Domain 2 of 4 · Chapter 5 of 5

Gen AI Agent Tooling

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing Generative AI Leader premium course — no separate purchase.

14-day money-back guarantee — no questions asked.

Included in this chapter:

  • From model to agent: what a tool is
  • Who runs the call: extensions, functions, and data stores
  • Pre-built AI APIs: match the API to the capability
  • The building blocks a tool runs on
  • Google AI Studio vs Vertex AI Studio
  • Exam-pattern recognition

The three functional tool types (a plugin is a packaged extension)

AspectExtensionFunctionData store
What it isAn API wrapper the agent can callA function signature the model fills inA body of your own content to retrieve from
Who executes the callVertex AI, automatically on the agent sideYour own application or client codeThe retrieval layer that grounds the answer
Reach for it whenThe agent should call a live API itselfYour app must control the action or side effectAnswers must be grounded in your own data
Typical backing serviceA published API endpointCloud Run or Cloud Run functionsCloud Storage or a database via Vertex AI Search

Decision tree

Answer from your ownprivate data?Standard perceptiontask?Who executes theexternal API call?Data storegrounding / RAGPre-built AI APIVision, Speech, Doc AI, ...Function callingyour app runs itExtensionVertex AI runs itYesNoYesNoyour appagent alone
Cheat sheet

Cheat sheet

  • A tool is how an agent reaches past the model's training data
  • Agent tools come in four types: extensions, functions, data stores, plugins
  • Reach for an extension when the agent should call a live API on its own
  • Reach for function calling when your app must control the action
  • A data store grounds an answer in your own content
  • A plugin is a packaged, reusable extension, not a fourth mechanism
  • Call a pre-built AI API before training your own model
  • Match the pre-built AI API to the capability
  • Use Document AI, not Vision, to turn a form into fields
  • An agent's tools run on Storage, databases, and serverless compute
  • Prototype in Google AI Studio: fast, free, just an API key
  • Move to Vertex AI Studio when you need enterprise controls
  • No tool needed when the model can already answer
  • Choosing the build tier is a separate decision from choosing tools
  • Vertex AI Studio is the no-code workspace for designing and testing prompts

Unlock with Premium — includes all practice exams and the complete study guide.

References

  1. Vertex AI Extensions overview
  2. Vertex AI function calling
  3. Vertex AI grounding overview
  4. Cloud Vision
  5. Video Intelligence API
  6. Speech-to-Text
  7. Text-to-Speech
  8. Cloud Translation
  9. Document AI overview
  10. Natural Language API
  11. Cloud Storage
  12. Cloud Run
  13. Cloud Run functions (Cloud Functions)
  14. Vertex AI
  15. Google AI Studio
  16. Vertex AI data governance
  17. Migrate from Gemini API to Vertex AI