Google previews Vertex AI Agent Orchestrator with policy guardrails

Google Cloud used its latest security and AI announcements to preview Vertex AI Agent Orchestrator, a managed layer for assembling multi-agent systems on top of Vertex models and tools. The service lets teams define agent roles, capabilities, and escalation rules, then binds those to Google Cloud IAM so each agent only gets the minimum required access to data stores, APIs, and backends. Workflows are represented as graphs, allowing orchestrator policies like “only the compliance agent can approve this action” or “never write to production without human sign-off.”

What changed. Vertex AI customers can now register agents, define their tools and scopes, and compose them into multi-step workflows under a single orchestrator that enforces IAM, logging, and policy guardrails.

Why it matters. Gartner and others have warned about “150k agents with no one in control”—this is one of the first hyperscaler moves to bake agent lifecycle, identity, and permissions into the platform rather than leaving orchestration entirely to app code.

Builder takeaway. If your current agent architecture on Vertex is a custom Python scheduler gluing tools together, consider migrating to Agent Orchestrator to offload routing, escalation, and access control to a managed service and focus your effort on domain-specific agent logic.