Google Launches AI Agents That Run for Days on Gemini Enterprise Platform
Updated
Updated · O'Reilly Media · Jun 8
Google Launches AI Agents That Run for Days on Gemini Enterprise Platform
3 articles · Updated · O'Reilly Media · Jun 8
Summary
Google has turned long-running AI agents into a formal enterprise product, folding Vertex AI into the Gemini Enterprise Agent Platform with Agent Runtime, Sessions and Memory Bank.
Agent Runtime is designed for agents that run autonomously for days, with sub-second cold starts and on-demand sandboxes, while Sessions persist event history and custom IDs can tie agent state to CRM or database records.
Memory Bank, now generally available, stores long-term user-scoped memory for later retrieval; Google said Payhawk cut expense-submission time by more than 50% using a Memory Bank-backed agent.
The launch packages operational controls enterprises need—sandboxed execution, agent orchestration, identity, gateway, observability and simulation—so teams do not have to build persistence, recovery and audit trails from scratch.
Google’s move reflects a broader shift from chat-style agents toward systems that can sustain multi-hour or multi-day work, as vendors race to productize durable state, verification and recovery for autonomous AI.
What new security is needed to stop autonomous AI from becoming the ultimate corporate insider threat?
How do we stop autonomous AI agents from secretly changing their goals during month-long tasks?
When an autonomous AI causes a billion-dollar disaster, who is legally held responsible for the damages?
Inside Google’s Gemini Enterprise Agent Platform: Transforming Enterprise Automation and Slashing AI Costs by 78%
Overview
In June 2026, Google launched the Gemini Enterprise Agent Platform, building on its earlier Antigravity agentic development platform. This new platform uses agentic AI and built-in workflow orchestration to help organizations automate and streamline complex workflows, moving from initial requests to instant execution. Deeply integrated with Google Cloud and powered by advanced eighth-generation TPUs with a dual-chip design, the platform meets the heavy computational needs of modern enterprise agents. This evolution marks a shift toward a unified control plane for intelligent agents, enabling businesses to achieve greater operational efficiency and unlock new levels of productivity.