Google Labs adds an agent step to Opal to turn static flows into agentic AI workflows

Google said on February 24, 2026 that it is launching a new agent step in Opal, available for all users, to turn static workflows into agentic AI workflows. Instead of manually wiring a single model call, users can now select an agent in the generate step and let Opal determine which tools and models it needs to reach the stated objective.

Google describes this as a move from static model calls to agentic intelligence. In the company’s explanation, the new step understands a user’s goal, chooses the best path, and can invoke different tools and models as needed. The examples in the post include Web Search for research tasks and Veo for video tasks, with the broader point being reduced manual configuration for more complex automation.

• The new agent step is available to all Opal users from launch.
• Google positions it as a way to build interactive experiences rather than one-way workflows.
• Opal now also adds Memory so an agent can retain names, preferences, and running lists across sessions.

Google’s before-and-after example is an interior design workflow. Before the update, an Opal could take a room image plus a style input and return a redesigned image in a mostly one-directional flow. With the new agent step, Google says the upgraded “Room Styler Opal” behaves more like a collaborative design partner, reaching back to the user when it needs input and selecting the best models and tools on its own.

That matters because many no-code AI workflow builders still depend on brittle step chaining. By pushing more orchestration into the agent layer, Google is trying to make multi-step workflows easier to author and more adaptive at runtime. The addition of Memory also points toward longer-lived, personalized agents rather than disposable one-off flows.

The competitive question is reliability. If the agent step consistently chooses the right tools and knows when to ask for human input, Opal becomes more than a prompt wrapper. If not, the underlying complexity simply moves from the user’s canvas into the system. Either way, the release is a clear sign that agentic product claims are shifting from demos into workflow builders.