Google just turned Opal from a simple mini-app builder into something far more interesting: an agentic workflow platform.

The new “agent step” replaces static model calls with something that actually thinks about your goal. Instead of manually selecting Gemini Flash and hoping for the best, you describe what you want—like an interactive storybook that adapts to user input—and the agent figures out which tools, models, and steps it needs to get there.

The examples in Google’s announcement are telling. Before: a storybook Opal where you hardcoded page counts and questions. After: a Visual Storyteller that autonomously decides what details it needs, suggests plot points, and shapes the narrative based on real-time creative decisions.

Same leap for interior design. The old Room Styler was a one-way street: upload photo, pick style, get result. The new version actually collaborates. It generates a concept, asks for feedback on specific elements, researches niche sub-styles if needed, and iterates until the output feels personal rather than templated.

What’s powering this is a bundle of new capabilities: memory across sessions (so your Opals learn your preferences), dynamic routing (the agent chooses different paths based on conditions), and interactive chat (it can ask follow-up questions when information is missing).

The underlying model is Gemini 3 Flash, but the interesting part isn’t the model—it’s the architecture. Google has built a system where you get “the power of an AI agent working towards your goal and the control of a step-by-step workflow you can customize.”

This feels like the right abstraction for most users. Rigid enough to be predictable, flexible enough to handle real complexity.

Source: https://blog.google/innovation-and-ai/models-and-research/google-labs/opal-agent/