Generative AI Code Examples: Practical Patterns for Developers

The best way to learn generative AI development is through real code patterns. These are the patterns I use daily in production.

Pattern 1: Basic API Call

The foundation of everything. Send a message, get a response. Works the same across every major API - only the import and model name change.

Key concepts:

• System prompt sets the AI's behavior
• User message is the input
• Temperature controls creativity (0 = deterministic, 1 = creative)
• Max tokens limits response length

Pattern 2: Streaming Responses

Never make users stare at a loading spinner. Stream tokens as they generate.

Why it matters:

• First token appears in ~200ms instead of waiting 2-5 seconds for the full response
• Users perceive the response as faster even though total time is the same
• Essential for any chat-like interface

Implementation approach: Use the streaming API option from your provider, read from the stream chunk by chunk, and render each chunk to the UI immediately.

Pattern 3: Structured Outputs

Get JSON instead of prose. Critical for building reliable AI features.

Use cases:

• Classification (sentiment, category, intent)
• Extraction (names, dates, amounts from text)
• Decision-making (approve/reject with reasoning)

Implementation approach: Define a JSON schema in your prompt or use the provider's structured output feature. Always validate the response against your expected schema.

Pattern 4: Tool Use

Let the AI call functions in your application. This is the bridge from generative to agentic AI.

Common tools:

• Database queries
• API calls to external services
• File operations
• Calculations and data processing

Implementation approach: Define tool schemas with names, descriptions, and parameter types. Send them alongside your messages. Handle tool call responses by executing the function and returning results.

Pattern 5: RAG (Retrieval-Augmented Generation)

Feed relevant context from your data into the AI prompt. The most common pattern for domain-specific AI features.

How it works:

1. Convert your documents into vector embeddings
2. When a user asks a question, find the most relevant documents
3. Include those documents in the prompt as context
4. AI answers using your specific data, not just general knowledge

When to use: Customer support bots, documentation search, internal knowledge bases - any time you need AI to answer from your data.

Pattern 6: Prompt Chaining

Break complex tasks into sequential AI calls. Each call builds on the previous output.

Example workflow:

1. Call 1: Extract key information from raw text
2. Call 2: Classify the extracted information
3. Call 3: Generate a response based on the classification

Why chain instead of one big prompt: Each step is simpler and more reliable. Easier to debug. Can use different models for different steps (cheap model for extraction, expensive for generation).

Putting It Together

These patterns combine to build real features:

• AI chatbot: Streaming + RAG + conversation history
• Content generator: Basic call + structured output + prompt chaining
• Data processor: Tool use + structured output + batch processing
• AI agent: Tool use + prompt chaining + ReAct loop

For the complete guide to building with AI, see How to Code AI. For API selection, see Best AI APIs.

Frequently Asked Questions

Which programming language is best for generative AI development?

How do I reduce AI API costs in production?

What's the most useful pattern to learn first?