Google crowns Jules to be its agent and spreads the AI love
Choc Fac brings gen mods to Android, Chrome, pretty much everywhere else
Google I/O Google technical folk laid out a menu of geeky delights on Tuesday at Google I/O, in the hope that software developers will pay to build upon the Chocolate Factory's platforms and services.
The developer-focused portion of what's notionally a developer conference followed the general interest keynote from Google CEO Sundar Pichai. After Pichai rendered the super-corp's accomplishments in the broad strokes of salesmanship, his stable of software and hardware experts took turns addressing the masses assembled at the Shoreline Amphitheatre in Mountain View, the chrome heart of Silicon Valley, to illustrate the kinds of applications that Google's tools and infrastructure now make possible.
Google Labs VP Josh Woodward kicked off the egghead parade by demonstrating Stitch, an experimental AI service for creating user interface designs for web and mobile apps.
"This is Stitch," he said. "It starts with design first. So you can go into it and just paste in a prompt, like 'Make me an app for discovering California, the activities, the getaways.' And basically, you can just click Generate Design. And it's going to go out and start making a design for you."
The results can be exported as CSS/HTML or into Figma for further refinement.
Stitch offers a choice of Google's Gemini 2.5 Pro and Gemini 2.5 Flash AI models, celebrated during the main keynote for their capabilities. These include support for building AI applications that hear and speak in 24 languages in real-time.
Google's asynchronous coding agent Jules has entered into public beta. Like the new GitHub Copilot coding agent, Jules can craft code, fix bugs, and perform tests on GitHub repos on its own Git branch. No human oversight is required, although it's up to the human minder – the developer, if that's still the right term – to create a pull request to merge the AI-authored changes.
Google's Gemini 2.5 Pro model has been integrated into Android Studio. This brings with it two agentic capabilities: Journeys, which enables Gemini to perform app testing; and Version Upgrade Agent, which automatically updates dependencies to the latest compatible versions, and builds the project to catch and resolve errors that may have occurred from the dependency changes.
Android devs also gained access to ML Kit GenAI APIs that allow Gemini Nano to be used for on-device tasks.
Web developers felt Google's love in the form of new CSS primitives for building Carousels – scrolling content areas – in only a few lines of CSS and HTML.
Pinterest tested the web tech, according to Google, and saw a 90 percent reduction in Carousel code, from about 2,000 lines of code to 200.
"It should not only be possible but simple to create beautiful, accessible, declarative, cross-browser UI," said Una Kravets, staff developer relations engineer for Google Chrome.
Along similar lines, there's now an Interest Invoker API, available as an origin trial, for toggling the presence of popover menus when visitors show active interest in a specific portion of a website. A Google demo of the API that incorporates the Anchor Positioning API and Popover API showed how it might be used to surface the price of a theater seat when a website visitor hovers over a graphic showing the auditorium.
- GitHub Copilot angles for promotion from assistant to agent
- LLMs can't stop making up software dependencies and sabotaging everything
- AI can improve on code it writes, but you have to know how to ask
- Bake an LLM with custom prompts into your app? Sure! Here's how to get started
- Techies propose the Agent Name Service: It's like DNS but for AI agents
Meanwhile, Chrome DevTools now has Gemini built-in. AI assistance is available to accelerate styling, performance optimization, and debugging.
With the arrival of Chrome 138 on Monday, web developers participating in Google's early preview program can try out built-in (client-side) AI capabilities enabled by Gemini Nano.
The Summarizer API, Language Detector API, Translator API, and Prompt API for Chrome Extensions are available in Chrome 138 Stable. The Writer and Rewriter APIs can be accessed in origin trials, and the Proofreader API and Prompt API with multimodal capabilities can be tested in Chrome 138 Canary.
Firebase, Google's backend-as-a-service for app developers, has been bestowed with several AI-oriented improvements, including Figma import support.
Speaking of Google...
Here's what else caught our eye from the IO event:
- Google AI Ultra was launched, a $250/month subscription plan with the highest usage limits that provides access to the web giant's most capable models; this is arriving first in the US. It includes Deep Think reasoning mode in Gemini 2.5 Pro, 30 TB of storage, and YouTube premium thrown in.
- Gemini in Chrome is rolling out for Windows and macOS desktops, with caveats: It's coming to Google AI Pro and Google AI Ultra subscribers in the US who use English as their Chrome language.
- Yet more Gemini – this time Gemini Live, which can use your phone camera and screen-sharing to answer queries in real time.
- There's a Deep Research generator for pulling together public and private reports, and a vibe-coding-like tool called Canvas for producing interactive infographics, quizzes, and other stuff.
Gemini Code Assist for individuals, a free version of Google's AI-coding assistant that entered public preview in February, has now entered general availability. There's also a paid version and both are now powered by Gemini 2.5 Pro.
Google Colab, a cloud-hosted Jupyter Notebook environment for data science and Python witchcraft, is also getting an AI facelift. Colab AI, currently being rolled out slowly, promises agentic assistant via Gemini 2.5 Flash.
And not to be eclipsed by Google's proprietary Gemini models, the company's open source model family Gemma has gained a few new additions. There's Gemma 3n, a preview model that can run on as little as 2GB of RAM, and MedGemma, for multimodal medical text and image comprehension. And coming soon: SignGemm, a sign language understanding model, and DolphinGemma, "the world’s first large language model for dolphins."
Those high-pitched chirps and clicks? That's how Dolphins say "thanks for the plastic." ®