Google Labs introduced a new feature to its vibe-coding app Opal, an interactive agent that plans and executes tasks based on the user’s goals.
The update transforms Opal from a static tool to an agentic model, Google said in a blog post At launch. Google Labs is the tech giant’s public testing ground for new and experimental products.
The app included predefined steps that required developers to manually specify the inputs and models required for the task.
With the upgrade released on February 24, Google is automating that process, with agents designed to “understand your objective” and automatically determine the best models and actions to achieve it.
“We believe this approach gives you the best of both worlds,” Google said. “The power of an AI agent working toward your goal and control of a step-by-step workflow that you can customize and refine at any time.”
Using the gemini 3 flash modelThe updated model can more effectively plan and execute tasks, tailor responses, and remember information across sessions to create a more personalized experience, according to Google.
For example, Google highlighted an executive briefing workflow at Opal, in which an agent adjusts content based on whether the user is meeting with a new or existing client.
Opel can now search the web for background information on new prospects or review internal meeting notes to provide relevant context.
The new agents are also natively interactive, Google said in a blog post. If additional details are needed, the system prompts users for clarification or presents options to guide next steps, rather than relying on a pre-defined workflow.
The vendor said the enhancements enable even non-technical users to create more complex, multi-step workflows within their apps.
Google first introduced Opal in July 2025 as a tool for US users to create mini web applications or remix existing ones without writing code. The tool expanded to 15 more countries in October 2025, before launching in more than 160 markets a month later.
Google is not alone in turning its attention to this vibe-coding; Major AI vendors like OpenAI and Anthropic are also offering similar systems as customers prefer easy-to-use, intelligent models.
