認證與學習路線Certifications and learning path
這一頁回答兩個問題：證照值不值得考、該考哪一張；以及從這門課出發，怎麼安排一份 12 週的學習計畫。
This page answers two questions: are certifications worth it, and which one; and how to plan 12 weeks of study starting from this course.
以下證照資訊依各家官方頁面整理，查詢日期為 2026-10-14。證照的名稱、考試版本、費用與資格變動很快（例如 2026 年 AWS、Microsoft 都改版或淘汰了好幾張 AI 證照），報名前請一定再到官方頁面確認。官方頁面沒有寫的欄位，這裡標示「官方未列」，不用第三方網站的數字補。
The certification details below come from each vendor's official pages, checked on 2026-10-14. Names, exam versions, fees and eligibility change quickly (AWS and Microsoft both revised or retired several AI certifications in 2026), so always re-check the official page before registering. Fields the official pages don't state are marked "not listed" rather than filled from third-party sites.
先講立場：證照什麼時候值得考First, a position: when certifications are worth it
對 FDE、AI 應用工程師這類職位，證照的分量通常不如作品（見作品集專案）。它證明你「知道」某些東西，但不證明你「做得出來」。不過，在下面幾種情況，證照有實際用處：
For FDE and AI application engineering roles, certifications usually carry less weight than projects (see portfolio projects). They show you know something, not that you can build it. They do help in a few situations:
客戶綁定某家雲端Customers on one cloud
你的目標產業（例如半導體、製造業）的客戶主要用 AWS、Azure 或 Google Cloud，證照能證明你懂他們的基礎架構。
If your target industry's customers mostly use AWS, Azure or Google Cloud, a certification shows you understand their infrastructure.
履歷還沒有 AI 經歷No AI experience on your CV yet
轉職時，證照可以當成「有系統學過」的門檻證明，讓履歷通過第一關篩選。
When changing careers, a certification proves structured study and can get a CV past first-round screening.
公司或合作夥伴要求Employer or partner requirements
有些顧問公司的合作夥伴資格，要求一定人數的認證人員（例如 Claude Partner Network）。
Some consulting partner programs require a number of certified staff (the Claude Partner Network, for example).
用考試大綱當學習地圖Using the syllabus as a map
就算不考，考試大綱也是很好的「該學什麼」清單。
Even if you never sit the exam, the syllabus is a good "what to learn" checklist.
不建議為了收集證照而犧牲做專案的時間。一張證照加一個有評測數字的作品，比三張證照有說服力。
Don't trade project time for collecting certificates. One certification plus one project with eval numbers beats three certifications.
主要的 AI 相關證照Main AI-related certifications
AWS
| 證照Certification | 適合For | 考試Exam | 費用Fee |
|---|---|---|---|
| AWS Certified AI Practitioner | 入門：驗證 AI/ML 與生成式 AI 的基礎知識Foundational AI/ML and generative AI knowledge | 90 分鐘、65 題90 min, 65 questions | USD 100 |
| AWS Certified Machine Learning Engineer – Associate | ML 模型的開發與上線Building and operating ML in production | 改版中：英文版 MLA-C01 已於 2026-09-28 停考；新版 MLA-C02 於 2026-09-29 開始 beta（170 分鐘、85 題），預計 2027-01-14 正式推出Being revised: English MLA-C01 ended 2026-09-28; MLA-C02 beta from 2026-09-29 (170 min, 85 questions), GA expected 2027-01-14 | beta USD 75beta USD 75 |
| AWS Certified Generative AI Developer – Professional | 進階：建置生成式 AI 應用。官方建議 2 年以上的開發經驗，加上 1 年生成式 AI 實作經驗Advanced: building generative AI applications. AWS recommends 2+ years building production apps plus 1 year of hands-on generative AI | 180 分鐘、75 題180 min, 75 questions | USD 300 |
另外，AWS 的 Solutions Architect – Professional（SAP-C03）與 Developer – Associate（DVA-C03）新版本也加入了生成式 AI 與 AI 安全的內容，2026 年 11、12 月開始提供。AWS's new Solutions Architect – Professional (SAP-C03) and Developer – Associate (DVA-C03) versions also add generative AI and AI security content, available from November and December 2026. AWS Training & Certification blog, 2026-09
Google Cloud
| 證照Certification | 適合For | 考試Exam | 費用Fee |
|---|---|---|---|
| Generative AI Leader | 任何職務，不需技術背景；偏商業與決策Any role, technical or not; business and decision focus | 90 分鐘、50–60 題；效期 3 年；英、日、西、葡文90 min, 50–60 questions; valid 3 years; EN, JA, ES, PT | USD 99 |
| Professional Machine Learning Engineer | 進階：官方建議 3 年以上業界經驗，其中 1 年以上在 Google Cloud；最新版本已改為以 Gemini Enterprise Agent Platform 為主Advanced: 3+ years industry experience, 1+ on Google Cloud; latest version centres on the Gemini Enterprise Agent Platform | 2 小時、50–60 題；英、日文2 hours, 50–60 questions; EN, JA | USD 200 |
Microsoft Azure
| 證照Certification | 適合For | 考試Exam | 費用Fee |
|---|---|---|---|
| Azure AI Apps and Agents Developer Associate（AI-103） | 中級：用 Python 與 Microsoft Foundry 建置生成式 AI 與 agent 應用；取代舊的 AI-102Intermediate: generative AI and agent apps with Python and Microsoft Foundry; replaces the older AI-102 | 120 分鐘；提供繁體中文120 min; available in Traditional Chinese | 依考試地區而定Varies by country |
Databricks
| 證照Certification | 適合For | 考試Exam | 費用Fee |
|---|---|---|---|
| Generative AI Engineer Associate | 在 Databricks 上建置 RAG 等生成式 AI 應用；官方建議 6 個月以上實作經驗。考試範圍包含「評測與監控」（12%）Building RAG and other generative AI apps on Databricks; 6+ months hands-on recommended. Includes "Evaluation and Monitoring" (12%) | 90 分鐘、45 題計分；效期 2 年90 min, 45 scored questions; valid 2 years | USD 200 |
Anthropic（Claude）
Anthropic 有四張 Claude 證照：Associate: Foundations、Developer: Foundations、Architect: Foundations、Architect: Professional，考試皆為監考形式。但目前只開放給「Claude Partner Network」成員組織的人員報考，個人無法直接報名。合作夥伴網路是給「組織」加入的，加入免費。考試時間與費用，官方部落格未列。
Anthropic offers four Claude certifications: Associate: Foundations, Developer: Foundations, Architect: Foundations and Architect: Professional, all proctored. However, exams are currently available only to members of the Claude Partner Network, so individuals can't register on their own. The partner network is joined by organisations, and membership is free. Exam length and fees aren't listed on the official blog.
