Zero to Agentic: Building Workflows and Agent Systems in Dify

Date: 5 November 2026 (Thu)

Time: 15:30 - 17:00

Location: M407

Target Audience: All students

Medium of Instruction: Cantonese

Objective

This workshop introduces participants to the fundamental concepts of Generative AI and Agentic AI, with a focus on designing AI applications that are simple, reliable, and controllable.

Through hands-on exercises using Dify, participants will learn how to choose between workflows, single-agent, and multi-agent approaches, apply prompt and context engineering, and manage AI actions safely.

What you will learn

By the end of this workshop, participants will be able to:

• Understand the differences between GenAI, workflows, and Agentic AI.
• Choose suitable architectures using workflow, single-agent, or multi-agent approaches.
• Design AI systems with clearer prompts, structured outputs, validation, and human approval.
• Apply RAG and Context Engineering to provide relevant information to AI.
• Understand how to reduce risks from incorrect, unpredictable, or irreversible AI actions.

Agenda

1. Introduction to GenAI & Agentic AI

- What is GenAI and Agentic AI?

• Understand the basic concepts and differences between Generative AI and Agentic AI
• Learn how AI agents can make decisions and perform actions toward a goal using workflow
• Identify suitable scenarios for using GenAI, workflows, or agents

- Workflow vs Agent

• Understand the key differences between predefined workflows and dynamic agent behavior
• Learn when a workflow may be more suitable than an agent
• Recognize the trade-offs between stability, flexibility, and control

- Defining Goals and Boundaries

• Learn how to define clear goals, roles, and responsibilities for an AI agent
• Understand the importance of setting allowed and prohibited actions
• Learn how clear input and output requirements can improve agent reliability
  
  

2. Designing Reliable AI Agents

- Single Agent vs Multi-Agent

• Understand use case of single agent, pros and cons
• Understand use case of multiple agents, pros and cons
• Learn why using more agents does not automatically lead to better results

- Choosing the Right Architecture

• Learn how to choose between workflows, single-agent, and multi-agent approaches
• Understand how task complexity and responsibilities affect architecture decisions
• Apply the principle of using the simplest architecture that can achieve the goal

- Agent Awareness and Reliability and Safety

• Understand common risks and limitations when designing AI agents
• Learn how validation, structured outputs, and human approval can improve reliability
• Recognize high-risk action to prevent irrecoverable actions that require Human-in-loop action

3. Practice 1 – Christmas Card Website

- Build a Simple GenAI Workflow

• Learn how to build a simple GenAI workflow using Dify
• Understand how user input can be transformed into generated content
• Learn how to connect different workflow steps into a complete application

- Structured Output and Prompt Design

• Learn how System Prompts can guide AI behavior and output
• Understand how structured outputs can improve consistency and predictability
• Learn how input and output validation can reduce unexpected results

- Choosing Between GenAI and Agents

• Understand when a simple GenAI workflow is sufficient for a task
• Learn why unnecessary agent behavior can increase complexity
• Apply simple and reversible design principles to AI applications

4. Practice 2 – Travel Agent for Tourists

- RAG and Context Engineering

• Understand how RAG can provide external knowledge to AI applications
• Learn how relevant context can improve the quality of AI responses
• Understand how to prepare and provide useful data to AI agents

- Multi-Agent Architecture

• Learn how different agents can be assigned different responsibilities
• Understand how multiple agents can cooperate toward a shared goal
• Learn how to separate complex tasks into specialized agent roles

- Agent Cooperation and Task Decomposition

• Learn how to break a complex task into smaller and manageable tasks
• Understand how data and outputs can be passed between different agents
• Learn how structured communication can improve multi-agent reliability

5. Wrap-up and Q&A

- Review Core Concepts

• Reinforce the differences between GenAI, workflows, single agents, and multi-agent systems
• Review key concepts including RAG, Context Engineering, and structured outputs
• Consolidate the main design principles covered throughout the workshop

- Review AI Design Principles

• Start with the simplest solution that can achieve the goal
• Keep agent roles, actions, and outputs clearly defined
• Add Human-in-the-Loop controls for important actions

- Key Takeaway and Q&A

• Review the key lessons learned from both practical exercises
• Discuss common challenges when designing AI agents
• Clarify participant questions and explore possible real-world applications