The next step after Karpathy’s agentic engineering

agents-cli is a powerful tool that enhances coding agents for building, evaluating, and deploying agents on Google Cloud. It provides a CLI interface and skills that make it easier for coding agents to perform these tasks. While agents-cli is not a coding agent itself, it works seamlessly with popular coding agents such as Antigravity CLI, Claude Code, Codex, and any other coding agent.

To get started with agents-cli, users need to have Python 3.11+, uv, and Node.js installed. The tool works with any coding agent, allowing users to build agents that can compress verbose text into terse technical grunts. The Google Cloud agent stack that agents-cli builds on includes various Google Cloud services, which are used for deploying and running agents.

Although agents-cli is not a replacement for Antigravity CLI, Claude Code, or Codex, it provides a more comprehensive set of tools for building, evaluating, and deploying agents on Google Cloud. Unlike using ADK directly, agents-cli offers additional skills and commands that streamline the agent-building process. Users can also integrate agents-cli with their existing agent projects, enhancing their capabilities.

The tool can be used without a coding agent as well. It supports standalone use, with standalone commands that can be executed directly from the terminal. However, the skills provided by agents-cli make it more convenient for coding agents to utilize the tool. Users can extend agents-cli with other skill suites if needed, such as agent-skills for general software-engineering workflows and google/skills for Google Cloud foundations.

agents-cli emits usage signals for each command invocation to help improve the tool and understand how it's being used. These signals include metadata such as command name, exit code, error class, duration, CLI/Python versions, and OS. However, the tool does not record file contents, arguments, or other input to maintain user privacy. Users can opt out of sending telemetry by setting specific environment variables.

When deploying agents with agents-cli, users are responsible for the resources created in their Google Cloud project. They should review the Google Cloud Service Terms for any applicable terms and conditions. Overall, agents-cli is a valuable tool for turning any coding assistant into an expert at creating, evaluating, and deploying AI agents on Google Cloud, offering a wide range of features and customization options.