Gen AI Leader
Deep Dive
A language model on its own only generates text. For an agent to actually do something — query a system, execute code, read a file — it needs tooling: concrete mechanisms for interacting with the external environment. This block covers the types of tooling, the catalog of Google Cloud services available as tools, and when to use each of Google's two "Studios."
Contents
An agent combines a model with different types of tools to move from "answering a question" to "completing a task" in the real world.
Extensions
Connect the agent to an external API in a standardized way, encapsulating how to call it. Example: an extension that connects the agent to a flight-booking system's API to check availability in real time.
Functions
Through function calling, the model decides when and with what parameters to invoke a function defined by the developer, but it is the developer's own system that executes it. Example: a function calculate_tax(amount, region) that the model invokes when it detects it needs it.
Data stores
Give the agent access to your own information through retrieval (typically RAG), instead of relying only on the knowledge the model was trained with. Example: a data store with the company's HR policy manual, queried by the agent when answering employee questions.
Plugins
Reusable, installable functionality packages that extend what the agent can do without rebuilding the integration from scratch. Example: a plugin that adds the ability to generate and send PDF reports directly from the conversation with the agent.
Watch out for this on the exam
The most commonly tested distinction is extensions vs. functions: in an extension, the agent/platform itself executes the call to the external API; in function calling, the model only generates the instruction of which function to call and with what arguments — the actual execution happens in the developer's system, outside the model's direct control.
Here's what a complete call looks like in action: the round-trip cycle that occurs every time the agent decides to use a function to resolve a user's question.
Google Cloud offers both general-purpose compute infrastructure and prebuilt, specialized AI APIs, which an agent can use as tools to interact with the environment.
Cloud Storage
Stores and retrieves files (documents, images) that the agent reads or generates.
Cloud Functions
Runs lightweight custom logic, invoked by the agent via function calling.
Cloud Run
Runs containers or more complex services that the agent invokes as the backend of a tool.
Speech-to-Text API
Converts spoken audio into text, for agents that interact by voice.
Text-to-Speech API
Converts text into spoken audio, for the agent's voice responses.
Translation API
Translates text between languages within the agent's flow.
Document AI API
Extracts structured data from documents (invoices, contracts, forms).
Cloud Vision API
Analyzes images: object detection, OCR, and content labeling.
Cloud Video Intelligence API
Analyzes video content: scene detection, objects, and text in video.
Natural Language API
Analyzes text to extract entities, sentiment, and syntax without needing a full LLM.
The table below also includes services without a dedicated icon (databases, Agent Platform, Document Translation API, Google Cloud API Library) to have the complete catalog in one place.
| Service / API | What it does as an agent tool |
|---|---|
| Cloud Storage | Stores and retrieves files (documents, images) that the agent reads or generates |
| Databases (Cloud SQL, Firestore, BigQuery...) | Query and update the application's or business's own structured data |
| Cloud Functions | Runs lightweight, serverless custom logic, invoked by the agent via function calling |
| Cloud Run | Runs more complex containers/services that the agent invokes as the backend of a tool |
| Agent Platform | Orchestrates the agent itself: models, data stores, and tools in a single development environment |
| Speech-to-Text API | Converts spoken audio into text, for agents that interact by voice |
| Text-to-Speech API | Converts text into spoken audio, for the agent's voice responses |
| Translation API | Translates text between languages within the agent's flow |
| Document Translation API | Translates entire documents while preserving formatting, not just plain text |
| Document AI API | Extracts structured data from documents (invoices, contracts, forms) |
| Cloud Vision API | Analyzes images: object detection, OCR, content labeling |
| Cloud Video Intelligence API | Analyzes video content: scene detection, objects, and text in video |
| Natural Language API | Analyzes text to extract entities, sentiment, and syntax without needing a full LLM |
| Google Cloud API Library | Full catalog of Google Cloud APIs available to discover and enable as additional tools |
How to memorize this table for the exam
Group mentally: generic compute/storage (Cloud Storage, databases, Cloud Functions, Cloud Run) vs. specialized AI APIs by modality (voice: Speech-to-Text/Text-to-Speech; text: Translation, Natural Language; documents: Document AI, Document Translation; vision: Cloud Vision, Video Intelligence).
Both tools let you experiment with Google's generative AI, but they have different purposes and audiences: Agent Studio is oriented toward building agents, while Google AI Studio / Vertex AI Studio is oriented toward experimenting directly with models and prompts.
Agent Studio
A low-code/no-code environment for designing, configuring, and deploying full agents: defining instructions, connecting data stores, adding tools (extensions, functions), and orchestrating the agent's multi-step behavior.
Google AI Studio
A rapid experimentation environment for working directly with Gemini models: testing prompts, adjusting inference parameters, comparing model versions, and getting sample code — without necessarily building a full agent.
Practical rule for deciding
If the question is "does this model do what I need with this prompt?" — use Google AI Studio. If the question is "I need to assemble a system with tools, my own data, and multi-step logic" — use Agent Studio. AI Studio is the starting point for exploring the model; Agent Studio is where the final agent gets built.
Did you get this topic?
Test what you just learned
A Google Cloud developer is building a support agent that needs to: (1) invoke an internal function to check an order's status, and (2) transcribe customer voice messages before processing them. Which combination of tooling and API corresponds to each need?
Put what you learned to the test with original Gen AI Leader exam questions.
Start Gen AI Leader mock examUnlimited mock exams with the Pro plan
Log in to track your progress.