Google Cloud rebranded Vertex AI as the Gemini Enterprise Agent Platform at Cloud Next 2026 this week, consolidating what had been a collection of individual AI services into a single platform for building, scaling, governing, and operating autonomous agents. The conference drew over 32,000 attendees to Las Vegas and produced 260+ announcements across four days, according to Google’s wrap-up.
The rebrand is not cosmetic. Google absorbed Agentspace, its employee-facing AI assistant product, into a unified offering called Gemini Enterprise. The underlying platform gained 14 distinct agent infrastructure components that together form an end-to-end stack, from development to runtime to governance to security, The Next Web reported.
The Stack
The Agent Development Kit (ADK) reached v1.0 stable releases across four programming languages, enabling developers to organize agents into networks of sub-agents using a graph-based framework. Agent Studio provides a visual no-code-to-code progression: business users can build in a visual canvas, then export to ADK for full-code customization, according to Google.
Agent Runtime delivers sub-second cold starts and provisions new agents in seconds. Agent Sandbox provides hardened code execution for model-generated code and browser-based automation. Agent Memory Bank, now generally available, gives agents persistent learning across sessions with dynamically generated Memory Profiles for high-accuracy recall. Agent Sessions handles conversation history with custom session IDs that map directly to internal databases and CRM records, per Google’s blog.
Governance and Security
On the governance side, Agent Identity assigns every agent a cryptographic ID with auditable trails mapped to authorization policies. Agent Registry maintains a central library of indexed, approved tools and skills. Agent Gateway provides fleet management with unified connectivity, consistent security policies, and Model Armor protections against prompt injection and data leakage. Agent Anomaly Detection flags suspicious behavior using statistical models and an LLM-as-judge framework, working alongside Agent Threat Detection for malicious activity monitoring, Google wrote.
The Model Garden now hosts over 200 models including Google’s Gemini and Gemma families, Anthropic Claude, and open models like Meta’s Llama. Partner agents from Box, Workday, Salesforce, ServiceNow, and S&P Global are integrated directly into the platform, The Next Web noted.
The Competitive Positioning
Google Cloud CEO Thomas Kurian framed the strategy explicitly against competitors: other vendors are “handing you the pieces, not the platform,” he said during the keynote, leaving teams to integrate components themselves, according to The Next Web.
The positioning targets a real gap. OpenAI’s enterprise push relies on ChatGPT and Codex as application surfaces with partners handling integration. Anthropic’s Model Context Protocol has reached 10,000 servers and 97 million monthly SDK downloads, but Anthropic does not own the underlying infrastructure. Google’s bet is that vertical integration, owning the model, the silicon (8th-gen TPUs announced the same week), the runtime, and the distribution channel through Workspace’s 3+ billion user base, creates a lock-in advantage neither competitor can match.
The Agent2Agent (A2A) protocol, Google’s cross-platform agent communication standard, is now in production at 150 organizations. That metric matters because it measures interoperability adoption, the one thing a vertically integrated platform needs to avoid becoming a walled garden, The Next Web reported.