Google Releases Agent Development Kit (ADK) 2.0

What changed

Google has released version 2.0 of its Agent Development Kit (ADK) for Python, an open-source, code-first framework designed for building, evaluating, and deploying AI agents. This latest iteration emphasizes flexibility and control, applying software development principles to agent creation. Key features include a graph-based Workflow Runtime for composing deterministic execution flows with support for routing, loops, state management, and nested workflows. The Task API facilitates structured agent-to-agent delegation with multi-turn capabilities and human-in-the-loop options. ADK 2.0 supports modular multi-agent systems, allowing for the design of scalable applications by composing specialized agents into flexible hierarchies. It also boasts a rich tool ecosystem, enabling integration with pre-built tools, custom functions, OpenAPI specs, and MCP tools, with tight integration into the Google ecosystem. The framework prioritizes a code-first development approach for ultimate flexibility, testability, and versioning, though an Agent Config feature allows for code-free agent building. A tool confirmation flow provides an additional layer of security by guarding tool execution with explicit confirmation. ADK 2.0 is designed for deployment flexibility, supporting containerization for local deployment or scaling with Google Cloud services like Cloud Run and Vertex AI Agent Engine. The release cadence is approximately bi-weekly for stable releases. Developers can install the latest stable version via pip with pip install google-adk, requiring Python 3.10+. Optional integrations can be installed with pip install "google-adk[extensions]". For those needing the latest features before official release, installation from the main branch via git is possible: pip install git+https://github.com/google/adk-python.git@main.

Why it matters for builders

The ADK 2.0 empowers AI builders by offering a comprehensive toolkit that bridges the gap between AI capabilities and robust software engineering practices. Its modular design and code-first philosophy allow for greater control, testability, and versioning of agent logic and orchestration. The inclusion of advanced features like the Workflow Runtime and Task API simplifies the development of complex, multi-agent systems, enabling developers to build more sophisticated and reliable AI applications.

Practical impact

Developers can leverage ADK 2.0 to construct and deploy AI agents with greater ease and efficiency. The framework's model-agnostic and deployment-agnostic nature, coupled with its compatibility with other frameworks, offers significant flexibility. Builders can start by installing the ADK via pip and exploring the provided documentation and samples. For immediate testing, the adk run and adk web commands offer interactive CLI and web UI options, respectively. Evaluation tools are also bundled for assessing agent performance. Deployment options include local Docker containers and Google Cloud services, providing clear pathways for production readiness.

Caveats and source limits

The provided source material details the features and installation of ADK 2.0 but does not include specific benchmark results comparing its performance against other agent frameworks. Pricing information for enterprise-level deployment or advanced features is also not specified. While the documentation is extensive, the exact scope and limitations of the "Agent Config" feature for code-free development are not fully elaborated. The source also mentions community events and a community repo, but specific details on community contributions or engagement levels are limited.