Google's agent platform push

Google used Cloud Next on April 22 to roll out a broader enterprise agent stack, centering the event on what it now calls the Gemini Enterprise Agent Platform. (blog.google) An AI agent is software that can plan steps, call tools, and complete work with some autonomy instead of just answering a prompt. Google said the new platform combines model access, orchestration, security, and deployment in one environment for companies building those systems. (blog.google) Google is also folding earlier pieces into that pitch. In a July 31, 2025 post, the company said Agentspace is now part of Gemini Enterprise, and at Next ’26 it presented the platform as the roadmap for what it called the “agentic enterprise.” (cloud.google.com) (blog.google) One of the clearest product examples is Google Workspace Studio, which Google announced as generally available in December 2025 and expanded again at Next ’26. Google said employees can use it to design, manage, share, and invoke “skills” that automate repeatable work inside Workspace. (workspace.google.com 1) (workspace.google.com 2) Another piece is Project Mariner, a browser-based research prototype Google introduced in December 2024. Google said Mariner can understand what is on a browser screen and use web elements like text, images, and forms to take actions under user supervision. (blog.google) Google’s interoperability push runs through Agent2Agent, or A2A, an open protocol it launched on April 9, 2025 so agents built by different vendors can exchange information and coordinate actions. At launch Google said more than 50 partners backed the effort, including Salesforce, SAP, ServiceNow, Workday, Deloitte, PwC, and Accenture. (developers.googleblog.com) By July 2025, Google said support for A2A had grown to more than 150 organizations, and in April 2026 the Linux Foundation said the protocol had passed the same milestone and was seeing production deployments across Google, Microsoft, and Amazon Web Services platforms. (cloud.google.com) (linuxfoundation.org) Google is pairing that protocol with its own developer tooling. The company said at Next ’25 that it was introducing the open-source Agent Development Kit, and later said the kit had native A2A support so developers could build agents, expose them to other agents, and deploy them on Agent Engine, Cloud Run, or Google Kubernetes Engine. (blog.google) (cloud.google.com) The business case is that companies do not want separate systems for models, automation, security, and workplace software. Google said nearly 75% of Google Cloud customers now use its artificial intelligence products, and 330 customers processed more than 1 trillion tokens each over the past 12 months. (blog.google) Cloud Next’s own agenda showed how tightly Google tied the conference to that message: the April 22 opening keynote was billed as “The agentic cloud,” and the April 23 program included a session called “Get real: Agents in the autonomous era.” Google’s pitch is that the agent platform is no longer a side project inside Cloud or Workspace, but the frame for both. (googlecloudevents.com)