Knowing the products is how a leader turns ambition into an architecture. Google Cloud offers a layered stack -- Gemini models, Model Garden, Vertex AI Studio, and the managed Vertex AI platform. Choosing the wrong layer wastes money and time: building a custom model when a prompt would do, or hand-rolling MLOps that Vertex AI already provides. The exam tests whether you can map a business need to the right Google Cloud offering.

Vertex AI is Google Cloud's unified platform for building, deploying, and managing ML and generative AI. Gemini is Google's family of multimodal foundation models available through Vertex AI, with variants tuned for different speed, cost, and capability trade-offs. Model Garden is a catalog where you discover and deploy models -- Google's own, open, and third-party. Vertex AI Studio is a console for prompt design, testing, and tuning without heavy code.

The key leadership decision is managed versus custom. Prefer a managed foundation model accessed by API when a general model plus good prompting or grounding solves the task -- this is fastest and cheapest. Move toward tuning or a custom model only when you have domain data and quality gaps that prompting cannot close.

A bank wants an internal assistant over its policy documents. A leader maps this to the stack: use a Gemini model on Vertex AI, add grounding against the policy corpus rather than training a new model. They browse Model Garden to compare options, prototype prompts in Vertex AI Studio, then deploy through Vertex AI with its security, logging, and quota controls. No custom training is needed -- the managed path meets the goal in weeks, not months. Only if grounding still misses bank-specific tone would they consider tuning.

This stack is where fundamentals become real: tokens map to Vertex AI pricing, embeddings power grounding, and Agent Builder (next topic) rides on Vertex AI. Responsible AI controls -- safety filters, logging, IAM -- are built into the managed platform, which is a major reason to prefer managed offerings.