Most agent frameworks were built by startups racing for mindshare.
Then Google walked in and open-sourced its own β production requirements included.
A year later it's one of the fastest-growing agent frameworks on GitHub.
@GoogleOSS launched the Agent Development Kit (ADK) with one design goal: make building agents feel like software engineering, not prompt spaghetti.
20,000+ GitHub stars. 3,600 forks. Apache 2.0. ADK 2.0 hit General Availability in June 2026.
ADK is a code-first Python toolkit for building, evaluating, and deploying AI agents β from a single tool-calling bot to full multi-agent systems.
Here's what makes it different:
β Everything is plain Python β agents, tools, orchestration β so it's testable, versionable, and debuggable like normal code
β ADK 2.0's Workflow Runtime: a graph-based engine with routing, fan-out/fan-in, loops, retries, and human-in-the-loop
β Task API for structured agent-to-agent delegation instead of prompt hand-offs
β Model-agnostic: optimized for Gemini, but runs Claude, Ollama, and anything else via adapters
β Speaks the A2A protocol for remote agent-to-agent communication
β `adk web` gives you a built-in UI to test, debug, and showcase agents
β `adk eval` bakes evaluation into the workflow β measure agent quality, not vibes
β Compose specialized agents into hierarchies that scale from laptop to Cloud Run
β Now available in Python, Go, Java, and TypeScript
LangChain proved developers wanted agent frameworks. Google is betting they want one built like infrastructure.
Discovered on OSSphere :
ossphere.dev/google/adk-pythβ¦
What's your agent stack in 2026 β LangChain, CrewAI, ADK, or rolling your own? Drop it below π
#AIAgents #GoogleADK #Python #Gemini #AgentFrameworks #BuildInPublic #OpenSource