如果你的公司不是合作夥伴，可以先用 Anthropic Academy 的免費課程學習（課程完成證明不等於正式證照）。
If your employer isn't a partner, Anthropic Academy's free courses are a good start (course completion certificates aren't the proctored credential).
Claude blog: Four role-based certifications (2026-07-23) · Claude Academy FAQ
怎麼選How to choose
| 你的情況Your situation | 建議Suggestion |
|---|---|
| 目標客戶主要用某一家雲端Target customers use one cloud | 考那一家的 AI 開發類證照（AWS Generative AI Developer、Azure AI-103、Google PMLE）That vendor's AI developer certification (AWS Generative AI Developer, Azure AI-103, Google PMLE) |
| 還沒決定雲端，想先建立基礎No cloud chosen yet, want foundations | 先讀完本課程中級篇、做一個作品；證照可以晚點再考。真的需要入門證照，AWS AI Practitioner 或 Google Generative AI Leader 範圍較廣、費用較低Finish this course's intermediate track and build a project first; certify later. If you need an entry certificate, AWS AI Practitioner or Google Generative AI Leader are broad and cheaper |
| 希望用中文應考Want to take the exam in Chinese | Azure AI-103 有繁體中文版本Azure AI-103 is offered in Traditional Chinese |
| 客戶資料平台是 DatabricksCustomers' data platform is Databricks | Databricks Generative AI Engineer AssociateDatabricks Generative AI Engineer Associate |
| 在 Claude 合作夥伴公司工作You work at a Claude partner | Claude Developer 或 Architect 系列Claude Developer or Architect track |
證照範圍和本課程的對應How certification topics map to this course
各家 AI 開發類證照的共同主題，本課程大多有涵蓋。差別在於證照會考該家雲端的特定服務，那部分要另外讀官方教材。
Most topics shared by the AI developer certifications are covered here. The difference is that exams test each vendor's specific services, which you'll need to study from official material.
| 常見考試主題Common exam topic | 本課程章節Course chapters |
|---|---|
| prompt 設計、結構化輸出Prompt design, structured output | B3、I2、04 |
| RAG、資料準備、向量搜尋RAG, data preparation, vector search | B4、I3、05 |
| agent、工具呼叫Agents, tool use | I4、06 |
| 評測與監控Evaluation and monitoring | I6、07、08 |
| 部署、安全、治理、成本Deployment, security, governance, cost | I5、I7、08 |
12 週學習計畫A 12-week study plan
以每週 5 到 8 小時估算。已經有程式基礎的話，第 1 週可以很快帶過。重點是每個階段都有做出來的東西，不是只讀完。
Assumes 5–8 hours a week. With programming experience, week 1 goes quickly. The point is that every phase ends with something built, not just read.
| 週Weeks | 讀Read | 做出來的東西What you build |
|---|---|---|
| 1 | 入門篇 B1–B5（有基礎可略讀）Basics B1–B5 (skim if experienced) | 跑完 labs/beginner 的四個程式Run the four labs/beginner scripts |
| 2–3 | 中級篇 I1–I4Intermediate I1–I4 | 完成各章練習；用 --fake 與（可選）--ollama 跑機器人Chapter exercises; run the bot with --fake and optionally --ollama |
| 4 | 中級篇 I5–I7Intermediate I5–I7 | 把機器人部署成網站，取得第一個公開網址Deploy the bot as a website: your first public URL |
| 5–6 | 作品集專案Portfolio projects | 作品 1：換成自己的資料，寫 20 題評測，照 README 範本整理Project 1: your own data, 20 eval questions, README per the template |
| 7–9 | 進階篇 01–07Core 01–07 | 作品 2（主力作品）：最好結合你的專業領域Project 2 (your main piece): ideally drawing on your domain expertise |
| 10 | 進階篇 08–09Core 08–09 | 作品 2 加上監控、成本估算、模型比較（雲端 vs 本機）Add monitoring, cost estimates and a cloud-vs-local model comparison to project 2 |
| 11–12 | （選擇性）證照官方教材與模擬考(Optional) official certification material and practice exams | 報考一張與目標客戶相關的證照；或把時間拿來完善作品與撰寫技術文章Sit one certification relevant to your target customers, or polish your projects and write a technical article instead |
12 週結束時，你會有：兩個附評測數字的作品（其中一個有公開網址）、一份讀完的課程，以及（如果選擇考）一張證照。這三樣放在一起，比任何單一項目都有說服力。
After 12 weeks you'll have two projects with eval numbers (one with a public URL), a completed course, and optionally a certification. Together they're more convincing than any one alone.