generation of Siri has not yet been implemented, App Intents are enough to provide good results.
The same goes for Google's AppFunctions.
For example, a user may issue an instruction to find a recipe from a friend’s email and add relevant ingredients to their shopping list. After receiving the command, the AI first calls the "Search" function entry of the Mail App to retrieve and extract relevant content, and then calls the "Shopping List" entry of the memo to fill in and organize the data.
Some AppFunction functions have been implemented in Samsung Galaxy S26 and One UI 8.5 systems. For example, users can command Gemini to find specific photos in their albums and send them to friends via text message.
It should be noted that during the entire process, Gemini does not need to open the photo album and SMS App, or even leave the Gemini App. Instead, it uses AppFunctions to capture the corresponding entry into Gemini to perform operations, which is more efficient.
In essence, the implementation based on AppFunctions is the same as the past API path logic. This is a kind of "say hello" problem-solving idea.
However, not all apps have made relevant adaptations. It doesn't matter, Google has made another preparation.
In an article posted on the Android Developer Blog yesterday, Google made it clear that the company is also developing a UI automation framework that allows AI assistants and third-party applications to imitate humans and directly open the app and perform step-by-step operations.
——This is a replica of the "bean bag mobile phone".
However, although Google said that UI automation will take on the real "heavy work" in the future, in this Galaxy 26 series, UI automation is only an "early preview version."
▲ Doubao mobile phone helps me grow grass and compare shampoo prices
If AppFunctions requires App developers to perform additional adaptation work, then the UI automation framework leaves all the work to the AI agent without any additional adaptation. However, the effect depends very much on the capabilities of the AI agent. The advantage is that it can cover a large number of applications as soon as it is launched online.
Now you can see that in Google's Android Gemini agent plan, AppFunctions and UI automation are two routes that complement each other: ensuring maximum compatibility through standardized and traceable interface methods, while laying the foundation for a screen-reading interaction model that truly represents the future.
Google also said that this will not be an exclusive feature of Gemini, but a feature of the Android system.
This also means that in the future, whether it is the mobile phone manufacturer's own built-in AI assistant or third-party applications such as ChatGPT, they can call AppFunctions to perform tasks or "read" the mobile phone UI for automatic operations.
It is worth mentioning that even when the National Bank cannot use Gemini, the Bixby assistant of Samsung Galaxy S26 can also realize the functions of ordering takeaways, hailing taxis, and e-commerce price comparisons.
We can reasonably infer that Samsung has also found a model supplier in China to replace Gemini. As for who among these large model dragons, it may depend on who has achieved more outstanding results in mobile phone smartphones in the past year.
The road to AI mobile phones will not be limited to “lonely warriors”
Last year, the "Bean Bag Mobile Phone" made a stunning debut, but died prematurely due to regrettable circumstances. While deeply regretful, it also makes us think about, is the AI automation model the ideal model for AI mobile phones?
This question will not be answered for three to five years. At least Doubao mobile phone is not alone. Google, which owns the Android system, has also chosen this route and has a much greater say.
In fact, when Doubao mobile phones became popular overseas, some netizens began to imagine that if Google promoted this technology on Pixel and Android phones, the prospects would be very broad.
Although I think Google does not have a very clear answer to the proposition of "AI mobile phone". It is more like having AI, system and hardware at the same time. If you try it in every direction, maybe there will be a way through.
But at least, Google has laid a good example of "system-level automation" for Android, and many new phones will have the potential to become "bean bag phones".
This wave may not stop at the Android camp. Don’t forget, Apple has reached a cooperation with Google, and Gemini will become the technical support for Siri. And App Intents and AppFunctions are very similar...
Demo of AI Siri
Looking a little further: Gemini agents aren’t even limited to AI phones. In Sammer Samat’s vision, in the future smart glasses, AI pendants, and even cars, as long as they have Gemini, they can use it to complete complex tasks—of course, such a scenario is still far from being implemented.
However, Google has only taken the AI automation route at the technical level, and the establishment of the paradigm does not mean that the problem disappears. The various contradictions Doubao encountered at the time will also become challenges that latecomers will have to face.
The first, of course, is privacy and security. Google's pie is very big. In the future, it will not be limited to Gemini to call and operate mobile apps. Some third-party AI applications can penetrate deeper into the user's data core. If there are disguised malicious applications that take advantage of these interfaces, it will also cause greater losses.
Image source: 9To5Google
The more intense conflict is the competition among mobile phone hardware manufacturers, model/agent capability providers, and large platform applications over the new "entrance" to the AI era. This was also the original bean bag mobile phone, which was once the most difficult wall to overcome.
After all, using Gemini to hail a ride may mean that users no longer have to see Uber’s membership promotions, advertising recommendations, or even form brand stickiness, directly damaging the revenue of application service providers/advertising industry.
China has Internet/AI giants, so why not overseas? Old rivals like Meta and Amazon still have strong platforms and ecosystems, but they may not be willing to open up to Google and let Gemini automate everything.
Whether it is based on privacy, security, or platform rules, setting restrictions and raising access thresholds will inevitably lead to gaming and the battle will become more intense.
At least Google is confident about the future.Sammer Samat believes that AI technology has entered the "ongoing phase", and developers, rather than racking their brains to fight against it, should think of a suitable way to embrace it.
The confrontation between the new and the old is inevitable, and even national-level applications with a large number of users will not be immune forever. The final winners are likely to be those players who have been bravely pursuing the change on the eve of the change.
References:
https://android-developers.googleblog.com/2026/02/the-intelligent-os-making-ai-agents.html