At Google Cloud Next '26, Google folded its AI platform pieces together, renaming Vertex AI to the Gemini Enterprise Agent Platform and absorbing the Agentspace product into it, and alongside that rebrand it shipped agents-cli, a command-line package aimed specifically at coding agents rather than human developers. The pitch is that tools like Claude Code, Gemini CLI, or Cursor can use agents-cli as a set of skills that turn them into competent operators of Google's Agent Development Kit (ADK) and the broader Google Cloud agent stack, rather than requiring a human to hand-hold the deployment process. Concretely, it provides scaffolding templates for standard ADK agents, agents that speak the Agent-to-Agent (A2A) protocol, and agentic RAG setups, so a coding agent can generate a working skeleton instead of starting from a blank file. It also wires in an LLM-as-judge evaluation loop: a coding agent can generate eval sets, run them against the agent under development, look at what failed, and iterate automatically. On the security side, agents-cli integrates with Google's Agent Gateway and Model Armor, an inline layer that intercepts calls between agents and tools to enforce access control and screen for prompt injection and data leakage - a direct response to the kind of cross-agent attack surface that has been a growing concern this year. Deployment targets Cloud Run, letting teams containerize an A2A agent and get Google's serverless scaling without managing the underlying infrastructure themselves. Google also disclosed that A2A, the interoperability protocol it launched with over 50 partners, has reached roughly 150 organizations running it in production rather than pilots, suggesting cross-vendor agent communication is moving from proof-of-concept to something teams are actually shipping. For developers, the practical takeaway is that Google is explicitly designing tooling for AI agents as the primary user rather than a human typing commands, a notable shift in who a CLI is meant to serve.