If you’ve been waiting for enterprise AI agents to grow up, Google just made a strong argument that the wait is over. On July 29, Google pushed several major Gemini Enterprise Agent Platform features into general availability, moving well beyond the announcement-stage version of the platform it showed off at Cloud Next in April.

The practical implications are significant for any business that’s moved past AI experimentation and is now trying to deploy agents that actually do real work.

What Actually Shipped

Agents that work for seven days straight. The new Agent Runtime lets agents run asynchronously for up to seven days. That might sound like a technical detail, but it changes what you can automate. Multi-stage sales sequences, onboarding workflows, procurement approvals that span several departments — those are the kinds of processes that previously required a human to kick off each step. An agent that can stay alive and on-task across a full business week can handle them end to end.

Agents that remember people. Agent Memory Bank gives agents structured schemas that automatically extract and maintain context from conversations. User preferences, account history, past interactions — the agent holds onto this across sessions. The practical effect is that your AI agents stop feeling like they’re meeting your customers for the first time every single call.

Agents with their own identities. This is the one that matters most for security teams. Agent Identity is a native IAM type built on open standards that enforces least-privilege access, mitigates token theft by binding credentials to the agent runtime, and produces non-repudiable audit logs of everything the agent did and touched. When an auditor asks what your AI agent accessed and when, you can actually answer.

A control point for all agent traffic. Agent Gateway sits between your agents and your systems, enforcing granular access controls and natural language rules. It includes Google’s Model Armor protection against prompt injection attacks and data leakage — two of the most common ways badly configured AI agents create security incidents.

A registry to stop agent sprawl. Agent Registry gives organizations a centralized library of all their agents, making them discoverable and reusable. This matters because the biggest AI governance problem right now isn’t security — it’s that nobody knows how many agents they actually have running. A recent OutSystems survey found 94% of organizations are worried that AI sprawl is increasing their technical debt and security risk. A registry doesn’t solve sprawl by itself, but it makes it visible, which is the first step.

Automated code security. CodeMender, a managed code security agent, handles automated code remediation and zero-day risk reduction. It’s an interesting signal about where Google sees AI agents going: not just as customer-facing tools, but as internal engineering and security infrastructure.

Why This Matters for Businesses Right Now

Most enterprise AI deployments still live in a fragile middle ground. The AI can answer questions and generate content, but it can’t actually complete multi-step work independently. You’re still relying on humans to monitor progress, approve next steps, and restart anything that stalls.

The features Google shipped change that equation for organizations that run on Google Cloud. An agent with memory, a week-long runtime, a proper identity, and a governance layer surrounding it can take on the kind of process work that actually moves a business — not just the kind that saves a few minutes on email.

The harder question is whether organizations are ready to use it. Deploying agents with this level of capability requires thinking clearly about what they’re allowed to access, what they’re supposed to do, and how you’ll verify they did it correctly. The infrastructure for that is now available. The organizational readiness is the next problem.

What This Means for Business

For organizations already using Google Cloud, these features remove the main technical objections to deploying production-grade AI agents — and they’re available now, not on a roadmap.

For organizations still evaluating AI agent platforms, it’s a meaningful competitive shift. The gap between enterprise-grade and experimental AI infrastructure just got wider, and Google is now squarely on the enterprise end of that gap.

If you’re building an AI-powered business and wondering which platform can carry the weight of real production workflows, Google’s move to general availability is a clear signal that agentic AI infrastructure has matured faster than most expected.

For Enterprise DNA clients deploying Omni AI agents, the principles here — memory, identity, governance, runtime duration — apply regardless of which underlying platform you’re building on. These aren’t Google-specific concerns. They’re the fundamental requirements for any AI agent deployment that’s supposed to actually work in production.