Gemini Enterprise Agent Platform is Google Cloud's managed platform for building with models. Everything a team needs once it has chosen a model sits there, meaning the catalogue it chose from, the tuning, the retrieval, the evaluation and the controls a business wants in place before anything reaches a customer. A product person meets the platform at the moment a prototype has to become something the business can run.
The sections below start with the rename, because it decides which name a correct answer uses, and then take Model Garden, Agent Platform AutoML, the two RAG offerings, custom agents and the two build environments in turn.
The rename from Vertex AI
Vertex AI is now Gemini Enterprise Agent Platform, usually shortened to Agent Platform, and Vertex AI Search is now Agent Search. Google updated the exam for the new names while a great deal of published study material still carries the old ones. A candidate therefore has to recognise both names and answer with the current one.
Model Garden
Model Garden is the catalogue. Google's own models, open models such as Gemma and third party models are all reachable through one platform under one set of controls.
The business value of one catalogue is comparison without commitment. A team can evaluate several models against its own task and change its choice later without rebuilding the software it wrote around the call. Model Garden is also the concrete thing behind Google's claim to an open approach, since third party and open models sit in the catalogue beside Google's own.
Agent Platform AutoML
Agent Platform AutoML trains a custom predictive model from an organisation's own labelled data, with no architecture to choose and no training code to write. The team supplies the examples and the service produces the model.
AutoML belongs to classical machine learning, and the exam includes it because the right answer to a business problem is often a prediction. Forecasting demand, scoring leads and classifying records are all work of that kind. A question describing labelled historical data and a number or a category to predict is pointing at AutoML, and Gemini is the wrong answer to it.
The two RAG offerings
Google Cloud offers retrieval augmented generation in two forms, and the difference between them is how much of the pipeline a team assembles.
Prebuilt RAG with Agent Search is the managed form. A team points it at the content and the service handles chunking, embedding, indexing and retrieval. This is configuration, and it is the right default for most projects.
The RAG APIs expose the stages so a team can control each one, which means its own chunking strategy, its own embedding model, its own vector store and its own ranking. The extra work is worth it where retrieval quality is the product itself and the managed defaults have already been tried on the task.
The order matters. Teams routinely build a retrieval pipeline by hand and discover afterwards that the managed one would have done the job.
Custom agents on Agent Platform
Agent Platform is also where agents are built. An agent is given a goal, a set of tools it may use and instructions on how to behave, and it then plans a sequence of steps and carries them out.
The business value is work that has steps in it. Finding an order, checking the policy, issuing the refund and writing to the customer is one piece of work with four stages, and an agent can carry all four.
The cost is that the security questions change. An agent that can act on systems means a prompt injection now reaches something that changes data, so what the agent is permitted to do matters more than what it is told.
Agent Studio and Google AI Studio
Two build environments carry similar names and serve different audiences.
Google AI Studio is the fast way to try the Gemini API. A developer prompts in a browser, takes the generated code away and has a prototype the same afternoon.
Agent Studio sits within Agent Platform, and it is where an agent is designed, given its tools, tested and deployed under enterprise controls.
A question about prototyping points at Google AI Studio. A question about something that will run inside a business points at Agent Studio.