Google Cloud's gen AI offerings · Topic 6 of 6
Agent Tooling, Agent Studio and Google AI Studio
How gen AI agents use extensions, functions, data stores and plugins, which Google Cloud APIs act as tools, and Agent Studio versus Google AI Studio.
- 4 min
- Agents use tools to act: extensions, functions, data stores, plugins.
- Functions: the model proposes the call; the application executes it.
- Tools include Cloud Storage, databases, Cloud Run functions, Cloud Run, Agent Platform, and AI APIs for speech, language, documents, images and video.
What it is
A model on its own can only produce text or other content. An agent becomes useful when it can use tools to interact with the outside world: look up data, call a system, take an action. The exam names four kinds of tooling (extensions, functions, data stores, plugins), lists Google Cloud services and pre-built AI APIs that agents use as tools, and asks when to use Agent Studio versus Google AI Studio.
How it works
How agents use tools
The agent's model reads the request, decides which tool can help, prepares the input for that tool, receives the result, and continues until the goal is met. The four kinds of tooling:
| Tool type | What it is | Example |
|---|---|---|
| Extensions | Connectors that let the agent call an external API directly | Calling a weather service or a booking system |
| Functions | The model selects a function and produces its arguments; the application runs the function and returns the result (often called function calling) | The model asks to run check_order_status(order_id); the app runs it securely |
| Data stores | Sources of information the agent can search to ground its answers | A store of policy documents or product data |
| Plugins | Packaged add-ons that give an agent a new capability | A plugin that adds calendar scheduling |
With functions, the application keeps control of execution: the model only proposes the call. That makes functions a good fit when actions must pass through the organization's own checks.
Google Cloud services and APIs as tools
| Category | Services and APIs | What the agent can do with them |
|---|---|---|
| Storage and data | Cloud Storage, databases (such as Cloud SQL, Firestore, BigQuery) | Read and write files and records |
| Run custom code | Cloud Run functions (previously Cloud Functions), Cloud Run | Execute business logic as a tool |
| AI platform | Gemini Enterprise Agent Platform | Call models, search, and manage agents |
| Speech | Speech-to-Text API, Text-to-Speech API | Listen and talk |
| Language | Translation API, Document Translation API, Natural Language API | Translate text and documents, analyze sentiment and entities |
| Documents and vision | Document AI API, Cloud Vision API, Cloud Video Intelligence API | Extract data from forms and invoices, analyze images and video |
| Discover more | Google Cloud API Library | Find and enable further Google APIs |
Agent Studio or Google AI Studio
| Google AI Studio | Agent Studio (on Gemini Enterprise Agent Platform) | |
|---|---|---|
| Purpose | Fast, simple prototyping with Gemini models | Building and running gen AI for the enterprise |
| Typical user | Individual developers, students, startups experimenting | Enterprise teams moving to production |
| Access | Simple sign-in and an API key | Google Cloud project with IAM |
| Enterprise features | Limited | Security, data governance, privacy controls, scaling, monitoring and integration with Google Cloud services |
| When to choose | Try ideas quickly, learn, prototype | Production use with company data and compliance needs |
A common path is to prototype in Google AI Studio, then move to Agent Studio and Agent Platform when the idea needs enterprise controls.
A worked scenario
A logistics company builds an agent that answers customer emails in many languages and updates deliveries. The agent uses the Translation API to read and reply, a data store of shipping policies for grounding, a function reschedule_delivery that the company's application executes after checking permissions, and Document AI to read uploaded proof-of-delivery forms. The team prototyped prompts in Google AI Studio and built the production agent with Agent Studio on Agent Platform for IAM, logging and scale.
How the exam asks about this
- "Let the model request an action that our own application executes." Functions (function calling).
- "Give the agent documents to ground its answers." A data store.
- "Extract fields from invoices as a tool." Document AI API.
- "Quick prototype by a single developer" is Google AI Studio; "enterprise production with governance" is Agent Studio.
What to remember for the exam
- Agents use tools to act: extensions, functions, data stores, plugins.
- Functions: the model proposes the call; the application executes it.
- Tools include Cloud Storage, databases, Cloud Run functions, Cloud Run, Agent Platform, and AI APIs for speech, language, documents, images and video.
- Google AI Studio for fast prototyping; Agent Studio for enterprise production with security and governance.
Do not choose Google AI Studio when the scenario needs enterprise data governance, IAM and compliance. That is Agent Studio on Agent Platform.
Common mistakes
Thinking an agent can act without tools. Tools are how agents reach data and systems.
Confusing data stores with functions. Data stores provide information; functions perform actions.
Using the Vision API for long documents with forms. Document AI is built for structured extraction from documents.
Check yourself
What is the difference between a data store and a function as agent tools?
A data store provides information for grounding; a function performs an action that the application executes.
When should a team choose Agent Studio over Google AI Studio?
When it needs enterprise security, governance, privacy controls and production scale with company data.
Which API should an agent use to extract fields from invoices?
The Document AI API.
Topic 6 of 6 in Google Cloud's gen AI offerings
Test yourself on Google Cloud's gen AI offerings
The full Google Cloud Generative AI Leader question bank has 512 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Google Cloud. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.