Work in teams of 2–3 students. This is an investigation: you search, read, and use AI tools to understand (not to copy). No answers are given here.

Before you start

  • You manage your time and divide tasks as you want.
  • No “AI magic”: every claim must be tied to a mechanism (even explained simply).
  • Agentic AI ≠ Chatbot: an agent plans + uses tools + takes actions (often multi-step), so risks are higher.
  • Governance is mandatory: you must discuss human validation gates, logs/traceability, and failure modes.
  • Support key claims with at least one credible source (paper, textbook chapter, reputable technical explainer).
  • Examples must cover all three sectors: Tourism + Hospitality + Food Service.

1) Define Agentic AI

Write your definition in your own words. Keep it short and precise. You must include:

  1. The core function: an AI system that can pursue a goal by planning steps and executing actions via tools (APIs, software, workflows).
  2. What makes it “agentic”: autonomy level, multi-step behavior, tool-use, and feedback/iteration (not just one response).
  3. What it is NOT: list two common misunderstandings (e.g., “it always acts correctly”, “it replaces managers”).
Investigation task: find one credible definition of “agentic AI” (or “AI agents”) and cite it. Then rewrite it for a manager (simple wording, no jargon).
Examples requirement: add one concrete example sentence for each sector: Tourism / Hospitality / Food Service. (Your example must include at least one action the agent would take, not only “answering”.)

2) Explain the main algorithms (focus: planning, tool-use, control)

Explain how an agentic system works at a manager level, but with correct technical logic. Keep it clear: no math, no code. You must explicitly address how actions are chosen and controlled.

Required questions (you answer them — no copying)

  1. From goal to plan: What does it mean to “plan” in agentic AI? What is a “task decomposition” or “step-by-step plan”?
  2. Action space: What is an “action” for a software agent (examples: query a database, send a message, create a ticket, draft a document, call an API)?
  3. Tool use: How does the agent decide which tool to use, and what information it needs to call it safely?
  4. Memory / context: What is the difference between short-term context and longer-term memory? Why does it matter for consistency and errors?
  5. Grounding: What is “retrieval” (RAG) and why does it help in reducing wrong actions based on wrong information?
  6. Feedback loops: How can an agent detect failure and retry? What is “self-check” or “verification” in an agent loop?
  7. Guardrails: What are human validation gates, rules, and constraints? Where do you place them in the workflow?
  8. Failure modes: Give two technical/operational reasons agents can fail (not “AI is bad”).
Diagram task (no answers): draw a simple “agent loop” diagram (one block per step): Goal → Plan → Choose tool → Act → Observe result → Check/Revise → Log → Human approval (if needed). Add 2–3 “risk points” where mistakes would be costly.

3) Identify agentic tools (capabilities table — not use cases)

Fill this table with agentic capabilities (planning, tool-use, workflows, orchestration, approvals, logs). Do not write detailed use cases here (that’s section 4). You may add tools, but keep at least four.

Tool / Platform Agentic capabilities (planning / actions / orchestration) Tourism (fit) Hospitality (fit) Food Service (fit)
ChatGPT (GPTs / Assistants)        
Claude (tool use / workflows)        
Gemini (agent / tool use)        
Microsoft Copilot Studio        
Perplexity (search + actions)        
NotebookLM (document-grounded assistant)        
Mistral (API + assistants)        
Zapier (AI actions / automation)        
Mini-task: choose one tool/platform and write a 6-line “agentic capabilities checklist”. Then list 3 risks when the system can act (wrong actions, privacy/security, compliance/liability, etc.).

4) Propose possible uses (2–3 only)

Propose two or three realistic agentic use cases. Your examples must cover Tourism, Hospitality, and Food Service (at least one example per sector). Your use case must include at least one action the agent executes (not only “writing”).

Use-case card (copy/paste ×2 or ×3)

  • Sector: Tourism / Hospitality / Food Service
  • Use case title: ______________________________________________
  • Goal (business): ______________________________________________
  • Actions performed by the agent: _________________________________
  • Tools needed (systems/APIs): __________________________________
  • Human approvals (what must be validated): _______________________
  • Logging/traceability (what is recorded): _________________________
  • Limits & risks: _______________________________________________
  • What must remain human: _____________________________________
Mandatory safety requirement: for at least one use case, specify exactly where you place a “human-in-the-loop” gate (approval required) and what happens if the agent is uncertain or fails.

5) Reflective synthesis (short, critical)

Write 10–12 lines answering the questions below. Be specific. Be critical.

  1. Value vs hype: what is the realistic value of agentic AI across the three sectors?
  2. Main risk: what is the biggest risk when an AI can take actions (and why)?
  3. Accountability: who is responsible if the agent causes harm or costly mistakes?
  4. Minimum safeguards: list 3 safeguards you would impose (approvals, permissions, monitoring, audit logs, escalation rules).
  5. When NOT to use it: name one situation per sector where agentic AI should not be deployed (or only with strict limits).
Reminder: the written final exam questions will be built from students’ submissions. Anything you write can become an exam question.