Google ships 13 Gemini Agent Platform demos for ADK

Key Takeaways

• Google released 13 demos covering the full agent lifecycle: build, scale, and govern
• Agents CLI integrates with coding assistants like Claude Code and Codex for plain-English agent scaffolding
• Demos include production patterns like human-in-the-loop approval, durable state machines, and checkpoint-resume for long-running workflows

Google Cloud published 13 hands-on demos for its Gemini Enterprise Agent Platform, walking developers through the full lifecycle of building, scaling, and governing AI agents. The demos cover everything from a first ADK agent to production deployment with human-in-the-loop approval flows, durable state machines, and governance guardrails.

The most notable addition: Agents CLI. Install it into your coding assistant of choice, whether that's Claude Code, Codex, or Antigravity, and it gains seven skills that make it an expert in ADK. Describe what you want in plain English, and the assistant scaffolds, evaluates, deploys, and monitors the agent without leaving your editor.

What do the build demos cover?

Four demos focus on foundational agent construction. The ADK Foundation codelab is the entry point: set up your environment, define a conversational agent powered by Gemini, configure settings, and test through CLI and web UI. If you've never touched ADK, start here.

The ambient expense agent codelab is the most complete demo in the set. You build a corporate expense agent using ADK 2.0's graph-based workflow API. Expenses under a threshold auto-approve in plain Python. Anything above passes through a pre-LLM security screen (PII redaction, prompt-injection defense), a Gemini compliance analysis, and pauses for human review before finalization. The agent mounts behind FastAPI, triggers from Pub/Sub events, and grades itself with an LLM-as-judge evaluation. This agent reappears in later demos.

The MCP codelab shows how to build reusable tools using the Model Context Protocol, an open standard. These tools let Gemini query BigQuery, search files, and call APIs. Because MCP is vendor-agnostic, the tools work across different frameworks.

The A2UI codelab tackles dynamic frontends. The agent renders actual interface components, layouts, charts, interactive menus, that update in real time as the conversation flows. The agent assembles the UI the user needs on the fly.

How do the scale demos handle production workloads?

Four demos address what happens after the prototype works on your laptop. The Stateful Data Science Agent codelab builds a BigQuery agent that remembers user preferences across sessions via Memory Bank, then deploys to Agent Runtime (formerly Agent Engine). Infrastructure, scaling, and session management are handled automatically.

Real enterprise workflows take days or weeks, not seconds. One tutorial covers three architectural patterns for long-running agents: durable state machines, event-driven idle time handling, and checkpoint-and-resume with persistent sessions. The example is an onboarding coordinator that survives container restarts and picks up exactly where it left off.

The Deploy to Agent Runtime codelab takes the expense agent from the build section to production. Scaffold your deployment config with Agents CLI, preview with a dry run, deploy live. Cloud Trace, Cloud Logging, and BigQuery Agent Analytics wire in automatically. The agent auto-registers in Agent Registry, making it discoverable across your org immediately.

The frontend codelab ties everything together. Build a manager dashboard on Cloud Run, connect it to Agent Runtime through an OIDC-authenticated Pub/Sub pipeline, and give managers the ability to resume paused human-in-the-loop sessions from the browser. This completes the end-to-end enterprise architecture.

What governance patterns are included?

The governance demos address access control, endpoint tracking, and traffic filtering at scale. The Secure Agentic Coding codelab demonstrates how to secure an agent's lifecycle from the first commit. While the source text cuts off here, Google's broader Agent Platform messaging emphasizes pre-deployment security screening, audit logging, and policy-based access controls.

For teams already using tools like n8n, Make, or Zapier for automation workflows, Google's Agent Platform represents a different approach. Those tools excel at connecting APIs through visual interfaces. ADK targets developers who want code-first control over agent behavior, state management, and LLM orchestration.

Why does Agents CLI matter for DevOps teams?

The Agents CLI integration shifts agent development from a separate workflow to something embedded in your existing coding environment. You describe intent in natural language, and the CLI handles scaffolding, evaluation setup, deployment configuration, and monitoring instrumentation.

This matters for teams that already use AI coding assistants. Instead of context-switching to a separate agent-building tool, the agent becomes another artifact your coding assistant manages. Google claims seven skills install automatically: likely covering project setup, tool definition, deployment config, testing, monitoring, registry publishing, and eval framework integration.

Which demo should you start with?

If you've never used ADK, start with the ADK Foundation codelab. It's explicit on-ramp material.

If you're evaluating Agent Platform for production use, skip to the ambient expense agent codelab. It covers the most ground: graph-based workflows, security screening, compliance checks, human-in-the-loop, event triggers, and LLM-based evaluation. You'll see what a complete agent looks like before committing to the full tutorial series.

For teams already running agents elsewhere and considering migration, the Deploy to Agent Runtime and frontend codelabs show the operational model: auto-instrumentation, registry publishing, and dashboard integration.

Frequently Asked Questions

What is the Gemini Enterprise Agent Platform?

A Google Cloud platform for building, scaling, governing, and optimizing AI agents. It includes the Agent Development Kit (ADK) for code-first agent construction and Agent Runtime for managed deployment.

Do I need to follow the demos in order?

No. Google designed them as standalone tutorials. Start with ADK Foundation if you're new, or jump to specific demos based on what you need to learn.

What coding assistants work with Agents CLI?

Google mentions Antigravity, Claude Code, and Codex by name. The CLI installs seven skills that make any compatible coding assistant an ADK expert.

How does Agent Platform handle long-running workflows?

Through three architectural patterns: durable state machines, event-driven idle time handling, and checkpoint-and-resume with persistent sessions. Agents survive container restarts.

Is MCP vendor-specific to Google?

No. Model Context Protocol is an open standard. Tools you build work across different vendors and frameworks.

Source: Cloud Blog