Google announced Gemini 3.5 Flash at I/O yesterday, and the company is positioning it as an agent-first model rather than another chatbot upgrade. This one skipped the usual preview phase and went straight to general availability. It’s already powering the Gemini app, AI Mode in Google Search, and the company’s new developer platform.
The model is designed for autonomous task execution and coding workflows. Google says it can build software from scratch and handle complex multi-step tasks without constant supervision. That’s the pitch, anyway. The real test will be how it performs compared to Claude 3.7 Opus and GPT-5, both of which have been competitive on agentic benchmarks.
Pricing is higher than Gemini 3.0 Flash. Google hasn’t published the exact numbers yet, but Simon Willison notes it’s “more expensive” and that Google plans to use it for everything. The llm-gemini plugin already supports it as of version 0.32.
Google also launched Antigravity 2.0, its agent-first development platform. The update includes a redesigned desktop app and a new CLI tool. Developers can now build and deploy agentic workflows directly from the terminal, which should make it easier to integrate with existing toolchains.
The bigger news is the new AI Ultra plan at $100/month. That’s the same price as ChatGPT Pro, and it gives you 5x the usage limit of the $20 AI Pro plan. Google is clearly betting that developers will pay for higher limits as they build more complex agent workflows.
Antigravity isn’t a new product. Google has been building it for a while, but this is the first major version bump since launch. The 2.0 release adds better support for long-running tasks, improved error handling, and tighter integration with Google Cloud services.
Google also shipped Android CLI, a command-line tool designed to work with AI coding agents like Claude Code and OpenAI’s Codex. The idea is that agents can now build Android apps without needing to navigate Android Studio’s GUI.
This is a logical move. If you believe AI agents are going to write a lot of code, then your tooling needs to be agent-friendly. Command-line interfaces are easier for models to use than point-and-click workflows. Android CLI provides scaffolding, build management, and deployment commands that agents can call directly.
It’s still early, but this could lower the barrier for spinning up Android prototypes. If an agent can handle the boilerplate and project setup, developers can focus on the parts that actually matter.
Google also announced Gemini Spark, a personal assistant that runs continuously in the background. It’s built on Gemini’s base models and uses an agentic harness from Google Antigravity.
The Gmail integration is the key feature. Spark can read your email, monitor threads, and surface relevant information without you asking. Google is positioning it as a proactive assistant rather than a reactive chatbot. It can alert you to updates, track ongoing conversations, and pull in context from other Google services.
This is useful if you trust Google with that level of access. If you don’t, it’s a hard pass. There’s no pricing information yet, but it will likely be bundled with the AI Pro or Ultra plans.
Google is also rolling out AI-powered information agents that monitor topics in the background and alert you to updates. You can set up an agent to track a specific subject, competitor, or research area, and it will notify you when something changes.
This is different from Google Alerts. These agents are supposed to understand context and filter for relevance, not just keyword matches. If it works, it could be genuinely useful for staying on top of fast-moving topics. If it’s just glorified keyword monitoring with extra compute, it’s not worth the effort.
If you’re building agentic workflows or experimenting with AI coding tools, the Antigravity 2.0 and Android CLI updates are worth checking out. The CLI especially makes sense if you’re already using Claude Code or similar tools.
Gemini 3.5 Flash is interesting if you need a model optimized for tool use and long-running tasks. It’s not clear yet how it stacks up against the competition, but Google’s willingness to ship it across all their products suggests they’re confident in the benchmarks.
The $100 Ultra tier is expensive, but if you’re hitting rate limits on the $20 plan, it might be worth it. That said, you should compare it against ChatGPT Pro and Claude’s highest tier before committing.
If you’re not building agents or using AI coding tools regularly, most of this is skippable. Gemini Spark and the information agents are consumer-facing, but they’re not available widely yet. Wait for reviews before giving Google that much access to your email.
One email at dawn. The five stories that mattered, with the bits removed and the meaning kept. Free, for now.