1.6 The Four Gen AI Use-Case Patterns: Create, Summarize, Discover, Automate
Key Takeaways
- The exam guide groups gen AI business value into four verbs - create, summarize, discover, and automate - and scenario questions expect you to name the pattern before naming a product.
- Create covers text, image, code, video, and synthetic data generation; the business metric is production throughput and cost per asset.
- Summarize compresses long or multimodal source material into decision-ready output; the business metric is hours of reading avoided.
- Discover uses semantic search, embeddings, and analysis to surface knowledge that keyword search cannot reach; the business metric is time-to-answer.
- Automate chains reasoning with tool calls so an agent completes multi-step work end to end; the business metric is cases resolved without human touch.
1.6 The Four Gen AI Use-Case Patterns: Create, Summarize, Discover, Automate
Executive Summary: The exam guide asks candidates to "identify business use cases where gen AI can create, summarize, discover, and automate." These four verbs are the exam's organizing frame for value, and they recur in the business-strategy section as well. A leader who can classify an incoming request into one of the four patterns can immediately predict the architecture, the likely Google Cloud product, and the metric that will prove return on investment.
Why Four Patterns Instead of a Product List
Business stakeholders bring requests in the language of their function: "help legal review contracts faster," "stop customers calling us," "write more product copy." Those requests do not map cleanly to products. They do map cleanly to patterns, and each pattern implies a different architecture and a different measure of success.
| Pattern | The job to be done | Representative Google Cloud fit | Business metric that proves value |
|---|---|---|---|
| Create | Produce new content that did not exist | Gemini on Agent Platform, Imagen, Veo, Gemini Code Assist | Assets produced per period; cost per asset |
| Summarize | Compress long or multimodal material into decision-ready form | Gemini long-context, Gemini for Google Workspace, Gemini Enterprise | Reading hours avoided; time to decision |
| Discover | Surface knowledge that keyword search cannot reach | Agent Search, Agent Platform Vector Search, grounding | Time to answer; self-service resolution rate |
| Automate | Complete multi-step work end to end with tools | Agent Platform custom agents, Customer Engagement Suite | Cases resolved without human touch; deflection rate |
Pattern 1: Create
Create covers generating new artifacts: marketing copy, product descriptions, images, video, code, synthetic training data, and first-draft documents.
- Text generation. A retailer generates 40,000 SKU descriptions in brand voice, a task that previously required an agency.
- Image generation. A catalog team uses Imagen to produce seasonal lifestyle backgrounds and to inpaint product variants, avoiding repeat photoshoots.
- Video generation. A training team uses Veo to produce short instructional clips with controlled camera behavior.
- Code generation. Engineering uses Gemini Code Assist for boilerplate, tests, and legacy language migration.
- Synthetic data. A bank generates realistic but non-identifying transaction records so a fraud team can develop against production-like data without exposing customer information.
What leaders must watch: create-pattern outputs are the highest-volume producers of brand, legal, and factual risk, because content ships externally. This is the pattern that most needs review workflows, watermarking such as SynthID, and clear accountability for what is published.
Pattern 2: Summarize
Summarize compresses volume into a decision.
- A claims adjuster receives a two-page synthesis of a 300-page medical file.
- A sales leader gets an account brief assembled from twelve months of email, call transcripts, and support tickets.
- A compliance officer receives a change summary between two regulatory drafts.
- A manager gets meeting notes and action items from a recorded call in Google Meet through Gemini for Google Workspace.
The economics are unusually easy to defend: multiply the number of people, the hours of reading avoided per week, and the loaded hourly rate. This is why summarization is so often the first funded use case - the baseline is measurable and the change is immediate.
Multimodality changes what "long" means. Because Gemini natively accepts video, audio, images, and text in the same prompt, summarization now extends to a recorded plant inspection or an hour of customer calls, not just documents.
Pattern 3: Discover
Discover is about finding what you already own. Enterprises typically cannot answer their own questions, not because the answer is missing but because keyword search cannot reach it.
Semantic search using embeddings matches on meaning rather than exact words, so a query for "what happens if a shipment is damaged in transit" retrieves a clause titled "Carrier Liability for Goods in Transit" that contains none of the query terms. Agent Search provides this as a managed capability, and Agent Platform Vector Search provides the underlying similarity search at scale.
Discovery also covers analytical exploration: asking a natural-language question over structured data in BigQuery and receiving both the answer and the query that produced it.
Business metric: time-to-answer, and the proportion of questions answered without escalation to an expert.
Pattern 4: Automate
Automate is the pattern where the system does not merely produce text - it acts. An agent reasons about a goal, calls tools and APIs, observes results, and iterates until the task is complete.
- A customer-service agent checks an order in the ERP system, issues a refund within policy limits, and emails confirmation.
- An IT agent triages a ticket, provisions access, and updates the service management record.
- A finance agent extracts invoice fields, matches them against purchase orders, and routes only genuine exceptions to a human.
Automation carries the highest value and the highest governance burden, because the system changes state in real systems. This is where human-in-the-loop thresholds belong: agents act autonomously inside defined limits and escalate beyond them.
Business metric: the share of cases resolved end to end without human touch, alongside a quality bar that the automation must not degrade.
The Cross-Cutting Pattern: Personalized User Experience
The exam guide also names personalized user experience among its examples. Personalization is best understood as a modifier applied across the other four: creating copy tailored to a segment, summarizing an account for one specific relationship manager, discovering products for one shopper's stated intent, or automating a next-best-action per customer. When a scenario stresses per-individual adaptation, expect personalization to be the differentiator between otherwise similar options.
Classifying a Request in Practice
Ask one question: what leaves the system?
| What leaves the system | Pattern |
|---|---|
| A new artifact that did not exist before | Create |
| A shorter version of something that already existed | Summarize |
| A pointer to existing knowledge, plus the answer it supports | Discover |
| A completed action in another system | Automate |
Worked example. A hospital network asks for "an AI to help our nurses." Decomposed: drafting discharge instructions is create; condensing a patient's chart before handover is summarize; finding the current protocol for a rare presentation is discover; scheduling a follow-up appointment in the electronic health record is automate. Four patterns, four architectures, four metrics - and a portfolio conversation instead of a single vague project.
Strategic Leadership Guidance: Exam Tips and Common Pitfalls
[!TIP] Exam Tip: Identify the pattern before you pick the product. Most incorrect answers in scenario questions are products that serve a different pattern from the one the scenario actually describes.
Pitfall 1: Calling retrieval-augmented generation a "create" use case. RAG is the enabling technique for discover, and it supports grounded summarization; the value is finding and citing the right source, not producing novel content.
Pitfall 2: Treating automation as summarization with extra steps. The dividing line is whether the system writes back to another system. If it changes state, it is automation and it needs tool permissions, audit logging, and escalation thresholds.
Pitfall 3: Proposing a single enterprise-wide "AI project." Real portfolios contain several use cases across different patterns, each with its own owner and metric.
A telecom asks for a system that reads an inbound support email, looks up the account in the billing platform, applies a credit when the customer is within policy, and updates the ticket - escalating to a human only when the credit exceeds a threshold. Which use-case pattern does this describe?
A pharmaceutical company wants researchers to ask questions in plain language over 400,000 internal study reports and receive answers with citations to the source documents. Keyword search currently fails because researchers and authors use different terminology. Which pattern and enabling capability match this need?
Which statement best captures why a leader should classify a request into a use-case pattern before selecting a Google Cloud product?