Google’s New Gemini AI Model and Tools Are All About Agents Now

Google’s got a new Gemini AI model, and it’s built to be agentic. The company debuted Gemini 3.5 Flash Tuesday at its Google I/O annual developers conference, with a more powerful Gemini 3.5 Pro expected next month.

Google says its new AI models are comparable with other frontier and near-frontier models — like OpenAI’s GPT-5.5 or Anthropic’s Claude Opus 4.7 — but also focused on being faster and better capable of handling tasks for you. 

You’ll be able to use those models in lots of different ways, including an OpenClaw-like 24/7 agentic assistant and an upgraded Antigravity coding platform. Here’s what’s coming in Gemini from Google I/O.

A new pair of Gemini models

The first new model, Gemini 3.5 Flash, rolls out to all users starting now. Google CEO Sundar Pichai told reporters during a briefing ahead of I/O that the model is “an incredible delight to use” and a “game changer” internally at Google. It’s trained to operate quickly and accomplishes tasks at about half the cost of competitor models, he said.

Gemini 3.5 Flash is particularly good at running agents — autonomous AI instances that can handle particular tasks. That includes operating multiple agents at the same time to tackle larger, long-term projects. Koray Kavukcuoglu, CTO of Google DeepMind and Google’s chief AI architect, said it can handle sessions lasting multiple hours on its own, completing whole coding research projects. 

Still to come is Gemini 3.5 Pro, which Pichai said Google has been testing internally.  

Spark: A 24/7 AI assistant

In what looks like a challenger to OpenClaw, the autonomous AI agent platform that blew up at the start of the year, Google is preparing an assistant called Gemini Spark. Josh Woodward, vice president for Google Labs, the Gemini app and AI Studio, said the company is releasing Spark “deliberately,” with a group of trusted users getting access this week, followed by a beta for AI Ultra subscribers next week.

Spark is an always-on AI agent that can act on your behalf and follow directions. It’s based in Google’s cloud — so no need to buy one of those fast-disappearing Mac Minis to run OpenClaw — and can perform tasks in Workspace apps like Gmail and Docs. 

“When you use it, it almost feels like you’re tossing things over your shoulder, Spark’s catching them and getting the job done,” Woodward said.

Google testers are using it to plan parties, track school schedules and monitor inboxes for questions, Woodward said. Google plans more connections over the summer so that Spark can operate a wider variety of third-party apps and websites through Chrome. 

Expanding Antigravity

Coding is one area where Google has lagged behind rivals Anthropic and OpenAI, which have thrived with Claude Code and Codex, respectively. Now Google has reimagined its platform, Antigravity, as an agent-first development tool. Google’s vision for Antigravity goes beyond coding, however. The new direction is a “platform to develop and manage teams of autonomous AI agents,” Kavukcuoglu said.

That includes a standalone desktop application for Antigravity to serve as a home for agentic AI. The tasks for these agents can vary. Kavukcuoglu gave the example of one agent creating a website, another generating brand assets and a third planning products. Antigravity will also be available in the command line. 

Gemini app updates

With all of these agentic changes to Google’s AI products, expect a big makeover for the Gemini app. Woodward said the app will include the new Gemini 3.5 Flash and Gemini Omni models, along with a few new features.

The new design, called Neural Expressive, includes a new set of animations, colors and haptic feedback when you tap buttons. Gemini will also respond with more images, timelines or visualizations when appropriate, Woodward said. 

Another feature, called the Daily Brief, is a personalized presentation of information Gemini’s AI will collect based on your inbox, calendar tasks and other data. It will be available this week for Google AI Plus, Pro and Ultra subscribers in the US. Woodward described it as “a great way to introduce people to the power of agents.”