New agent step in Opal figures out the right tools and models it needs to accomplish the user’s objective, Google said. Credit: Shutterstock Google has added an agent step to its Opal tool for building AI-powered mini-apps. Powered by the Gemini 3 Flash model, the new agent in Opal enables autonomous workflows that plan, reason, and execute on the user’s behalf, Google said. Introduced February 24 and available to all Opal users, the agent step upgrades Opal workflows from static model calls to agentic intelligence, according to Google. Now, instead of manually picking a model, developers can select an agent in the “generate” step. The agent then triggers the right tools and models, such as Web Search for research or Veo for video, needed to accomplish the user’s intended mission. The agent can also make use of persistent memory, dynamic routing, and interactive chat with the user. With persistent memory, the agent can use Google Sheets to remember information across sessions, such as style preferences or ongoing lists, thus making mini-apps smarter the more they are used. With dynamic routing, the agent evaluates work and decides which steps to trigger next, bringing autonomy to mini-apps. With interactive chat, the agent can initiate a chat with the user to gather missing information or offer choices before moving to a plan’s next stage. Google presented an example in which creating a storybook Opal had required predefining page counts and user questions. Users now can build a Visual Storyteller Opal where the agent step autonomously decides which details it needs and suggests plot points to help direct where a story goes. This marks a shift from rigid formats to dynamic, unique narratives shaped by creative decisions, Google said. Artificial IntelligenceGenerative AISoftware DevelopmentDevelopment Tools