Google Cloud AI agent development and infrastructure

Google Cloud has shifted its focus toward the implementation of AI agents, emphasizing that successful integration requires companies to transform their business models and operational processes rather than just adopting new models. To support this, Google Cloud is integrating OpenAI models and the Codex coding tool into its development suite, including Gemini Agent.

To facilitate autonomous execution, Google Cloud introduced the Agent Compute Engine. This specialized infrastructure hosts agents in isolated, virtualized environments using micro virtual machines (microVMs) and Cloud Hypervisor to ensure security while agents execute code and manage complex workflows. The architecture utilizes Titanium hardware and Infrastructure Processing Units to maintain resource separation. Key features include the ability to pause workspaces to reduce resource consumption and the use of Hyperdisk storage for persistent memory states. A preview version is expected in the coming months, with the technology already being used internally for Gemini tasks.

Maureen Costello, Google Cloud VP, has characterized this shift as a “third wave” of workplace technology, where AI evolves from a tool into a “digital colleague” capable of coordinating work across business functions with minimal human intervention. This transition is reflected in a reported 14-fold increase in AI token volumes on Google’s platform between February and August 2026. To support these workflows, Google Workspace is introducing native markdown (.md) support in Google Docs and Drive, allowing developers and agents to collaborate on structured documentation without manual conversion.