- Agent CLI simplifies the entire AI agent development lifecycle, from building to deployment.
- ADK 2.0 introduces graph-based workflows for deterministic, reliable agent execution.
- New long-running, ambient, and resume agents enable persistent state and human-in-the-loop processes.
Google Cloud is fundamentally changing how developers build and manage AI agents with the introduction of Agent CLI and significant enhancements to the Agent Development Kit (ADK 2.0), pushing AI from experimental demos to robust, production-ready systems.
At the forefront of these innovations is the new Agent CLI, serving as the universal entry point for the Gemini Enterprise Agent Platform. This powerful command-line interface integrates a suite of skills and commands, empowering developers to build, scale, govern, and optimize the entire agent development lifecycle. It seamlessly integrates with various coding agents like Gemini CLI, dot code, and Codex, providing a unified context and the necessary skills to scaffold, build, and deploy agents with unprecedented ease. The concept of 'skills' is central to this advancement, making agents smarter by providing them with distinct capabilities that can be executed at runtime, moving beyond cumbersome, monolithic prompts.
The Agent Development Kit 2.0 introduces a critical feature: graph-based workflows. This innovation addresses the long-standing challenge of non-determinism in AI agents, offering precise control over routing and task execution. This newfound reliability is particularly vital for production environments in sectors like financial services and insurance, where predictable outcomes are paramount. Furthermore, Google Cloud has unveiled long-running AI agents, capable of maintaining state for up to seven days, a significant leap from previous limitations. This persistence is complemented by ambient agents, which can trigger based on events, schedules, or prompts, and resume agents, ensuring workflows can pause and restart seamlessly, even after interruptions, making human-in-the-loop processes more practical and efficient.
To further accelerate development, Google Cloud has launched Agent Garden, a comprehensive library of pre-built templates and multi-agent patterns. This resource provides developers with expert-designed blueprints for common use cases, including sequential, loop, and parallel agent workflows. The vision is clear: enable developers to build agents with agents, fostering a paradigm where complex AI systems are assembled and managed through natural language commands to the Agent CLI. This holistic approach, from flexible development tools to robust runtime capabilities, marks a pivotal shift towards building intelligent, reliable, and scalable AI solutions that can solve real-world problems for days, not just minutes.
Shubham Saboo, an AI Product Manager at Google Cloud, emphasized the transformative nature of these releases, stating, "We are moving from demo to something that is production that can run in production reliably and can really solve real problems." This sentiment underscores Google Cloud's commitment to providing developers with the tools needed to transition their AI projects from conceptual stages to fully operational, enterprise-grade applications.
“We are moving from demo to something that is production that can run in production reliably and can really solve real problems.”
- Shubham Saboo, AI Product Manager
