Google shipped three things at I/O 2026 that belong in the same sentence: Gemini 3.5 Flash, Google Antigravity 2.0, and Managed Agents in the Gemini API. Taken together, they represent Google's clearest public statement yet about how it plans to make Gemini the default layer underneath products it didn't build itself. The home is where that bet eventually lands.
The I/O announcement framed the combined release as a shift "from prompts to action." That framing is earned. These are not incremental API additions. They are the developer-facing version of the infrastructure Google already runs inside its own products.
What Google Antigravity 2.0 includes: desktop app, SDK, CLI, and Gemini API managed agents
Antigravity 2.0 arrives as three distinct surfaces. The desktop application gives developers a central hub for orchestrating multiple agents running in parallel, with support for scheduled background tasks and integrations across AI Studio, Android, and Firebase.
The Antigravity CLI offers the same capabilities without a graphical interface, and Google is now directing Gemini CLI users to migrate there. The Antigravity SDK gives developers programmatic access to the same agent harness powering Google's own products, with the option to host on the infrastructure of their choosing, per the announcement.
Managed Agents in the Gemini API are the complement. With a single API call, a developer can spin up an agent that reasons, calls tools, and executes code inside an isolated Linux environment. Sessions are stateful and resumable files, and context persists across turns, so developers don't have to write their own context management plumbing. Managed agents run on Gemini 3.5 Flash and are available through the Interactions API and in Google AI Studio, the announcement confirmed.
The model underneath matters. Google says 3.5 Flash outperforms Gemini 3.1 Pro across almost all benchmarks while running four times faster than comparable frontier models. Speed and cost are what make broad deployment viable, including in-home devices where a two-second lag feels broken. Gemini 3 Flash Preview is priced at $0.50 per million input tokens and $3 per million output tokens, with context caching enabling up to 90% cost reductions in repeated-token scenarios, according to Google's Flash.
The distribution path Google is building around these tools is direct. AI Studio now supports building Android apps from a prompt and publishing them directly to the Google Play test track. Entire projects can move from AI Studio to local development in Antigravity to production in a single click. Agents can natively call Google Workspace APIs and embed them in applications. These are not abstract platform capabilities; they are a pipeline from idea to consumer product running through surfaces Google already controls, as detailed in the I/O highlights.
The home is already the target market
The developer platform announcement doesn't exist in isolation. Google has been working to bring Gemini models into Nest cameras, Google Home, and Google Assistant on speakers and displays since announcing those plans in 2024, with the initial rollout scoped to Nest Aware subscribers in public preview before broader expansion, per a Google Nest blog post from that period.
The goal described then: cameras that move from recognizing "animal detected" to understanding "the dog is digging in the garden," and a Home app that lets users describe automations in plain language rather than configure them manually.
The gap between that announcement and a fully deployed consumer experience is real. Just last week, Google announced that Gemini for Home should respond more quickly to voice commands and provide more consistent responses. That's a meaningful improvement, but it's also a sign that the consumer side is still catching up to the developer tooling.
The developer platform and the home product are on converging tracks. Developers building on Antigravity and Managed Agents are building for Android, for Workspace, for Play, the same surfaces Google has been pointing at living rooms and kitchens for the past two years.
What the openness delivers, and what it actually costs
Google uses the language of openness here in ways that deserve some unpacking.
The genuinely open parts: Google open-sourced DeepSearchQA last December, a 900-task multi-step research benchmark across 17 fields, complete with a dataset, leaderboard, and technical report. Google is trying to shape how the field measures agent quality by publishing the evaluation standard itself, not just claiming a leaderboard win, per the Gemini Deep Research post.
The Agent Development Kit and Agent2Agent protocol now support the Interactions API, and Google has committed to extending compatibility to the broader open source ecosystem, with wider tool support expected in the coming months. The Interactions API itself supports remote Model Context Protocol tool calls, as Google's Interactions API post explains.
The parts worth holding with some care: the Interactions API, which sits underneath Managed Agents, has been in public beta since late last year and remains subject to breaking changes. Google says generateContent remains the primary path for standard production workloads, while the Interactions API is in public beta and subject to breaking changes.
Google says the Antigravity SDK lets developers host agents on the infrastructure of their choosing. That's accurate. It's also true that the zero-friction path AI Studio to Antigravity to Android to Play runs entirely through Google-owned surfaces. Workspace integration, Cloud billing, Android distribution: each one deepens the dependency. The blueprints are open. The smoothest construction site is Google's.
The bet, and the gap
The strategy visible in this week's announcement is a recognizable one. Google is not trying to win the home by building every device. It is trying to become the agent layer underneath products that other people build, the same approach that put Android on billions of phones. Externalizing Antigravity, publishing DeepSearchQA as an open benchmark, and supporting MCP and A2A are how you seed an ecosystem, not how you ship a single product.
The near-term gap is still real. The developer platform is ahead of the consumer experience it is supposed to power. Last week's confirmation of basic speed improvements for Gemini in smart home controls is progress, but it's incremental. The consumer reality and the developer ambition are not yet in the same place.
The signal to watch is concrete: the Interactions API and Gemini Deep Research are still listed as coming soon to Vertex AI, according to Google's own December post. Enterprise adoption at scale on Google Cloud is what distinguishes a strategic platform from a well-designed developer funnel. If the Vertex AI rollout happens and independent benchmarks validate Google's performance claims about 3.5 Flash, the gap between "open platform" and "Google-centered ecosystem with accessible entry points" starts to close. If neither happens quickly, that distinction becomes the argument against the whole thesis.
Comments
Be the first, drop a comment!