Google ADK: deploy agents to Vertex AI Agent Engine

This tutorial follows a Google ADK weather agent through local testing and deployment to Vertex AI Agent Engine. The sample tool returns hard-coded weather for New York, so the exercise focuses on configuration and deployment rather than live weather retrieval.

The presenter replaces Google API key configuration with Vertex AI settings, a Google Cloud project ID and a location, using us-central1 in the example. Setup includes installing the Google Cloud CLI and authenticating through a browser. An initial request in ADK Web fails because the project has not enabled the Vertex AI API. The presenter enables it and notes that the project also needs billing before retrying successfully. The local interface still uses Google Cloud services; this is not an offline inference setup.

For deployment, the presenter supplies the project, region, agent folder and an optional display name to the ADK deployment command. The demonstration shows automatic packaging and temporary files without a manually written Dockerfile. Deployment takes roughly three to four minutes in this example.

The final section tests the deployed AI agent in the cloud playground and points out runtime latency metrics and session records. It also distinguishes the project number in the resource URL from the project ID and recommends deleting test deployments to avoid additional charges.