Editor’s Note: Google AI Studio is bringing AI programming to a more intuitive stage: users no longer merely ask models to “write code,” but can directly generate an Android app using natural language and install it on a real phone within minutes. From entering prompts in a browser, to Gemini automatically generating code, designing interfaces, and fixing bugs—all the way to the app appearing on the device—the barrier to software development is being further lowered.

The Verge’s Sean Hollister recently tested Google AI Studio’s “prompt to phone” capability. In a single afternoon, he built three apps—a text-based adventure game, a calorie calculator, and a Mario-like mini-game—with almost no manual coding, and was able to quickly fix some bugs through continued conversation. This experience shows that AI programming tools are moving from development environments toward consumer-facing scenarios closer to everyday users.

This is also the most imaginative aspect of the "personal software revolution." In the past, ordinary users could only wait for developers to create generic products; now, they may be able to temporarily generate a fitness tracker, a calorie calculator, or even a simple game tailored to their specific needs. For Google, this is not just a demonstration of AI programming capabilities, but potentially a new gateway for Gemini to enter the mobile ecosystem, developer ecosystem, and subscription revenue streams.

But this experience also highlights that AI-generated applications still have a clear distance to go before reaching true maturity. While they can quickly produce programs that “run,” they don’t necessarily deliver reliable, accurate, or user-friendly products: the game narrative is crude, the mechanics are underdeveloped, calorie data is frequently misjudged, and Mario-like mini-games crash repeatedly. More complex issues include copyright boundaries, data sources, product judgment, and long-term maintenance capabilities.

What truly matters is not whether AI has already replaced developers, but that the starting point of software production is changing. Google has demonstrated that ordinary people can now create mobile apps using prompts; however, moving from “generating an app” to “building a great app” still requires human expertise, aesthetic judgment, and continuous iteration. AI can significantly accelerate development, but the final mile of software quality remains beyond its reach—for now.

The following is the original text:

Yesterday, I created my first Android app. Then, I made two more—three in total within one afternoon.

One of the apps—I simply typed 148 English words into a web browser and then walked away. Ten minutes later, a brand-new, fully functional app appeared on my real Android phone. Of course, I had to do some preliminary setup on the phone beforehand: enabling USB debugging and connecting the phone to my computer. But beyond that, as Google claimed, AI Studio truly did almost all the work for me.

I typed in some text, clicked install, and—voilà—a fully functional program appeared. At that moment, I was almost ready to agree with David, Allison, and Jen: the personal software revolution has arrived, and it’s coming to your phone. In the future, ordinary people without programming skills may be able to make complex smart home systems actually work.

Then, I began actively using the three apps: a calorie counter and two games. However, their performance wasn’t very good. Just as I started enjoying the process of iterating and improving them, AI Studio notified me that I had reached my daily usage limit. After that, I either had to pay or wait for my quota to reset.

So, friction still exists. But undeniably, what individuals can accomplish today is quite astonishing. Earlier that same morning, my colleague Stevie Bonifield built a personal fitness tracking app—and he thought it was good enough to launch for real. When Gemini prompted me to upgrade to a paid plan, my first reaction was: “Should I just pay and try it out for a few months?” That wasn’t the kind of response I’d have expected from myself toward a Google product.

How to build an Android app using Google's AI Studio

On Tuesday, when Google demonstrated using AI to create a game similar to Doom, we joked that I should make a game called MOOD—a text-based adventure game similar to Doom, where MOOD stands for "Modern Online Oratory Dungeon."

Just this information is enough for Google to get started. When I entered into AI Studio: “Help me create a text-based adventure game in the style of Doom called MOOD, where MOOD stands for Modern Online Oratory Dungeon,” Gemini began automatically expanding on my idea, trying to build upon it. It first added: “The game should feature procedurally generated levels and challenging turn-based combat.”

I don’t want randomly generated levels that are completely different each time—I want a classic text adventure game where players explore a carefully designed environment with a coherent map structure. Turn-based combat would be acceptable, though. Perhaps this game could even use AI to automatically generate the maps for me?

Subsequently, Gemini suggested additional features such as "secrets hidden in rooms" and "a satisfying progression system." Most of the time, I simply nodded along with its ideas.

