Karpathy’s Full Agentic Engineering Lifecycle using Google’s Agents-CLI

Google's Agents-CLI is a powerful tool for building, scaling, governing, and optimizing enterprise-grade AI agents on Google Cloud. This command-line interface, or CLI, provides a suite of skills and commands that enhance any coding assistant, making it an expert in creating, evaluating, and deploying agents.

To use the Agents-CLI, you'll need Python 3.11 or later, uv, and Node.js installed on your system. Once these prerequisites are met, you can launch Antigravity CLI, Claude Code, Codex, or any other coding agent of your choice. With the Agents-CLI, you can instruct your coding agent to build various types of agents, such as a caveman-style agent designed to compress verbose text into terse, technical grunts.

It's important to note that the Agents-CLI is not an AI agent itself, but rather a tool that provides the necessary commands and skills to improve your coding agent's capabilities. The CLI works seamlessly with other popular coding assistants, including Antigravity CLI, Claude Code, Codex, and any other coding agent you prefer.

Unlike the ADK (AI Development Kit) framework, which is an agent framework, the Agents-CLI offers the tools to build, evaluate, and deploy ADK agents end-to-end. This means that if you're already using ADK for your agent development, the Agents-CLI can streamline your workflow and provide additional features for deployment and cloud integration.

The Agents-CLI is compatible with existing agent projects. By using the "scaffold" command, you can enhance your current project with deployment and continuous integration/continuous deployment (CI/CD) capabilities. Even if you don't have a coding agent, the Agents-CLI can be used standalone, allowing you to run various commands directly from your terminal.

The skill set of the Agents-CLI covers the entire agent-building lifecycle, including scaffolding, ADK code patterns, evaluations, deployment, publishing, and observability. For additional concerns outside the agent-building scope, you can install other skill suites alongside the Agents-CLI. For example, the "agent-skills" suite focuses on general software engineering workflows like ideation, specification gates, planning, and code review, while the "google/skills" suite covers essential Google Cloud foundations such as BigQuery, Cloud Run, Firebase, and GKE.

When deploying agents using the Agents-CLI, you'll be creating resources in your own Google Cloud project. It's essential to review the Google Cloud Service Terms to understand the responsibilities and implications associated with deploying agents on Google Cloud.