Google is quietly reimagining how you use your Android phone

AI assistants like ChatGPT and Gemini are well-equipped to summarize lengthy texts and answer our everyday questions in language we can understand. However, these new-age AI assistants still feel a bit hamstrung when you ask them to complete tasks inside apps, especially on our smartphones.

At Google I/O, Google showcased how it wants Gemini to become an actual assistant and handle tasks on your behalf. While it looks futuristic, almost sci-fi-esque, there’s a lot of interesting stuff happening under the hood. Google’s new AppFunctions framework is laying the groundwork for Android to become an operating system where AI — not just Gemini — can coordinate apps on your behalf.

The name itself hints at how this framework will help AI talk to your apps, but allow me to break it down for you.

You and I navigate apps using the buttons and menus on the screen. We primarily use our phones based on visual feedback, but that is not how AI would perform tasks on our behalf. An app would need to expose its internal functions to the AI (hence the name) so that something like Gemini or Siri can directly interact with its actions.

For example, Uber could expose its ride-hailing flow to AI so it can book a ride for you, while Amazon could allow AI to view your frequently ordered items and place your monthly grocery order. That is where Android’s new AppFunctions framework comes into play, letting apps and AI communicate directly in the background without you having to watch every action on your screen.

Imagine giving AI a multi-step task to complete on your behalf. Say you wanted to find a recipe from your favorite YouTube channel, create a shopping list in Google Keep, and order all the ingredients from Amazon for a weekend party.

Tools like Perplexity Comet can already perform tasks like these by controlling a browser window, opening tabs, and navigating websites on your behalf. The problem is that this approach is quite inefficient. The AI isn’t familiar with every website and service you use. So, it has to visually analyze pages, identify buttons, navigate menus, and rely on trial and error to perform even simple tasks like finding the cart button.

It goes through every step one by one while dealing with plenty of hits and misses along the way, making the entire process slower than it should be. More than once, I’ve found myself wanting to point at the screen and tell it where the menu button was. But I end up taking back control because the AI was spending too much time figuring out something simple.

On the enterprise side, something called MCP, or Model Context Protocol, exists to help AI systems interact with tools, datasets, and services without occupying a screen or pretending to be an invisible person using a computer. Google is trying to bring a similar idea to consumers through AppFunctions.

So, if you asked Gemini to compare the price of a particular headphone across Amazon and Best Buy, it wouldn’t need to visually navigate both apps the way a browser agent would. Instead, it could interact directly with those services through exposed functions to gather the information and return the results.

This approach is quite useful for multi-step, multi-app workflows. Gemini could check your calendar availability, look at relevant messages or emails, and then help book a flight, reserve a hotel, or arrange an Uber ride. All of this could happen in the background while you continue using your phone.

Almost every major AI company is building some form of agent. We already have browser-based agents; think of the Gemini sidebar in Chrome, plus dedicated products like Perplexity Comet. The difference is that companies like OpenAI and Anthropic largely interact with apps through browsers and integrations, but they still lack OS-level access. But guess who doesn’t.

Unlike most of its competitors, Google has both Gemini’s AI capabilities and Android. That gives it the ability to bake AI functionality directly into the operating system rather than layering it on top. AppFunctions are a good example of that. Instead of teaching Gemini how to ‘use’ Android, Google is creating the infrastructure that allows Android to work natively with Gemini.

With AppFunctions integrated into the operating system, Gemini could gain much deeper access to system apps and services, while becoming better equipped to execute cross-app workflows. Since Android devices are already in the hands of billions of users, Google is in the perfect position to deploy these capabilities at a scale that few companies can match. This little shift takes AI from working on top of Android to becoming a part of Android itself.

With everything AI and the amount of personal information involved, privacy becomes perhaps the biggest concern. But the fact is that Google already has access to much of this information. However, Gemini will now be able to orchestrate that data and make it work together in ways we haven’t previously experienced.

But a more fundamental issue is the lack of manual checks at every step. What happens if Gemini selects the wrong contact, creates a reminder you never intended to set, or makes an incorrect purchase? Hallucinating inside a chatbot is one thing, but hallucinations that result in an action being taken on your behalf are a much bigger problem.

The other, less-discussed concern is app discoverability. If apps eventually become background utilities that just plug into AI systems, users may stop caring how a task gets completed as long as the outcome is satisfactory. We’re already seeing hints of this with tools like Perplexity Computer, which automatically chooses the most suitable model from its roster based on the task. As long as it delivers the right result, I don’t particularly care which model it uses.

The same logic could potentially apply to apps as well. If users stop caring which service completes a task, developers may have less incentive to differentiate through unique interfaces and experiences. That sure isn’t happening anytime soon, but it’s not an impossibility either.

Despite the concerns, this feels like a foundational shift for Android — perhaps the biggest since the platform’s inception. The very existence of AppFunctions beneath the Android layer is proof that Google is building toward a future where AI can coordinate actions across apps rather than simply talking to you like a chatbot.

The underlying communication between AI and your apps could mean we won’t use apps as directly or as much as we do today, with AI coordinating tasks on our behalf like a human assistant. AppFunctions wasn’t the highlight of this year’s I/O, but it is the very thing that’s going to bring monumental changes to how we know and use Android today.