What changed in Google AI Studio after I/O 2026: Build mode, Android, Workspace and Gemini 3.5 Flash

For a long time, Google AI Studio sat as "the playground where you prompt Gemini." I/O 2026 turned it into a proper app-building environment: not just web, not just mobile, internal tools sitting directly on top of Workspace data — all in the same interface. This post collects the last few weeks of updates in one place.

TL;DR

• Build mode now covers native Android. Kotlin + Jetpack Compose phone apps, in the same vibe-coding flow as the web side.

• Workspace is directly inside. Dashboards on Sheets data, internal tools talking to Drive, apps that work with team documents — all from prompts.

• One-click Cloud Run + Firebase. Full-stack runtime; you don't need to open a separate terminal to deploy.

• AI Studio now has a mobile app. Iterate on builds from your phone.

• Custom assets via Nano Banana. The build agent generates app-specific images itself; an annotation tool lets you draw on components inside the preview window.

• Export to Antigravity. Conversation history, files and secrets all travel with you into Google's new agentic dev environment.

• New default model: Gemini 3.5 Flash. Outperforms the older 3.1 Pro on most benchmarks while running roughly 4× faster.

Why Build mode got so big

The old "chat + model parameters" surface of AI Studio is still there, but the spotlight has moved to Build mode. You describe an app in natural language, sprinkle in AI Chips to attach capabilities ("add image generation here", "add Google Maps data there"), and iterate over the preview. Before, this was web-only. Now:

• The web side runs on a full-stack runtime — backend logic doesn't need a separate Vercel/Cloudflare setup; one-click deploy to Cloud Run.

• The Android side emits Kotlin + Compose; you can ship an APK, or export to Android Studio if you want to keep iterating in a real IDE.

This blurs the line between "prompt UI for demos" and "studio you actually ship from." It expands the "how far can a solo dev go alone" game — and it's basically Google's version of the trend I wrote about in Claude's design choices for solo builders.

Workspace integration: dashboards from prompts

The quietly biggest news, in my read. Inside Build mode, Google Workspace is now a first-class data source:

• A few prompts get you a dashboard on Sheets data.

• You can build internal tools that talk to the docs your team already lives in on Drive.

• You can ship shareable apps that handle each user's own Drive/Sheets — straight from prompts.

This is AI Studio quietly stretching out of "for developers only" and into the internal tools market. Calling it a direct Retool/Glide competitor would be premature, but the direction is clear.

A mobile app: AI Studio in your pocket

There's now an AI Studio mobile app; the full build-mode experience works from the phone. You can iterate code, preview a build, even check an Android build — without going back to the desktop. I haven't gotten my hands on it yet; once I do, I'll add a short first-impressions note at the bottom of this post.

Design side: Nano Banana + preview annotation

The build agent generates app-specific visuals (illustrations, heroes, icons) with Nano Banana. The new edit tool lets you draw directly inside the preview — circle a component and say "make this softer", or ask for a brand-new visual to drop in. That's a sharply different proposition than the "design in Figma, dump into code" loop: design lives inside the build itself, not as a separate step.

Export to Antigravity

When AI Studio starts to feel too small — when you want deeper agentic control or a proper IDE — you can export the project as-is into Google's new dev environment, Antigravity. Conversation history, files and secrets all come along. Antigravity is a topic on its own (it sits in the Windsurf / Cursor / Devin lane); I wrote about where the AI coding tools market is heading and that Google was going to land in this lane with real weight — that landing has now happened.

New default: Gemini 3.5 Flash

Post-I/O, the default model is Gemini 3.5 Flash:

• Beats the older 3.1 Pro on most benchmarks.

• Runs roughly 4× faster.

• Its cost/latency profile fits Build mode's "type, see, change" loop — the magic of vibe-coding dies the moment the model is slow, and Google was deliberate here.

For more complex agent flows, Gemini 3 Pro is still there; but having Flash as the default for everyday building is the right call.

What it adds up to

Three lines, the way I see it:

1. AI Studio is no longer Vertex AI's "little sibling." It's becoming a standalone product for solo devs and small teams. 2. Designers, ops people, internal-tool builders are now in the audience too. Workspace + Nano Banana + preview annotation together open a door for the "I'm not a developer but I need an app" segment. 3. Google's "model + IDE + cloud" line is now complete. Gemini 3.5 → AI Studio Build mode → Antigravity → Cloud Run; you start and finish inside the same wall. That's a directly comparable stance against OpenAI's Codex + Apps SDK or Anthropic's Claude Code + Claude Platform line.

Closing note — Singrey

Honestly I'd been using AI Studio mostly as a "let me poke at Gemini" place, not as a serious build tool. After this update wave, my plan for the first free weekend is to build a tiny internal tool — most likely a small dashboard that reads my own blog metrics from Sheets — and actually ship it. How smooth the Workspace integration really is in practice (auth, permissions, edge cases) will matter way more than how it looks in a keynote demo. Once I try it, a short first-impressions post will follow.