At the core of the release are two development paths: a low-code interface called Agent Studio and a code-first Agent Development Kit. Google said the platform also includes a reworked runtime that supports long-running agents capable of maintaining state over extended periods, alongside a Memory Bank system that preserves context across interactions.
The system also expands model access through Google’s Model Garden, offering more than 200 models, including Gemini 3.1 Pro and third-party options such as Anthropic’s Claude. This allows enterprises to choose different models depending on specific workloads while keeping operations within a unified framework.
A central focus is governance. Features such as Agent Identity, Registry, and Gateway are designed to track and control agent behavior, enforce policies, and provide auditability across deployments. Additional tools monitor performance and flag anomalies, giving organizations visibility into how agents operate in real time.
Google is positioning the platform as a way to move beyond isolated AI tasks toward systems that can handle broader business processes. Companies including Comcast, PayPal, and L’Oréal are already using the technology to support applications ranging from customer service to internal operations.
The launch comes as enterprises shift from early AI experiments to production use cases, creating demand for infrastructure that can manage complexity, security, and scale in agent-based systems.
This analysis is based on reporting from Google Cloud Blog.
Image courtesy of Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
