Google Cloud: Gemini Enterprise Agent Platform

Google's managed AI platform was called Vertex AI until April 2026, when Google announced its evolution into the Gemini Enterprise Agent Platform and said all Vertex AI services and roadmap would be delivered through it. Most documentation, code samples and job descriptions still say "Vertex AI", so you need both names. This note maps the platform's parts to the concepts in this course and to their AWS and Azure counterparts.

Naming, Then and Now

You may read	Current framing (2026)
Vertex AI	Gemini Enterprise Agent Platform ("the evolution of Vertex AI")
Vertex AI Agent Builder, Agent Engine	Agent Studio (low-code), ADK (code-first), Agent Runtime (managed execution)
Agentspace	Folded into the Gemini Enterprise product for end users
Vertex AI Model Garden	Model Garden (200+ models)

---

The Components

flowchart TD
    MG["📚 Model Garden<br/>Gemini, Gemma, partner models<br/>(e.g. Claude), open models"] --> BUILD
    subgraph BUILD["🛠️ Build"]
        AS["🧩 Agent Studio<br/>low-code"]
        ADK["💻 Agent Development Kit<br/>code-first, model-agnostic"]
    end
    BUILD --> RUN["🚀 Agent Runtime<br/>managed execution, sessions,<br/>long-running agents"]
    RUN --> MEM["🧠 Memory Bank<br/>long-term memory"]
    RUN --> SBX["📦 Agent Sandbox<br/>isolated code execution"]
    RUN --> GOV["🛡️ Governance<br/>Agent Identity · Agent Registry ·<br/>Agent Gateway + Model Armor"]
    RUN --> OBS["🔭 Evaluation, simulation<br/>and observability"]

    style MG fill:#e8e2d9,stroke:#ccc4b8
    style RUN fill:#d8dfe8,stroke:#b0bac8
    style GOV fill:#ddd8e4,stroke:#b8b0c8
    style OBS fill:#dde4dc,stroke:#b0c4b0

Component	What it does	Course link
Model Garden	Catalog and deployment of Google, partner and open models	Model Landscape
Agent Studio / ADK	Build agents visually or in code; ADK is Google's open-source agent framework	Agent Frameworks
Agent Runtime	Hosts agents with sessions, scaling and support for long-running work	Production Agents
Memory Bank	Managed long-term memory across sessions	Agent Foundations
Agent Sandbox	Isolated execution for model-written code	Production Agents
Agent Identity, Registry, Gateway	Per-agent identities, an approved catalog of tools and skills, and a policy-enforcing gateway with Model Armor protection against prompt injection and data leakage	Security & Compliance
Evaluation, simulation, observability	Test agents before release and trace them in production	Evaluation & Benchmarks
Managed RAG and search (RAG Engine, Vertex AI Search, Vector Search)	Managed retrieval pipelines and search over your documents	Managed RAG on Cloud Platforms

---

Where Workloads Run

Workload	Managed option	Build-your-own option
Call a hosted model	Model Garden endpoints	-
Serve an open model	Model Garden one-click deploy	vLLM/SGLang on GKE with GPUs or TPUs (see LLM Serving on Kubernetes)
Run an agent	Agent Runtime	Your framework on Cloud Run or GKE
Train or fine-tune	Managed tuning for supported models	GKE or TPU slices (Ironwood/TPU7x pods scale to 9,216 chips - see Accelerators & Interconnects)

---

Check Yourself

Exercises

Study Notes

Must-know:

• Vertex AI → Gemini Enterprise Agent Platform (announced April 2026); expect both names in the wild
• Components: Model Garden, Agent Studio, ADK, Agent Runtime, Memory Bank, Agent Sandbox, Agent Identity/Registry/Gateway with Model Armor, evaluation and observability
• Counterparts: AWS Bedrock + AgentCore; Microsoft Foundry + Foundry Agent Service
• Self-managed path: open models on GKE with GPUs or TPUs

References

• Google Cloud, Introducing Gemini Enterprise Agent Platform (April 2026)
• Google Cloud, Agent Platform overview
• Google, Agent Development Kit

Last reviewed: 2026-09