Google just made every coding assistant smarter at building agents than most engineers who've been doing it for years.
The Summary
- Google released agents-cli, an open-source toolkit that teaches any AI coding assistant how to build, evaluate, and deploy production-grade agents on Google Cloud
- Works with Claude Code, GitHub Copilot, and any other coding agent — give them skills once, get agent expertise forever
- Includes eight specialized skills covering the full lifecycle: scaffolding, ADK Python APIs, evaluation frameworks, deployment to Cloud Run/GKE, observability, and Gemini Enterprise registration
The Signal
Google just shipped the meta-tool. Not an agent builder. Not another framework. A system that makes AI coding assistants fluent in building other AI agents.
Here's what matters: agents-cli packages Google's Agent Development Kit (ADK) knowledge into modular "skills" that any coding assistant can absorb. Install it once with `npx skills add google/agents-cli` and your Claude or Copilot suddenly knows how to scaffold agent projects, wire up tools and callbacks, write evaluation harnesses with LLM-as-judge patterns, and deploy to production infrastructure. No more copy-pasting documentation. No more "how do I structure this callback handler?" The coding assistant just knows.
"Turn your favorite coding assistant into an expert at building and deploying agents on Google Cloud."
The eight skills break down like this:
- Workflow and code preservation: Development lifecycle rules, model selection logic
- ADK Python API: Core agent primitives, orchestration patterns, state management
- Scaffold: Project creation and enhancement commands
- Evaluation: Metrics, datasets, adaptive rubrics, LLM-as-judge methodology
- Deploy: Cloud Run, GKE, CI/CD pipelines, secrets management
- Publish: Gemini Enterprise Agent Platform registration
- Observability: Cloud Trace integration, logging, third-party monitoring
This is Google making a bet that the bottleneck in agent development isn't ideas or infrastructure — it's the gap between "I want to build an agent" and "I know how to structure callbacks in this specific framework." By encoding that knowledge as skills instead of docs, they're collapsing setup time from days to minutes.
The example in the repo is telling: "Build a caveman-style agent that compresses verbose text into terse, technical grunts." You tell your coding assistant that sentence, it spins up the scaffold, wires the ADK patterns, writes eval criteria for "caveman-ness," and preps deployment configs. All from a conversational prompt because it absorbed the skills.
What's notable is the portability. This isn't locked to Google's own AI. It works with Claude, Copilot, Antigravity CLI, any agent that can consume skill definitions. Google is distributing agent-building literacy across the entire coding assistant ecosystem. That's either brilliant platform strategy or an admission that the real moat isn't the tools — it's Cloud infrastructure at scale.
The observability and deployment modules reveal the real target: enterprises trying to move agents from prototype to production. Setting up Cloud Trace, managing secrets across environments, wiring CI/CD for agent updates — this is the unglamorous work that kills 80% of internal agent projects. Google just automated teaching your coding assistant how to handle it.
The Implication
If you're building agents on Google Cloud, install this today and let your coding assistant learn the ADK patterns instead of you. If you're on a different stack, watch how fast the "skills for coding assistants" pattern spreads. OpenAI, Anthropic, and AWS will ship equivalents within months.
The deeper play: Google is training a generation of developers to build agents conversationally. When "just tell Claude what you want" becomes the default workflow, the framework that Claude knows best wins the deployment volume. That's platform strategy disguised as open source.