Memory hook: Buy the workflow; build the differentiator.
Must remember
- Gemini in Workspace supports everyday writing, meetings and analysis; the Gemini app provides conversational assistance. Managed enterprise offerings add organizational access and administration controls; do not assume consumer and enterprise terms match.
- Gemini Enterprise brings enterprise search and agent experiences to organizational information. Notebook-style experiences ground work in selected sources; connectors still need identity-aware access and freshness.
- Customer experience agents combine conversation, knowledge retrieval, escalation and business-system actions. A deterministic flow suits strict transactions; generative conversation suits flexible requests.
- Code assistants help explain, generate and review code. Human review, tests, dependency scanning and secret protection remain necessary; generated code is not automatically production-ready.
- The managed AI platform supplies model access, development, evaluation and deployment. Model Garden offers model choices; managed APIs avoid building every capability from scratch. TPU/GPU infrastructure helps demanding training and inference workloads.
- An agent combines a model, instructions, tools, context and an execution loop. Function calling proposes structured tool arguments; application code authorizes and executes them. Search grounds answers, while tools can change external state.
Review details
For customer experience, Conversational Agents supports customer-facing conversation, Agent Assist helps a human support representative, and Conversational Insights analyzes conversations for trends. Do not choose an analytics tool when the requirement is live transaction handling. Google AI Studio is a model/prompt prototyping environment; enterprise Agent Studio/designer and platform capabilities address agent construction, deployment and governance under their supported models.
Speech-to-Text transcribes; Text-to-Speech synthesizes audio; Translation/Document Translation handles language; Document AI extracts structured meaning from documents; Vision/Video Intelligence interprets images/video; Natural Language handles supported text analysis. Select an existing API when it meets the task before commissioning new model training.
Choose under exam pressure
| Requirement | Choice and reason |
|---|---|
| Standard meeting summaries | Workspace AI capabilities before a bespoke model pipeline. |
| An assistant must update an order | Authorized tool integration with validation, audit and appropriate approval. |
Traps
- Adding a connector does not authorize every employee to every connected document.
- A chatbot that only returns text is not necessarily an autonomous agent.
Active recall
1. What distinguishes a tool from a model answer?
A tool executes an external capability; an answer is generated content.
2. When is a prebuilt API preferable?
When it meets the business need with less implementation and maintenance.
3. Why use an enterprise search connector?
To retrieve relevant organizational content while respecting permissions.
4. Does a code assistant replace tests?
No; tests verify observable behavior and catch plausible but wrong code.
5. Why are TPUs relevant?
They are accelerators optimized for machine-learning workloads, not a requirement for every AI application.