Google has unveiled Android Studio Panda 2, a significant update that introduces an AI agent capable of generating applications from the ground up, alongside an AI-enhanced version upgrade assistant. This latest iteration of Android Studio is built on the community edition of JetBrains IntelliJ IDEA, providing a robust foundation for developers.
AI Integration and Features
The AI capabilities within Android Studio are primarily powered by Gemini, Google’s family of large language models (LLMs). While the default setting utilizes Gemini, developers have the flexibility to configure alternative AI providers through the settings. The introduction of Gemini in Android Studio comes with a free tier, offering access to a “lightweight version of Gemini 2.5 Pro,” which is designed to handle most tasks effectively. For professional developers, a subscription to the business tier is recommended to unlock advanced features.
According to Senior Product Manager Matt Dyor, the new AI agent allows developers to create a working prototype with just a single prompt. This agent meticulously constructs a detailed project plan and automates the workflow, which includes generating code, building the project, analyzing errors, and self-correcting until a complete implementation is achieved.
Getting Started with Gemini
To begin using these features, developers must sign into Gemini and enable AI integration within Android Studio. A privacy dialog informs users that Google will collect data such as “chat text, prompts, related code, generated output,” and that this information may be reviewed by human analysts. Developers are cautioned against including any confidential information in their prompts. A practical approach to safeguard sensitive data involves listing it in a .aiexclude file, ensuring that it remains outside the AI’s processing scope.
Testing the New Features
In a hands-on demonstration, we challenged Gemini to create a bridge deal analyzer, designed to accept a bridge hand in Portable Bridge Notation (PBN) format and conduct a double dummy analysis. The AI quickly generated a promising project plan, outlining features such as “effortlessly importing bridge deals” and performing “instant calculations of the maximum tricks available for each strain.” Its grasp of bridge terminology was notably impressive.
Upon approval to proceed, the agent spent approximately 15 minutes iterating through its development loop, seeking consent for each file modification. Ultimately, it announced that “the application has been thoroughly tested and verified by the critic_agent… it meets all functional requirements.”
However, while the bridge deal analyzer successfully built and ran in the Android emulator, a closer examination of the generated test deal revealed discrepancies, including one hand containing 14 cards instead of the standard 13, and the 5 of clubs appearing twice. Attempts to rectify this issue resulted in an “invalid rank” error. Further inspection of the code indicated that the double dummy analysis was based on a “simplified heuristic,” leading to inaccuracies, although it functioned with invalid input.
Given that this was our initial experience with the free tier, it is essential to approach Gemini’s performance with a degree of understanding. The iterative nature of vibe coding encourages developers to experiment and refine their prompts. More precise instructions, such as utilizing established open-source double dummy code available on GitHub, could yield improved outcomes.
Additional Updates in Android Studio
Aside from the AI advancements, Android Studio remains a mature product, albeit with some performance hiccups. During basic usage, we encountered errors, including a Gradle plugin error: “java.lang.IllegalStateException: This method is forbidden on EDT because it does not pump the event queue,” along with frequent notifications that “sync is taking a significant amount of time to download dependencies.”
Another noteworthy change is the deprecation of the Custom View preview, which previously allowed for the creation of custom controls without Jetpack Compose. The latter is now the recommended approach, reflecting a shift in development practices. Additionally, the 3D mode in the layout inspector has also been deprecated, as usage metrics suggest a decline in its popularity.
