Task 2.5: Define the purpose and types of tooling for gen AI agents.

• AThe model connects to the order API over the internet on its own

  Incorrect

  The model only proposes the call; your application decides whether to make it and makes it.

• BThe model names the function and arguments; the app calls the API

  Correct

  With function calling, the model does not execute anything itself: it returns a structured request naming the function and its arguments, the application calls the order API, and the result is passed back to the model for the answer.

• CThe model looks up the order in data it memorized during tuning

  Incorrect

  The model has no record of this customer's order; only the order API knows its current status.

• DThe model retrieves similar past orders from a vector database instead

  Incorrect

  Similarity search finds related passages; the customer needs the exact live status of one order.

Tools such as functions let an agent reach live systems, but the model only asks; the application acts.

• ADocument AI

  Correct

  Document AI turns unstructured documents into structured data: it digitizes with OCR, extracts fields and key-value pairs, and has pretrained parsers for documents such as invoices.

• BCloud Vision API

  Incorrect

  The Vision API analyzes images in general; it does not map an invoice's contents to named fields such as due date and line items.

• CCloud Natural Language API

  Incorrect

  The Natural Language API analyzes text that is already digital; it does not read scanned documents or their layout.

• DCloud Translation API

  Incorrect

  The Translation API translates text between languages; it does not extract fields.

For document field extraction, Document AI is the purpose-built tool; the other APIs cover images, text analysis and translation.

• AText translation in Cloud Translation Basic

  Incorrect

  Basic translates text strings; the agent would have to extract the text and lose the layout it must keep.

• BThe Cloud Text-to-Speech API with a German-language voice

  Incorrect

  Text-to-Speech turns text into spoken audio; it does not translate documents.

• CDocument translation in Cloud Translation Advanced

  Correct

  Cloud Translation Advanced can translate whole documents in formats such as DOCX and PDF while preserving their original formatting and layout.

• DDocument AI form parsing of each contract

  Incorrect

  Document AI does not translate documents.

When the format must survive translation, use document translation rather than plain text translation.

• AAgent Studio on Gemini Enterprise Agent Platform

  Incorrect

  Agent Studio needs a Google Cloud account with billing, which is more setup than this afternoon's personal prototype calls for.

• BGemini Enterprise

  Incorrect

  Gemini Enterprise is a licensed work assistant for organizations, not a developer playground for prototyping with the API.

• CGoogle AI Studio

  Correct

  Google AI Studio is the fastest way to start building with the Gemini API: you sign up with a Google Account, authenticate with an API key, and no-cost usage is available where applicable.

• DCX Agent Studio

  Incorrect

  CX Agent Studio is not a general place to experiment with Gemini prompts.

Google AI Studio is the quickest way to start experimenting with the Gemini API.

• ACloud Text-to-Speech

  Incorrect

  Text-to-Speech goes the other way, turning text into natural-sounding audio.

• BCloud Translation

  Incorrect

  Translation converts text between languages; it does not transcribe audio.

• CCloud Natural Language

  Incorrect

  Natural Language analyzes text for sentiment and entities; it needs the text first.

• DCloud Speech-to-Text

  Correct

  Speech-to-Text converts audio into a text transcription, including streaming and long audio, which the agent can then process.

Voice in is Speech-to-Text; voice out is Text-to-Speech.

• AA Compute Engine VM the team patches

  Incorrect

  A VM works, but the team would manage the server, which it wants to avoid for a small function.

• BCloud Run functions

  Correct

  Cloud Run functions is a lightweight compute option for single-purpose, stand-alone functions that respond to events or HTTP calls without managing a server or runtime.

• CA Gemini Notebook Enterprise notebook

  Incorrect

  Notebooks are for research over documents; they do not host code that an agent can call.

• DAn Agent Search data store

  Incorrect

  A data store gives an agent information to read; it cannot create a refund record.

Functions are a natural home for small tools: the agent calls them, and the platform handles the servers.

• ADocument AI

  Incorrect

  Document AI extracts structured fields from documents such as forms and invoices, not landmarks from travel photos.

• BCloud Vision API

  Correct

  The Vision API offers image labeling, face and landmark detection, and optical character recognition, which covers landmarks and sign text.

• CCloud Speech-to-Text

  Incorrect

  Speech-to-Text transcribes audio; the inputs here are photos.

• DCloud Natural Language

  Incorrect

  Natural Language analyzes text that already exists; it cannot read images.

For general image understanding tasks such as landmarks and OCR, the Vision API is the pre-built option.

• AA data store

  Correct

  Data stores give an agent access to information, such as indexed documents, that it can retrieve when answering.

• BA function

  Incorrect

  Functions define specific actions the agent can take, such as calling an order API; reading a document library is the job of a data store.

• CA Code Execution sandbox

  Incorrect

  Code Execution lets an agent run generated code in an isolated sandbox, for tasks such as calculations; it does not hold the manuals.

• DA safety filter

  Incorrect

  Safety filters block harmful content; they do not give the agent any knowledge.

Match the tool to the need: functions take actions, data stores provide information, and sandboxes run code.

• ACloud Translation API

  Incorrect

  Translation converts text between languages; it does not score sentiment.

• BCloud Vision API

  Incorrect

  The Vision API analyzes images; these posts are text.

• CCloud Natural Language API

  Correct

  The Natural Language API provides sentiment analysis, entity analysis, entity sentiment and content classification for text.

• DCloud Text-to-Speech API

  Incorrect

  Text-to-Speech produces audio from text; it does not analyze meaning.

Pre-built APIs give agents ready-made skills: sentiment and entities come from the Natural Language API.