Integrating other coding agents
While Antigravity is our recommended agentic coding solution, Google Cloud is designed to work well with any coding agent you choose. Our platform is open, and we provide tools to ensure flexibility:
-
The Agent CLI and Agent Development Kit (ADK) allow you to build and interact with agents from various sources, including tools like Claude Code. This means developers can often keep their preferred interfaces while running the underlying AI inference on Google Cloud. This approach ensures your workflows benefit from Google Cloud’s security, compliance, and infrastructure.
-
Our Skills for Google products, launched at Next, are designed to be compatible with multiple coding tools, enabling you to enhance different agents with a consistent set of capabilities.
This flexibility allows teams to integrate their existing favorite tools and models, ensuring seamless and compliant operation within their established workflows.
Rung four: Agent Development Kit (ADK 2.0)
Code-first, low floor, high ceiling. If Managed Agents are configuration-first, ADK is engineering-first. This is for software engineers who want to build custom agent meshes from the ground up – any architecture, any model, unconstrained.
ADK enhancements launched at Google Cloud Next are now available for everyone. It introduces a unified graph-based engine that gives you a slider from dynamic, model-led reasoning to strict, deterministic workflows. The framework handles the heavy lifting of multi-agent coordination, managing how sub-agents, tools, and data pass between one another.
-
Collaborative workflows (Python v2.0.0): Previously called the Task-based Agent Collaboration API, this is how you build self-managing agent teams. A coordinator delegates to subagents using explicit operating modes:
- chat: Full user interaction, manual return to parent, this is “handoff conversation to sub-agents”.
- task: User interaction for clarifications, automatic return to parent, this is a new “collaborate for this assignment” which is the best of both other options.
- single-turn: No user interaction, parallel execution, automatic return, this is “agent as tool”.
-
Dynamic workflows: Dynamic workflows in ADK allow you to put aside graph-based path structures and use the full power of your chosen programming language to build workflows. With Dynamic workflows, you can create workflows with simple decorators, invoke workflow nodes as functions, and build complex routing logic.
-
ADK Kotlin (Beta): “ADK for Android.” Kotlin support joins Python, Go, and Java, increasing language coverage so your on-device mobile agents can seamlessly coordinate with your backend Python agents.
Finally, the Agents CLI packages Google’s expert skills for ADK, eval, deploy, observability, and publishing – turning any AI coding agent (like Antigravity, Gemini CLI, Claude Code, or Cursor) into an expert at agent app building as well as agent ops. It gives your AI Agent skills to understand the Google Cloud agent stack, turning an expansive ecosystem into a seamless assembly line for developers hillclimbing their agent builds.
