Deployment and Runtime
Explore the trade-offs between Agent Runtime, Cloud Run, and GKE for hosting enterprise agents. Learn to choose a runtime based on the customer’s operational capacity, networking, security, and workload needs, and to encode those constraints in deployment configuration.
An agent that works on a developer’s laptop is still a long way from production. In a customer environment, the agent has to run reliably for many users at once, reach private systems inside the customer’s network, meet security and compliance requirements, and fit the skills of the team that will operate it after the FDE engagement ends. The choice of runtime affects all of these, and a choice that seems convenient during a pilot can create serious problems once the system goes live.
Google Cloud offers three main options for running agents: Agent Runtime, Cloud Run, and Google Kubernetes Engine (GKE). Each is a sound choice in the right situation. FDE interviewers rarely ask which option is best in general. They ask which option fits a specific customer, and they listen for whether the answer is driven by the customer’s constraints rather than by the candidate’s personal preference.
The three runtime options
The options differ mainly in how much of the infrastructure Google manages and how much control the customer retains.
Agent Runtime is the managed runtime within Gemini Enterprise Agent Platform, previously known as Vertex AI Agent Engine. It is designed specifically for agents. It hosts the agent as a managed service and provides agent-specific capabilities such as Agent Sessions for conversation history, Agent Memory Bank for long-term memory, and Agent Sandbox for safe code execution. It integrates directly with the platform’s identity, governance, evaluation, ...