Google just held their annual IO event, revealing a bunch of new features and products aimed at developers.
While some announcements may seem flashy, many are about coding experiences and streamlining our workflows ...
Jules: Autonomous AI Coding Agent
Jules is an autonomous AI coding agent currently in public beta. It is designed to function in a secure virtual machine, providing:
- Project Understanding: It clones your GitHub repository to gain insight into your entire project.
- Task Management: Jules can write tests, build new features, fix bugs, and even provide audio change logs.
- Future Collaboration: Imagine having stand-up meetings with AI agents, potentially transforming the daily operations within software development teams.
Jules operates on the Gemini 2.5 Pro model.
Gemini Code Assist
For those who prefer a coding assistant rather than a fully autonomous agent, Google has rolled out Gemini Code Assist. This co-pilot extension is now generally available and integrates seamlessly with popular development environments like VS Code and JetBrains. This tool aids developers by offering suggestions and coding assistance, also powered by Gemini 2.5 Pro.
UI Development with Stitch
Google has also introduced Stitch, a novel experiment aimed at simplifying UI design and development. This tool allows users to express their ideas via text and image prompts.
Stitch can convert these inputs into complex UI designs, which can be iterated quickly. Users can also paste their designs back into design tools like Figma or export the front-end code.
Gemini Diffusion
Among the more intriguing updates was the introduction of Gemini Diffusion, an experimental research model that generates text and code via diffusion techniques.
This model claims to be really fast, even completing code generation before a user finishes reading the prompt.
A waiting list is currently in place for those interested in testing this.
Upgraded Gemini 2.5 Models
Gemini 2.5 Pro has received upgrades:
- Integration with Learn LM, developed alongside education experts, this model excels in educational applications.
- DeepThink, enables the model to explore multiple hypotheses before generating responses, demonstrating a boost in performance in math and code.
Gemini 2.5 Flash model is improved in reasoning, multimodality, and handling of long context scenarios.
API Features
Gemini API:
- Conversational Interactions: Enhanced natural language capabilities include audio output that can adapt tone and emotional nuance, allowing for more engaging conversations.
- Asynchronous Function Calling: This feature enables long-running functions to operate without disrupting the main conversational flow.
- URL Context: This allows users to retrieve the full context of web pages directly from URLs without needing additional search tools.
Tools in AI Studio
Google's AI Studio has been updated with new generative media models like Imagen and VO, along with the capability to include audio in generated videos.
Also, integration with Gemini 2.5 Pro allows developers to create web applications directly within the AI Studio's code editor.
Mobile-First AI: Gemma 3N
One of the more surprising announcements was the preview of Gemma 3N, a mobile-first AI model optimized for on-device tasks.
This model promises faster responses and improved understanding across various media types. It operates effectively with a memory footprint of 2GB, enabling mobile devices.
Advancements for Android Developers
For Android developers, new generative AI tools, including MLKit genai APIs powered by Gemini Nano, offer on-device capabilities.
Material 3 Expressive allows for adaptive application development across a variety of devices, including phones, tablets, and even cars.