Google Agents-CLI: Deploy AI Agents to Production in 10 Minutes

By Deepak Bagada, CEO at SaaSNext. I deployed 3 production agents using Google Agents-CLI in the first week after its July 8, 2026 launch, measuring setup time, iteration speed, and the gap between prototype and production.

Google released Agents-CLI on July 8, 2026, and it hit #1 on Product Hunt the next day. The premise is compelling: a command-line tool designed not for humans but for AI coding agents to use. Your coding agent (Claude Code, Codex, Antigravity) runs agents-cli to scaffold, evaluate, and deploy production AI agents on Google Cloud. The three commands — create, eval run, deploy — collapse what traditionally takes 2-3 weeks of production engineering into a single agent session. This guide walks through installing Agents-CLI, running each command, and the patterns that make it work for production workloads.

What Is Google Agents-CLI Agents-CLI is a command-line tool from Google that scaffolds, evaluates, and deploys production AI agents on Google Cloud. It is optimized for Google ADK (Agent Development Kit) but supports any model or framework. The tool is designed to be driven by AI coding agents in headless mode, meaning your coding agent can scaffold a project, run evaluations, iterate based on scores, and deploy without human intervention. The output of agents-cli create includes: agent code and tool definitions, unit tests, Dockerfile, Terraform infrastructure, Cloud Build CI/CD configuration, and OpenTelemetry observability setup. Every scaffolded project uses expert-crafted templates from Google's own production agent deployments. Agents-CLI is open source under the MIT license on GitHub.

Why Production Agent Deployment Is Hard Building a prototype AI agent takes an afternoon. The agent logic is usually simple: receive input, process with an LLM, maybe call a tool, return output. Production deployment is where the time disappears. You need authentication (service accounts, API keys), infrastructure (Docker, Terraform, Cloud Run), CI/CD (build, test, deploy pipeline), observability (logging, tracing, metrics), and integration (MCP servers, tool auth, secret management). Each requires reading separate documentation, configuring different tools, and manually stitching everything together. Agents-CLI eliminates this entirely by generating all production infrastructure from the first command. A coding agent running agents-cli create gets a project that is ready for production deployment, not just local development.

Quick Start: Install and Deploy in 10 Minutes Step 1: Install with one command: uvx google-agents-cli setup. This installs the CLI and its dependencies. Step 2: Ask your coding agent to build an agent. For example: Build an SRE agent that reads logs and drafts incident reports. Step 3: The coding agent runs agents-cli create sre-agent, which scaffolds the complete project with agent, tools, tests, Dockerfile, Terraform, CI/CD, and observability. Step 4: The coding agent runs agents-cli eval run to test the agent against a sample dataset and get a score. Step 5: The coding agent iterates on the agent based on eval scores, re-running eval until it passes the success threshold. Step 6: The coding agent runs agents-cli deploy to ship the agent to Vertex AI Agent Engine, Cloud Run, or GKE. Total time: under 10 minutes for the first deployment, under 5 minutes for subsequent iterations.

The Three Commands agents-cli create [agent-name] scaffolds a complete production agent project. The template includes agent code (ADK-based), tool definitions (MCP or function tools), unit tests (pytest), Dockerfile (multi-stage, production-optimized), Terraform (Cloud Run, VPC, IAM, secrets), CI/CD (Cloud Build, artifact registry, deploy), and observability (OpenTelemetry, Cloud Trace, Cloud Logging). agents-cli eval run evaluates the agent against a test dataset. The eval framework scores the agent on accuracy, latency, and cost. The coding agent uses these scores to iterate: improving prompts, adjusting tools, or optimizing model selection. The headless auto-optimization loop continues until scores meet the specified success criteria. agents-cli deploy ships the agent to production. Deploy targets include Vertex AI Agent Engine (managed agent runtime), Cloud Run (serverless containers), and GKE (Kubernetes). Deployment is zero-downtime with automatic rollback on failure.

Real-World Example: Building an SRE Agent I used Claude Code to drive Agents-CLI and build an SRE agent that reads Cloud Logging, identifies incident patterns, and drafts incident reports. The coding agent scaffolded the project in 30 seconds using agents-cli create sre-agent. It evaluated the initial agent: accuracy 62%. It iterated on the prompt and tool definitions, re-ran eval: accuracy 78%. Two more iterations: accuracy 89%. It deployed to Cloud Run with agents-cli deploy. Total time from blank terminal to deployed production agent: 18 minutes. The same agent built manually would require approximately 40 hours of infrastructure work.

Limitations Agents-CLI is optimized for Google Cloud. Deploying to AWS, Azure, or on-premises requires manual template adaptation. The tool currently optimizes for Google ADK; using LangGraph, CrewAI, or other frameworks requires customizing the scaffolding templates. The headless auto-optimization loop can overfit to eval datasets if the eval data lacks diversity. You are responsible for ensuring eval datasets represent real-world conditions. As an early release (July 8, 2026), community templates and third-party integrations are still developing.

FAQ Q: How much does Agents-CLI cost? A: Agents-CLI is free and open source under MIT license. You pay only for Google Cloud infrastructure (Agent Runtime, Cloud Run, or GKE) and model inference costs. Q: Does Agents-CLI work with non-Google models? A: Yes. The default is Google ADK with Gemini, but you can customize templates for any model provider. Q: Can my coding agent use Agents-CLI autonomously? A: Yes. Headless mode is the primary design. Your coding agent runs all three commands without human intervention. Q: What happens when the agent fails eval? A: The coding agent iterates automatically based on eval scores until success criteria are met or iteration limits are reached. Q: How long does Agents-CLI take to set up? A: One command: uvx google-agents-cli setup. Under 2 minutes.