Google Cloud Agents CLI: From Prototype to Production in One Command

I have spent more time than I care to admit wiring together infrastructure for AI agents. The model works. The logic is sound. But getting from a local Python script to a production service that scales, has proper CI/CD, and passes security review is still a multi-week slog. The problem is not the agent itself. It is the gap between the code and the cloud.

Google Cloud just closed that gap with the Agents CLI. It is not another generic CLI wrapper. It is a dedicated programmatic backbone for the entire Agent Development Lifecycle, built specifically for the tools we already use in our terminals.

What It Actually Does

The Agents CLI lives inside the new Agent Platform on Google Cloud. Its job is to turn your coding agent from a smart autocomplete into a full-stack deployment engine. The CLI exposes the entire Google Cloud agent stack - Agent Platform, Cloud Run, A2A Integration, and more - in a format that your agent can read, understand, and act on without hallucinating its way through YAML documentation.

There are two modes. Agent Mode is where the CLI shines. Your coding agent can consume the CLI's bundled skills and scaffold entire projects with automatic defaults. Human Mode gives you the same commands in your own terminal when you want deterministic control.

The Three Phases: Build, Evaluate, Ship

The CLI breaks the lifecycle into three phases. Each phase maps to real commands.

1. Build: Scaffolding with Skills

The first command installs the CLI and injects bundled skills into your coding environment:

uvx google-agents-cli setup

From there, your agent can scaffold a complete project. Want a travel expense agent that auto-approves expenses under $50 and routes anything suspicious to a human? Tell your agent in plain English. The CLI handles the defaults:

agents-cli create finance-agent -y --deployment-target agent_runtime
cd finance-agent

The CLI provides the exact API references, sensory input, and project structure your agent needs to build something functional immediately. No more guessing which Cloud Run revision format is current.

2. Evaluate: Local Simulation and Rigorous Testing

Building the logic is half the battle. The other half is proving it works. The Agents CLI has built-in evaluation commands:

# Run evaluations against ground-truth datasets
agents-cli eval run

# Compare trajectory scoring across two runs
agents-cli eval compare evals/run_v1.json evals/run_v2.json

Your coding agent can orchestrate these tests natively, validate data retrieval, and benchmark different configurations without leaving the terminal.

3. Ship: Production Deployment in Minutes

The deployment phase is where most projects die. The Agents CLI provisions Infrastructure as Code, sets up CI/CD pipelines, and deploys directly to Agent Runtime or Cloud Run:

# Provision production infrastructure
agents-cli infra single-project

# Ship the agent to Google Cloud
agents-cli deploy

# Register with Gemini Enterprise for distribution
agents-cli publish gemini-enterprise

What used to take a 70-day enterprise deployment cycle now happens in one CLI session.

The Bigger Picture

Google is not just shipping a CLI here. It is shipping an opinion about how agent infrastructure should work. The Agents CLI is the missing layer between your coding agent and Google Cloud. Other platforms are still asking you to paste service account JSON into a chat window. Google is giving your agent the keys programmatically.

This also pairs nicely with their other recent moves. The Gemini CLI rebranded to Antigravity CLI. The AI Edge Gallery now speaks MCP. Google is building the full stack, from local agent to cloud deployment, with standardized interfaces between every layer.

Why This Matters

If you are building agents that need to run in production, not just in a Jupyter notebook, the infrastructure gap is your biggest bottleneck. The Agents CLI turns that bottleneck into a single command. That is not a marginal improvement. It is a change in what kind of agent projects are worth building.

I will be testing this on a real Cloud Run deployment this week. The fact that my agent can now provision its own infrastructure, evaluate its own accuracy, and ship itself to production is a shift in how I think about agent architecture.

Source: Google Developers Blog