L0
Casual AI use: useful for exploration, weak for repeatable work.
單次使用：適合發想，但很難支撐穩定工作。
August 16, 2026 | 09:00-16:20 | CSKM Global Learning Center
AI Agent Business Platform Workshop
AI Agent 商業平台實作工作坊
A full-day MBA workshop for working professionals who need to move beyond AI tool trials and design workflows that can be reviewed, handed over, and measured.
這是一堂給在職 MBA 學員的整日工作坊。課程重點是把 AI 構想整理成可檢查、可交接、可衡量的商業工作流程。
Real classroom screenshot: ChatGPT Projects as a workspace for instructions, files, and iterative work.
真實課堂截圖：用 ChatGPT Projects 把指令、檔案與反覆修正放在同一個工作空間。
Course Positioning
A management design studio for AI workflows.
管理者的 AI 工作流程設計室。
The course treats AI agents as part of business process design. Students will learn where AI should gather data, prepare judgment, call tools, hand work back to people, and leave records.
課程把 AI Agent 放回商業流程設計。學員會學習 AI 應該在哪裡整理資料、準備判斷、呼叫工具、交回人工審核，並留下可追蹤紀錄。
L1
Reusable instructions: turn a personal method into a stable task pattern.
可重複指令：把個人的做事方法整理成穩定任務。
L2
Workflows and skills: connect files, tools, checks, and team notes.
工作流程與 Skills：把檔案、工具、檢查與團隊紀錄串起來。
L3
Agent platforms: route data, make bounded judgments, trigger actions, and measure value.
Agent 平台：處理資料、做有邊界的判斷、觸發行動並衡量價值。
Full-Day Agenda
A workshop paced around decisions and outputs.
以判斷與產出推進的一日工作坊。
Every lecture block leads to a concrete field in the final project brief. Students do not leave with scattered notes; they leave with a proposal seed.
每一段短講都會接到 final project brief 的欄位。學員最後會帶走一份能繼續發展的 proposal seed。
09:00-09:20
Opening and problem framing開場與問題設定Choose one real process to work on throughout the day.先選定一個整天沿用的真實流程。
Candidate process
流程候選題
09:20-10:05
From chat to workflow從聊天到工作流程Hallucination, context limits, drift, verification, and the L0-L3 model.幻覺、上下文限制、輸出漂移、查核與 L0-L3 架構。
AI use-level self-check
AI 使用層級自評
10:20-11:10
Reusable instructions可重複使用的指令Task, constraints, steps, output format, and workspace choices.任務、限制、步驟、輸出格式與工作空間選擇。
First reusable instruction
第一版可重複指令
11:10-12:00
Agent platform logicAgent 平台邏輯Goal, data, judgment, tool, action, record, and human review.目標、資料、判斷、工具、行動、紀錄與人工審核。
Problem definition table
問題界定表
13:00-13:45
Case discussion案例討論Customer Signal Agent Platform, with side lenses from Starbucks/NomadGo, Ramp, and Palantir.Customer Signal Agent Platform，並對照 Starbucks/NomadGo、Ramp 與 Palantir。
Group topic
小組題目
13:45-15:35
Group design workshop小組設計工作坊Blueprint the workflow, review points, MVP, risk controls, and ROI.完成流程藍圖、審核點、MVP、風險控管與 ROI。
Project brief draft
Project brief 初稿
15:35-16:20
Presentation and closing發表與收斂Three-minute group presentations and assignment briefing.三分鐘小組發表與課後作業說明。
Revised brief
修正版 brief
Course Units
Each page works as a self-contained learning guide.
每一頁都是可獨立閱讀的學習講義。
Use each page as a focused reading path. Every unit keeps the same rhythm: learning goal, core concepts, tool demonstration, classroom activity, output, and checklist.
網站採獨立頁面設計，讓每個單元都能保留足夠脈絡。各單元都包含目標、概念、示範、活動、產出與檢核表。
Unit 1
From Chat to Workflow
Understand why single-turn AI use breaks under real business pressure.
理解單次聊天為什麼無法支撐真實商業流程。
Unit 2
Reusable Instructions
Write task instructions that can be reused, reviewed, and moved across tools.
寫出可重複、可檢查、可跨工具搬移的任務指令。
Unit 3
Agent Platform Logic
Define the minimum structure of a practical AI agent platform.
界定實務 AI Agent 平台的最小結構。
Unit 4
Customer Signal Case
Use a cross-industry case to reason about data, users, and business value.
用跨產業案例討論資料、使用者與商業價值。
Unit 5
Workflow Blueprint
Map trigger, inputs, tasks, outputs, and human review points.
畫出觸發情境、輸入、任務、輸出與人工審核點。
Unit 6
MVP, Risk, ROI
Constrain the first build and define how value will be measured.
縮小第一版範圍，定義如何衡量價值。
Classroom Output
Students build a practical project brief.
學員完成 project brief，不寫空泛的大系統想像。
The final brief is intentionally small. It should be concrete enough for a manager, teammate, or vendor to discuss the next step.
最終 brief 刻意保持精簡，但要具體到主管、團隊成員或外部供應商能接著討論下一步。
1. ProblemWhat business process deserves AI support?哪一個流程值得先導入 AI？
2. DataWhich sources can the agent read safely?哪些資料可以安全讀取？
3. Agent TasksWhat should the agent sort, draft, compare, or trigger?Agent 要整理、草擬、比較或觸發什麼？
4. ReviewWhere should a person approve the output?哪些輸出需要有人審核？
5. ValueHow will the team know the MVP is worth keeping?團隊如何判斷 MVP 值得保留？