Google Cloud expanded general availability of its Gemini Enterprise Agent Platform on July 30, adding seven features that treat autonomous agents as first-class enterprise workloads: Agent Memory Bank, Agent Runtime, Agent Identity, Agent Gateway, Agent Registry, Agent Evaluation, and Agent Observability.
The most significant addition is Agent Runtime, which allows agents to run continuously for up to seven days. That shifts agents from single-session chatbot interactions to multi-day workflows spanning sales prospecting, compliance monitoring, and IT incident response. Agent Memory Bank preserves structured context from previous interactions, so agents retain user preferences, prior decisions, and account history across sessions rather than starting from scratch each time.
Identity as Infrastructure
Agent Identity introduces a native IAM type specifically for agents, with permissions tied to the runtime and built around least-privilege access. This targets a problem that has plagued enterprise automation for years: orphaned service accounts, unclear ownership, and permissions that expand unchecked over time.
“At Best Buy, we see as more organizations adopt AI agents, agent identity is becoming just as important as human identity,” Kishor Patil, Senior Manager of Cloud Platform Engineering at Best Buy, told SecurityBrief Australia. “Agent Identity helps bring accountability and governance to autonomous systems by making it clear who an agent is, what it can access, and who is responsible for it.”
Agent Gateway acts as a policy enforcement point across an organization’s entire agent estate, while Agent Registry provides a central catalog of agents, servers, and connections to reduce duplication and give administrators visibility into what has been built.
Production Monitoring
Agent Observability provides tracing and dashboards around reasoning, tool use, and execution. Agent Evaluation monitors production performance and detects drift over time. The pairing addresses a recurring issue in enterprise AI: agents that behave differently in production than in testing, especially when they interact with live data, changing prompts, or connected tools.
Enterprise Deployments
AT&T is using Agent Memory Bank for sales applications across app, voice, and web channels. “Our autonomous and intelligent AI sales agents in the app channel can resume conversations after a gap by synthesizing key facts from prior customer interactions, effectively moving from guessing to remembering,” said Jeff Dixon, AVP of Digital Product Management and Development at AT&T, according to SecurityBrief Australia.
Commerzbank is evaluating Agent Registry and Agent Gateway as part of its governance strategy. “These services are key to our governance strategy, offering vital controls for agent discoverability, policy enforcement, and access management,” said Seenuvasan Devasenan, Cluster Architect at Commerzbank’s AI Transformation Office, per SecurityBrief Australia. Liberty Global and WellSky are also listed as users or evaluators.
The Governance Stack Race
The release positions Google Cloud alongside a growing cluster of vendors building the operational layer around agent deployments. Delinea and Neo launched runtime authorization platforms this week. Samaya published the FrontierFinance benchmark for measuring agent performance in production workflows. NanoClaw partnered with Echo on CVE-free agent runtimes. The pattern is consistent: the market is moving from “can this agent do the task” to “can this agent be audited, constrained, and governed while doing the task.” Google Cloud’s bet is that the cloud provider who owns identity, observability, and governance for agents will own the enterprise deployment budget, the same way AWS captured infrastructure spend by owning compute, networking, and IAM together.