Skip to content
NTPU AI Learn

AI 104 Apply

Design a real Agentic workflow, from first step to final output, with a Canvas

What the Canvas is for

The Agentic AI Workflow Canvas is the capstone of AI 104. For one real use case you decompose the task, choose the workflow pattern, classify the decision points, place verification and human control, and plan what happens when something fails, so that the next stage can build it.

What you fill in

  • Goal, and the Trigger / Input that starts the workflow
  • Task Decomposition: tasks, subtasks, inputs, outputs, dependencies and who executes each (human, AI or tool)
  • Execution Order: Sequential, Parallel, Routing or Loop
  • Decision Points: rule-based, model-based, agentic or human
  • Memory, Knowledge & Skills each step needs (connects to AI 103)
  • Tool / Action Steps, without designing the tool connection
  • Execution Output and Verification for every step
  • Human-in-the-Loop: review, approval, escalation and override
  • Failure / Fallback, and the Final Output

Warning

A high-impact action should never run without a person's approval. State which outputs a person must confirm before anyone relies on them.

Examples

The canvas page shows a worked example (preparing a course session) and ideas for other occupations: ESG report review, customer request, investment brief, policy review and administrative request. All use simulated data.

Before you submit

Warning

Do not enter secrets, credentials, personal data, client data or sensitive information you are not authorized to share. Use simulated or public examples only.

What comes next: AI 201

AI 104 teaches how an agent executes a task. The Tool / Action Steps you identified become the input to AI 201: MCP, which shows how an agent connects safely to tools and trusted data. As always, a person stays responsible for reviewing the result.

Academic mapping: M5276 Week 4

AI 104 maps primarily to Week 4 of M5276 Agentic AI Innovative Applications (Case Study on Agentic AI Innovative Applications I: Workflow & Execution Design). Suggested class activity: Use Case, Task Decomposition, Workflow Pattern, Decision Points, Memory / Knowledge / Skills, Execution, Verification, Human Review.

It builds on Week 2 (task interpretation and decomposition) and Week 3 (memory, knowledge and skills). Forward links: Week 5 (Tool Use, which becomes AI 201), Week 10 (planning and reasoning), Week 13 (multi-agent collaboration) and Week 14 (agent harness and loop engineering).

Go to the canvas

What you need to complete

  • Complete all six AI 104 modules
  • Score at least 80% on the course post-test
  • To earn the Agentic Workflow Design Capability Certificate, also submit an Agentic AI Workflow Canvas and have it reviewed by a Credential Admin against the rubric

How you are assessed

A Credential Admin scores the Agentic AI Workflow Canvas across seven weighted domains: Goal & Trigger Framing, Task Decomposition & Execution Order, Decision Point Design, Memory/Knowledge/Skills & Tool Steps, Execution Output & Verification, Human-in-the-Loop & Failure Handling, and Final Output & Reflection. Out of 21 points, 17 or more with no critical-fail rule triggered is a PASS; 13 or fewer, or any triggered critical-fail rule, is a FAIL.

Info

This scoring logic is public so you know what is being assessed. It is not the same as quiz answer keys, and it is a different, human-reviewed pass from the automated reference score you already see on the canvas page -- the Credential Admin makes the final call.

What counts as evidence

  • A recorded rubric result (PASS, REVISION_REQUIRED, or FAIL) with no critical-fail rule triggered
  • A PASS leaves individual capability evidence for all eight AI 104 fine-grained competencies (Prompt, Workflow & Agent Diagnosis; Task Decomposition & Execution Steps; Workflow Pattern Selection; Decision Spectrum Classification; Basic Agent Loop; Verification Method Selection; Human-in-the-Loop Design; Agentic AI Workflow Canvas Design)

What you can earn

AI 104 Course Completion Certificate: complete all six modules and reach the post-test threshold -- no Agentic AI Workflow Canvas required. Agentic Workflow Design Capability Certificate: also requires a submitted Agentic AI Workflow Canvas that passes rubric review. The two are earned independently of each other. This is separate from AI 104's own pre-existing legacy course certificate described elsewhere on this page (which additionally requires all practice activities and at least one submitted canvas) -- the two mechanisms are evaluated independently.

What this does not mean

This credential confirms the holder completed AI 104's required coursework and post-test, or (for the capability certificate) demonstrated task decomposition, workflow pattern selection, decision point design, verification, and human-control skills on the Agentic AI Workflow Canvas.

  • It does not represent a formal automation or systems-integration professional license or hiring qualification
  • It does not mean any specific workflow the holder designs later will necessarily be correct or safe
  • It does not mean human review can be skipped before a high-risk action (Human, AI, Human)

What next?