Sitemap
3 min readAug 23, 2025

--

ADK-Python v1.12.0: Streamlining Agent Creation and Tooling

The Agent Development Kit for Python (ADK‑Python) got a solid boost with its v1.12.0 release, offering new features, CLI enhancements, experimental tooling support, and useful refinements. Here’s what’s new:

  1. YAML-First Agent Definition

The star of this release is the ability to define agents using YAML configuration files, complementing traditional Python-based setup. This new feature makes agent specification more declarative and accessible to non-developers or quicker iterations. Support includes:

• Defining agents via YAML config files

• Deploying these config-based agents directly to the Agent Engine via the CLI

Read the release announcement for details.

2. Bigtable Toolset (Experimental)

A dedicated Bigtable toolset has been introduced to streamline interaction with Bigtable when building AI agent applications. Designed as an experimental feature, it’s aimed at simplifying agent workflows that involve large-scale data.

3. Custom Tool Name Prefixes

When generating Google API toolsets automatically, you now can customize the tool_name_prefix — helpful for naming clarity and standardization. The oauth_calendar_agent example illustrates this customization in action.

4. New build_image Option in CLI

Deploying agents to Cloud Run is more flexible now thanks to a new build_image flag in the adk deploy cloud_run command — allowing you to specify or override the container image build process.

5. setdefault() for ADK State

The ADK State object now supports a convenient .setdefault() method, making state management more Pythonic and user-friendly — ideal for initializing or retrieving state variables efficiently.

Bug Fixes and Improvements Summary

Bug Fix Highlights

• Lazy loading for VertexAiCodeExecutor and ContainerCodeExecutor — improved performance and reduced startup overhead.

• Several fixes targeting the A2A (Agent-to-Agent) demo, packaging logic, event ID merging, and path handling.

• Resilience enhancements: avoids crashes when optional token counts are absent, corrects version comparisons in the CLI, adds missing OAuth scopes (e.g., Spanner admin), and smoothes out typing and display glitches.

Quality-of-Life Improvements

• Suppress experimental feature warnings with the new ADK_SUPPRESS_EXPERIMENTAL_FEATURE_WARNINGS environment variable.

• Agent config schemas now use pydantic.Field, so the generated JSON schema includes descriptive metadata for each config field.

• Updated dependency on openai to reflect an upstream change.

• Added license headers and other minor housekeeping across the codebase.

Community Buzz

On X, developer Ivan Nardini shares a succinct summary: “ADK 1.12.0 lets you author agents with simple and declarative YAML files, adds a new Bigtable toolset for massive datasets, and improved …”

Why It Matters

• Faster prototyping: YAML-based agent config opens doors for low-code workflows, rapid iteration, and easier collaboration across teams.

• Scalable tooling: The Bigtable toolset hints at growing support for more integrated, large-scale data operations.

• Better developer ergonomics: CLI tweaks, state helpers, and suppression flags streamline workflows and reduce friction.

• Stability and clarity: Under-the-hood fixes and metadata improvements boost robustness, documentation clarity, and maintainability.

Example: YAML Agent Config (Hypothetical)

name: “search_agent”

model: “gemini-2.0-flash”

instruction: “Act as a helpful assistant using Google Search.”

description: “Search-based assistant”

tools:

. — google_search

Combined with a CLI command like:

adk deploy cloud_run — config agent.yaml — build_image my-image:tag

this workflow dramatically cuts down on boilerplate and accelerates deployment paths.

In Summary

ADK-Python v1.12.0 — released Aug 21, 2025 — offers:

• YAML-based agent authoring with CLI support

• Experimental Bigtable toolset

• Customizable tool naming

• build_image option in Cloud Run deployments

• Convenient state defaults via .setdefault()

• Numerous fixes and developer improvements

• Ability to suppress experimental warnings & enhanced schema documentation

--

--

Ali Arsanjani
Ali Arsanjani

Written by Ali Arsanjani

Director Google, AI | EX: WW Tech Leader, Chief Principal AI/ML Solution Architect, AWS | IBM Distinguished Engineer and CTO Analytics & ML