Google Cloud has announced the general availability of several major capabilities within its Gemini Enterprise Agent Platform, expanding the tools businesses can use to build, secure, and scale AI agents for long-running, production-grade workflows.
The update, announced on July 29, 2026, by Mike Clark, Director of Product Management for Gemini Enterprise Agent Platform, brings previously limited-access features to all users, alongside the recent launch of CodeMender, Google’s managed code security agent.
Longer-Running Agents With Better Memory
Google is rolling out two features aimed at helping agents handle extended, multi-step tasks:
- Agent Memory Bank Lets developers define structured schemas so agents automatically extract and retain key context, such as user preferences, past decisions, and account history, across long-running interactions without slowing down response times.
- Agent Runtime Enables agents to run continuously for up to seven days, supporting asynchronous processes like a week-long sales sequence or a multi-stage onboarding workflow, with agents making background decisions without constant human oversight.
Security, Auditing, and Centralized Control
To address enterprise security needs, three features are now generally available:
- Agent Identity A native IAM identity type built on open standards that enforces least-privilege access for agents, binds access directly to the agent runtime to reduce token theft, and automatically manages credential lifecycles to eliminate dormant accounts.
- Agent Gateway A central control point for securing and governing all agent interactions, offering granular access controls through IAM conditions and natural language rules, plus built-in protection against prompt injection, tool poisoning, and data leakage via Model Armor.
- Agent Registry A single organization-wide catalog of agents, servers, and connections, helping teams discover and reuse existing agents instead of rebuilding them, while giving administrators visibility into agent sprawl.
Performance Monitoring and Evaluation
Two additional tools give teams visibility into how agents behave once deployed:
- Agent Evaluation Continuously monitors agent performance in production using pre-built metrics, custom Python scoring, LLM-as-a-judge evaluation, or adaptive rubrics developed with Google DeepMind, flagging performance degradation or behavioral drift.
- Agent Observability Provides end-to-end tracing of agent reasoning, tool usage, and execution performance through real-time dashboards.
Enterprise Adoption
Several large organizations shared how they’re using the new capabilities. AT&T said its AI sales agents now use Agent Memory Bank to resume customer conversations across channels without losing context. Best Buy highlighted Agent Identity for solving issues around orphaned service accounts and unclear agent ownership. Commerzbank AG and Liberty Global both pointed to Agent Registry and Agent Gateway as central to their governance strategies, while WellSky said the tools help it catalog and lifecycle-manage agents across its healthcare platforms.
Availability
All the features mentioned are available now through the Gemini Enterprise Agent Platform documentation. Google has also published a livestream walking developers through the full agent lifecycle on the platform.
