This year's Google China Developer Conference took place in Shanghai, where developers flooded the venue with excitement matching the city's sweltering weather (though it might have been due to insufficient air conditioning). From my perspective, the event had one central theme: believing that zero-code alone makes you a qualified developer is naive; however, hesitating, worrying, and being afraid to take the first step toward becoming a developer is truly regrettable.

Can you at least talk? For someone like me who can't write a single line of code, I could previously only pretend to be an engaged participant. This time was different—Alphabet set up an "App Hotline Booth" at the venue, connected to Gemini 2.5 Pro on the other end. You describe your ideas in plain language, Gemini asks some clarifying questions about requirements, and after mutual confirmation, you hang up. In less than a minute, Gemini returns a front-end interface.

I asked it to generate a "dog recipe" app that could dynamically create daily meal and nutrition supplement ratios based on a dog's age, starting from 5 months old. How would I evaluate what Gemini produced? It was like my absent-minded response of "got it" to my boss. Of course, this was just a small demo showcasing the Vibe Coding concept—enabling people with no programming experience to transform creative ideas into applications using natural language to guide AI through complex coding tasks.

Alphabet has introduced Agent Mode across many development environments, including Firebase Studio and Android Studio. Based on developers' natural language descriptions, these agents autonomously complete multi-step tasks like building prototypes, fixing bugs, adding features, and refactoring components—like having a copilot beside you. Even tedious frontend debugging details can be handled by AI. Developers can ask Gemini about CSS layout issues in DevTools. For example, Alphabet demonstrated at the event: "How do I center a button?" Gemini responded, "Add translateX(-50%)," and you can simply choose to apply the suggestion.

What excited many developers was the ability to run agents directly in browsers. Alphabet has optimized for common tasks like text summarization, email drafting, and contextual prompting, launching 7 on-device AI APIs based on models like Gemini Nano that are completely cloud-independent, allowing developers to call AI functions like ordinary function libraries.

Also in on-device AI, Alphabet released the open-source model Gemma 3, which shares the same architecture as Gemini Nano and requires only 2GB of memory to run. In the live demonstration, a technical evangelist from Alphabet showed running the Gemma 3 4B model offline using the community tool LM Studio. He dragged in an image of a plane ticket and gave the model a multi-step Chinese prompt to read the information in the image, save it as JSON format, and add two new fields: "Thank you for attending today's event" and its English translation. The Gemma 3 4B model successfully completed all requests, demonstrating local multimodal and structured data output capabilities.

**Becoming a Qualified Developer: Putting Your Ideas into the Ecosystem**

Alphabet showcased an interesting application called Androidify. Users upload a selfie, and AI processes it based on skin color and accessories (even if you're walking a dog) to generate an "Android character." Since I didn't bring a dog, I took a photo holding a picture of a small dog.

The technical logic behind the Androidify app involves calling the Gemini 2.5 Pro model through Firebase AI Logic to generate text instructions for user selfies, then passing them to Imagen to create personalized avatars. You can understand that behind a new product isn't just a single-point technology update, but a complete "ecosystem" update.

From Android, Web, and cross-platform to Cloud, Alphabet provides entirely new toolsets and integrates AI into commonly used IDEs to assist program development. On the Android front, for example, Androidify was developed using the new Compose Material library, plus the aforementioned Agent Mode introduced in the Android Studio development environment.

For Web development, improvements focus on UI quality and cross-platform compatibility. New CSS primitives have been introduced for UI construction, while improved Baseline integration into mainstream development tools like VS Code and ESLint covers all Web platform features, simplifying cross-browser development.

On the Cloud side, the underlying Gemini 2.5 series models are fully available on the Vertex AI platform, with new Supervised Fine-Tuning (SFT) and Vertex AI Model Optimizer features (simplifying model selection) to optimize model training, evaluation, deployment, and monitoring.

The Firebase Studio development environment toolchain has also been upgraded, including Gemini Code Assist, based on Gemini 2.5 Pro, further enhancing automatic completion, generation, and code explanation, with integrated GitHub/GitLab access, Google Docs access, and code generation based on database schemas. The Gemini CLI "command-line tool" also serves as a component of Agent Mode, specifically responsible for executing the specific steps planned by Agent Mode in command-line environments.

Overall, AI seamlessly integrates into multiple development scenarios including Web, Android, and Cloud, with deep integration into development tools. AI Coding has progressed beyond improving productivity and evolved from simple "code completion tools" to becoming a "copilot" that can execute tasks like accessing local development environments, analyzing project errors, managing dependencies, and interacting with applications.

**The Core Value of Programmers Remains**

"Regarding discussions about Vibe Coding making backend frameworks irrelevant, I disagree," said Timothy Jordan, Director of Developer Relations and Open Source Platforms at Alphabet. "You can't just stay at the application layer; you need to ensure code security, scalability, and implementation of specific functionality." This relates to demonstrating programmers' core value.

Jordan shared data showing that as of March this year, Gemma celebrated its first anniversary with over 200 million downloads, spawning more than 80,000 derivative models. The most advanced version to date, the Gemma 3 series, supports over 140 languages. For example, ShowGemma is an application for users to securely deploy Gemma models; MedGemma is an open-source multimodal model specifically designed for medical and health fields; and even DolphinGemma, a model helping scientists research "dolphin language."

"Alphabet's advantage as a platform is providing end-to-end solutions, achieving vertical and horizontal integration, coordinating resources across cloud, mobile, generative AI, and AI assistant aspects for efficient operation."

Every major model company is trying to build its own platform ecosystem, and AI narratives have evolved from single model technology levels to more complex ecosystem-level competition.