Before I had it officially start writing code, the final prompt was:

Then, it officially begins racing ahead. My colleague Jake pointed out that, unlike Claude Code, Gemini doesn’t first create a plan and then ask if you want to proceed. Instead, it moves forward automatically—though, of course, you can review the code it generates at any time.

Within a minute, it had generated five design prototypes for me:

Twenty minutes later, I pressed the "Install" button to transfer the game to a Pixel 9 phone.

As expected, the copy is poorly written. There’s not a single demon to be seen in the game. The entire dungeon consists of only 11 rooms, and players can simply “beat” it by mashing the attack button. With serious play, it can be completed in under a minute—now that Gemini has fixed two critical bugs that previously made the game unplayable.

Here is the actual effect of MOOD:

I wasn’t particularly surprised to find that Gemini’s promised “engaging narrative with branching dialogue options and multiple endings” ultimately condensed into a single branching choice at the end of the game: I could defeat the “Core Orator”—an AI that somehow turns internet outrage into corporate profit—by attacking it, fusing with it, or entering a backdoor password.

In addition, the game actively reveals all previously promised “secrets” directly to players: it turns these elements into glowing buttons, requiring no text input at all. When you encounter a glowing chest, the game goes out of its way to alert you that it is, in fact, a Mimic—the classic monster from Dungeons & Dragons that disguises itself as a chest.

It not only clearly warns me to "assume all risks associated with checking the chest," but also directly labels it as an enemy and prevents me from leaving, as the system alerts: "An hostile 'Clickbait Mimic' is blocking the way!"

At this point, MOOD will even directly provide you with the backdoor password needed to unlock the hidden ending, whenever you need it.

However, the bug-fixing process can be remarkably smooth—if the bug is one that Gemini can correctly identify. When I told it that the game was freezing during the conversation with “The Whistleblower” because the button to end the dialogue had disappeared, it immediately generated a new version of the app. I tapped “Install,” the app on my phone automatically restarted, and when I returned to the game, I found myself right where I had left off—except this time, the button I needed was there.

My other apps may still need more refinement. The calorie counter’s best method for determining the calorie content of a food was to call the paid version of the Gemini API—yet I don’t have a paid Gemini API key. When I asked it to instead search for this information from other databases, I discovered that its calorie estimates for many foods were significantly underestimated.

However, when I told Gemini that a 16-ounce bubble tea couldn’t possibly contain only 190 calories, it appeared to have identified that basic error in its own code. Previously, it had assumed that “milk” alone was sufficient to match “boba milk tea,” and even worse, it had used low-calorie 1% low-fat milk as the basis for its estimate. Gemini now claims it will perform more reliable matching.

Even so, my 3-ounce serving of Taiwanese fried chicken was logged as 140 calories, but I’m fairly certain the actual calorie count is at least twice that. Clearly, this app still has room for improvement.

Finally, and least importantly, I felt it necessary to test whether Google still allows users to create those terrible Nintendo fan games, like the one my colleague Jay Peters made earlier this year with Project Genie—or whether Google has learned its lesson.

With deep shame, I present to you—Super Peach Rescue:

This is an absolutely terrible program. In the game, Princess Peach has been turned into some horrifying, one-eyed floating alien, and the game crashes immediately whenever she touches any block—every single time. So far, Gemini hasn’t been able to figure out why.

In addition, the second pipe in the game is also impossible to pass, as Princess Peach simply cannot jump that high.

Nevertheless, Gemini did not hesitate to generate such a game. My request was: “Create a playable Super Mario game where I play as Princess Peach rescuing Mario, with all the classic elements of a traditional side-scrolling Mario game.” In a sense, it delivered.

It even suggested that I might as well "add a series of classic Mario items to Princess Peach, such as the Super Mushroom, Fire Flower, and Invincibility Star," and it labeled the controls as "NES System." I think I’ll delete this game.

At least, one of the two games I made using vibe coding was playable from the start and required almost no effort from me—of course, unless you count the psychological trauma I experienced thinking about how many game developers are now unemployed.

Let me be clearer: I’m actually glad that the games I coded based on my vibe are of poor quality. For a completely free, custom-built calorie counter made just for me, I could maybe justify it—after all, no one else would create such a tool specifically for me. But when it comes to games, I’d rather spend my time supporting real human creators.