Google Cloud unveils persistent Gemini Agents for enterprise tasks

Google Cloud wants its AI agents to stop acting like chatbots and start acting like colleagues. The company has unveiled Gemini Agents that can run long tasks, move across multiple apps, and keep their own email inbox, calendar, and Drive storage. The announcement came on April 22, 2026, at Google Cloud Next, the company’s annual conference in Las Vegas. The agents sit inside a broader product called the Gemini Enterprise Agent Platform. It is built to help businesses build, scale, govern, and fine-tune autonomous AI agents powered mainly by Gemini models. What Google actually shipped Under the hood, it is an evolution of Vertex AI, Google’s existing machine learning platform. Instead of just hosting models, it now handles the full lifecycle of agents that are expected to operate on their own. These agents can work autonomously on complex enterprise tasks for up to several days. Subsequent updates set the maximum runtime at 7 days. A feature called Memory Bank lets agents retain context over time, so they don’t forget what they were doing halfway through a workflow. Google also built in a stack of developer tools. The lineup includes: Agent Development Kit (ADK): a code-based toolk...