AI Agents and Applications人工智慧代理人與應用

Course Description課程簡介

This course introduces the theory, tools and applications of AI agents from a sociological perspective, with no programming background assumed. The theory covers large language models, prompt and context engineering, the ReAct agent architecture, RAG, MCP, multi-agent systems, evaluation methods, information security and governance. The hands-on part compares AI agent tools (Claude Code, OpenAI Codex, Gemini CLI, Cursor, GitHub Copilot, Lovable, Bolt.new, v0, Replit Agent, n8n and others) and uses vibe coding to build websites, apps and research agents. The social science part explores social simulation with generative agents, qualitative coding with LLMs, and the effects of AI on work, inequality and accountability. Assessment includes biweekly assignments, a midterm, a final exam and a final project (theoretical analysis plus a working product). The tool list will be updated to reflect the versions and free plans available when the course runs.

Weekly Schedule各週重點

1. W01

   Introduction: from chatbots to AI agents
   Key terms: AI Agent, Agentic AI, Rational Agent, PEAS, Autonomy, Human-in-the-Loop, Sociotechnical System
   Hands-on: Set up the course tools and run an AI agent through a multi-step task.

2. W02

   How large language models work
   Key terms: LLM, Foundation Model, Transformer, Attention, Token, Context Window, Embedding, Temperature, Hallucination
   Hands-on: Compare outputs across models and temperatures using the same prompt. Assignment 1: model output comparison and hallucination check

3. W03

   Prompt engineering and context engineering
   Key terms: System Prompt, Few-Shot, Chain-of-Thought, Structured Output (JSON), Prompt Engineering, Context Engineering
   Hands-on: Write reusable system prompts and output formats. Final project groups formed

4. W04

   Core architecture of AI agents
   Key terms: ReAct, Agent Loop, Tool Calling, Planning, Memory, Reflection, Orchestration
   Hands-on: Build a first agent in n8n that can call external tools. Assignment 2: tool-using agent in n8n

5. W05

   Knowledge retrieval and tool protocols
   Key terms: RAG, Vector Database, Chunking, Semantic Search, MCP, Agent Skills, A2A
   Hands-on: Build a RAG question-answering agent from social policy documents and connect an MCP server.

6. W06

   Survey and comparison of AI agent tools
   Key terms: Terminal agents (Claude Code, OpenAI Codex, Gemini CLI); IDE agents (Cursor, GitHub Copilot, Windsurf, Google Antigravity); cloud autonomous agents (Devin); app builders (Lovable, Bolt.new, v0, Replit Agent); workflows and frameworks (n8n, Dify, LangGraph, CrewAI, AutoGen)
   Hands-on: In groups, try different tools on the same task and record cost, speed and quality. Assignment 3: AI agent tool review

7. W07

   Vibe coding: theory and methods
   Key terms: Vibe Coding, Agentic Coding, Spec-Driven Development, PRD, AGENTS.md / CLAUDE.md, Git, Version Control, Code Review, Technical Debt
   Hands-on: Write a requirements spec, have a coding agent produce a first version, and manage changes with Git.

8. W08

   AI agent applications (1): building websites
   Key terms: Frontend, HTML / CSS / JavaScript, React, Responsive Design, Supabase, Deployment (Vercel, GitHub Pages), API Key, Environment Variables
   Hands-on: Use vibe coding to complete and publicly deploy a website on a social issue. Assignment 4: social issue website goes live

9. W09

   Midterm exam Covers weeks 1–8 (key terms, essays and reading hands-on work)

10. W10

    AI agent applications (2): building apps
    Key terms: PWA, React Native, Expo, Backend, REST API, Authentication, UX, Accessibility
    Hands-on: Use tools such as Replit Agent or Lovable to build a phone-ready app prototype and run user tests. Assignment 5: mobile app prototype; final project proposal due

11. W11

    Multi-agent systems
    Key terms: Multi-Agent Systems, Orchestrator-Worker, Role Assignment, Handoff, Shared State, Emergent Behavior, LangGraph, CrewAI
    Hands-on: Design a three-role collaborating agent using division-of-labor concepts from organizational sociology.

12. W12

    Agent evaluation and observability
    Key terms: Evals, Benchmark, SWE-bench, Golden Dataset, Trajectory, Observability, Guardrails, Validity and Reliability
    Hands-on: Build an evaluation dataset for your own agent and calculate performance metrics. Assignment 6: agent evaluation report

13. W13

    Generative agents and social simulation
    Key terms: Generative Agents, Agent-Based Modeling, Silicon Sampling, Algorithmic Fidelity, Synthetic Respondents, External Validity, WEIRD Bias
    Hands-on: Simulate a Taiwanese public opinion item and compare it with real survey results.

14. W14

    AI agents as social research assistants
    Key terms: Computational Social Science, Research Workflow Automation, Codebook to Promptbook, LLM Annotation, Cohen's Kappa, Research Ethics
    Hands-on: Use an agent to code interview or news texts and compare its agreement with human coding. Assignment 7: LLM coding vs. human coding

15. W15

    AI agents, work and social inequality
    Key terms: Task-Based Approach, Automation vs. Augmentation, Deskilling, Productivity Paradox, Digital Divide, Algorithmic Management
    Hands-on: Analyze the tasks of one occupation in Taiwan and assess which could be replaced or augmented by agents.

16. W16

    Security, ethics and governance
    Key terms: Prompt Injection, Least Privilege, Sandbox, Data Leakage, Red Teaming, Moral Crumple Zone, Accountability, Taiwan's AI Basic Act, EU AI Act
    Hands-on: Red-team your final project and check personal data and permissions. Assignment 8: project security and ethics check

17. W17

    Final project presentations (Demo Day)
    Key terms: Proof of Concept, MVP, Peer Review
    Hands-on: Each group demonstrates its agent, website or app live and explains its sociological basis and evaluation results. Final project presentation

18. W18

    Final exam Covers weeks 10–16; final project report and product links due