In this video I test Google ADK’s new Visual Agent Builder, which basically lets AI create your AI agents. Google ADK (Agent Development Kit) is an open-source framework for building agents in Python, Go, and Java—but the big update here is you can now build multi-agent workflows with a visual canvas and an AI assistant (Gemini) on the right side. You just describe what you want in plain English, and it generates the agent graph for you—no coding required. If you still want control, you can drag-and-drop components like sequential agents, loop agents, parallel agents, tools, and callbacks (guardrails). I walk through setup end-to-end: pip install google-adk, then run adk web to launch the local UI. I also show how authentication works behind the scenes (either Vertex AI project auth via gcloud or an AI Studio key), because that’s what powers the Gemini assistant. Then I build a “deep researcher” style agent that uses built-in Google Search tooling, runs a loop with query refinement, and produces a final report with references. My favorite part is the transparency: you can see agent-to-agent/tool interactions and metadata live, and the builder generates YAML files so you can take it into CI/CD or share it with your team. Overall, I’m bullish on ADK for production-grade agent work, even though the UI still has some missing pieces like easier custom tool add-ons without extra plumbing.