Google · 2026-04-22 · major
Google Gemini Enterprise Agent Platform — Full-Stack Agent Infrastructure GA
Google rebrands and expands Vertex AI into the Gemini Enterprise Agent Platform — a four-pillar Build/Scale/Govern/Optimize system for autonomous agents. Includes Agent Studio, Memory Bank, cryptographic Agent Identity, Agent Gateway, and 200+ models in Model Garden. GA April 22.
Google rebrands Vertex AI into a complete enterprise agent platform — Build, Scale, Govern, and Optimize, GA April 22.
What is it?
Announced at Google Cloud Next 2026, the Gemini Enterprise Agent Platform (GEAP) replaces and significantly expands Vertex AI. It's organized into four pillars: Build (Agent Studio low-code canvas + Agent Development Kit for code-first), Scale (Agent Runtime with sub-second cold starts and multi-day workflows, Agent Memory Bank for persistent context), Govern (cryptographic Agent Identity and Registry, Agent Gateway with security and policy enforcement, anomaly detection), and Optimize (Agent Simulation, Evaluation, Observability, continuous Optimizer). Model Garden hosts 200+ models including Gemini 3.1 Pro, Gemma 4, and third-party models like Claude. Workspace Intelligence simultaneously ships as an AI-aware semantic layer across Gmail, Docs, Drive, and Calendar.
How does it work?
Developers build agents in Agent Studio or the ADK (Python/TypeScript), connect sub-agents via Google's A2A protocol, and register them in Agent Registry. Agent Gateway enforces unified security policies and routes traffic across the fleet. Each agent gets a cryptographic Agent Identity for complete audit trails and anomaly detection. Memory Bank provides persistent state across sessions. Agent Simulation lets teams test agent behavior before deployment, and Evaluator scores outputs against ground-truth datasets. Workspace Intelligence adds a semantic knowledge graph of organizational data as ambient context for agents running on GEAP.
Why does it matter?
GEAP is Google's clearest signal that enterprise AI competition has shifted from frontier model benchmarks to agent infrastructure. It directly challenges Anthropic Managed Agents and OpenAI's Agents SDK with a vertically integrated platform that bakes governance, security, and continuous optimization in by default. Google also committed $750M to its 120,000-partner ecosystem to accelerate enterprise agent adoption. The cryptographic Agent Identity system — giving each agent an auditable trail — addresses the enterprise compliance gap that has slowed AI adoption in regulated industries.
Who is it for?
Platform engineers and IT architects building and governing enterprise AI agents
Try it
google.github.io/adk-docs/ — Python/TS Agent Development KitKey numbers
- Models in Model Garden: 200+