Question 1

What is Prompt Chaining in Agentic AI?

Accepted Answer

Prompt Chaining is an agentic design pattern that decomposes a complex, multi-step task into a sequence of smaller, structured prompts, where the generated output of each step is systematically validated and passed as context input into the next step. Rather than asking a single LLM to execute a large, monolithic task in one go—which increases cognitive load, token usage, and the risk of hallucinations—Prompt Chaining constructs a modular pipeline. Each prompt performs a single sub-task (e.g. topic analysis, outline construction, draft generation, review). If any step fails or generates malformed output, the system can isolate the error and re-run that specific step rather than repeating the entire execution.

Question 2

Explain the Routing Pattern in Agentic AI systems.

Accepted Answer

The Routing Pattern dynamically directs incoming user queries to specialized agents, prompts, or databases based on an initial intent classification, input complexity analysis, or targeted semantic matching. In enterprise setups, a single general-purpose prompt cannot handle all scenarios efficiently. The Routing Pattern deploys a router (which can be a fast, lightweight LLM, a classifier, or semantic search embeddings) to analyze the query first. The query is then routed to the most qualified handler (e.g., billing queries go to the billing agent, system bugs to the engineering agent). This optimizes processing speeds, keeps token costs low, and ensures highly specialized answers.

Question 3

What is the Orchestration Pattern in Agentic AI?

Accepted Answer

The Orchestration Pattern uses a central coordinator (Orchestrator) driven by an LLM to dynamically decompose a high-level goal into individual sub-tasks, dispatch those tasks to specialized worker agents, manage dependencies, and aggregate the outputs into a final synthesized result. Unlike rigid chains where the sequence is pre-determined, the Orchestration Pattern allows for dynamic workflows. The orchestrator agent acts as a manager: it creates a plan of action based on the user's goal, dispatches individual tasks to specialized workers (e.g., flight booking agent, hotel search agent), monitors their execution, resolves any dependency bottlenecks, and combines their findings into a cohesive summary.

Question 4

Describe Memory Management Patterns in Agentic AI systems.

Accepted Answer

Memory Management Patterns handle how an agentic system stores, filters, retrieves, and preserves historical context across conversations and multi-step executions. Memory is divided into Short-Term (session context), Long-Term (persistent databases), and Episodic (semantic vector recollections). Agents require memory to deliver coherent long-term interactions:
1. **Short-Term Memory**: The active conversation window. It is maintained using sliding context buffers that truncate old messages to fit context limits.
2. **Long-Term Memory**: Stores user profiles, preferences, and facts in persistent databases.
3. **Episodic Memory**: Captures specific past execution cycles or conversational interactions, using vector embeddings and semantic search to retrieve them only when relevant to the current task.

Question 5

How does the Tool Integration Pattern work in Agentic AI?

Accepted Answer

The Tool Integration Pattern (or Tool Use/Function Calling) allows an LLM to interact with external systems by dynamically selecting and formatting parameters for APIs, databases, calculators, or code sandboxes based on natural language descriptions. LLMs cannot perform calculations or access real-time databases on their own. The Tool Integration pattern registers available tools using natural language descriptions (schemas). When a task requires a tool, the LLM generates a structured JSON block containing the name of the tool and the parsed arguments. The execution environment catches this block, runs the tool, and returns the result to the LLM to complete the generative response.

Question 6

Implement a basic Prompt Chaining function that processes a task through multiple sequential prompts.

Accepted Answer

Here is a production-grade Python implementation of Prompt Chaining. The function decomposes a high-level task by processing it sequentially through a series of prompt templates, using static type hints and docstrings. In this implementation, the `prompt_chain` function iterates through a list of templates. Each template is formatted using the output of the previous step. A callback function `llm_call` handles the actual interaction with the language model, separating the pipeline logic from the API implementation.

Question 7

Write a routing function that directs queries to specialized agents based on intent classification.

Accepted Answer

Here is an object-oriented Python implementation of the Routing Pattern, featuring protocol-based agent interfaces and a dynamic intent classifier with fallback safety. This implementation uses Python's `Protocol` from the `typing` module to define a strict structural contract for all agents. The router uses an intent classifier callback to evaluate the query, maps the intent to the corresponding specialized agent, and falls back to a default general-purpose agent if no matching intent is found.

Question 8

Create a simple orchestrator class that coordinates multiple agents for a multi-step task.

Accepted Answer

Here is a Python implementation of the Orchestrator-Workers pattern, where a central coordinator parses a high-level goal, executes individual sub-tasks through worker agents, and aggregates the results. In this architecture, the `AgentOrchestrator` uses a planner agent to deconstruct a user's goal into a list of structured task steps. The orchestrator loops through this plan, maps each step to a specialized worker, and aggregates all worker outputs. Finally, an aggregator agent combines these outputs into a single cohesive response.

Question 9

Implement a memory management system for an AI agent with short-term and long-term storage.

Accepted Answer

Here is a Python implementation of an agent memory management class, featuring a sliding context window for short-term conversation memory and a key-value store for persistent long-term memory. This `AgentMemory` class manages memory limit allocations. When saving messages to short-term memory, it automatically removes the oldest messages once a size limit is reached, protecting the LLM from exceeding its context window. It also provides a separate long-term memory store to persist user preferences and facts across sessions.

Question 10

Write a tool selection function that dynamically chooses appropriate tools based on task description.

Accepted Answer

Here is a Python implementation of dynamic, semantic tool selection using cosine similarity, allowing an AI agent to select the most relevant tool for a task based on text embeddings. This implementation uses a cosine similarity algorithm to measure the semantic closeness between a task description vector and registered tool description vectors. The function calculates the dot product and vector magnitudes to determine similarity, and returns the tool only if the similarity score meets a minimum threshold (e.g. 0.70).

Agentic AI
Design Patterns Hub

What is Prompt Chaining in Agentic AI?

Explain the Routing Pattern in Agentic AI systems.

What is the Orchestration Pattern in Agentic AI?

Describe Memory Management Patterns in Agentic AI systems.

How does the Tool Integration Pattern work in Agentic AI?

Implement a basic Prompt Chaining function that processes a task through multiple sequential prompts.

Write a routing function that directs queries to specialized agents based on intent classification.

Create a simple orchestrator class that coordinates multiple agents for a multi-step task.

Implement a memory management system for an AI agent with short-term and long-term storage.

Write a tool selection function that dynamically chooses appropriate tools based on task description.

Finished practicing?

Agentic AI Design Patterns Hub

What is Prompt Chaining in Agentic AI?

Explain the Routing Pattern in Agentic AI systems.

What is the Orchestration Pattern in Agentic AI?

Describe Memory Management Patterns in Agentic AI systems.

How does the Tool Integration Pattern work in Agentic AI?

Implement a basic Prompt Chaining function that processes a task through multiple sequential prompts.

Write a routing function that directs queries to specialized agents based on intent classification.

Create a simple orchestrator class that coordinates multiple agents for a multi-step task.

Implement a memory management system for an AI agent with short-term and long-term storage.

Write a tool selection function that dynamically chooses appropriate tools based on task description.

Finished practicing?

Agentic AI
Design Patterns Hub