跳到主要內容Skip to content
中EN
選單Menu

服務項目Service

AI 輔助程式開發：教學與企業內訓AI-assisted coding: training for teams

教工程團隊用好 Claude Code、Codex、Cursor、Copilot，也教測試、程式碼審查與資安把關；另有給主管與非工程同仁的工作坊。台北、新北、桃園、基隆可到府授課，也可以線上進行。

Hands-on training in Claude Code, Codex, Cursor and Copilot for developers, plus the testing, review and security habits that make them safe, and a workshop for managers and non-engineers. On-site in Taipei, New Taipei, Taoyuan and Keelung, or online.

預約免費 30 分鐘諮詢Book a free 30-minute consultation

為什麼團隊需要的不只是工具教學

Claude Code、Codex、Cursor、GitHub Copilot 上手只要一個下午，難的是讓團隊穩定地用好。Google 的 DORA 2025 報告指出 AI 是「放大器」：流程健全的團隊會更好，流程有問題的團隊問題也會被放大。METR 在 2025 年的隨機對照研究中，資深開發者使用 AI 工具後反而慢了約 19%，自己卻以為變快了。Veracode 的研究也發現，約 45% 的測試任務中，AI 產生的程式碼帶有安全缺陷。

所以我們的課程除了工具操作，更重視任務拆解、測試、程式碼審查、權限與密鑰管理，讓 AI 真正幫上忙，而不是製造更多需要收拾的程式碼。

課程形式與對象

① 主管與非工程同仁工作坊

對象：老闆、PM、設計、營運。時數：半天，約 3 小時，30 人以內。內容包括 AI 代理與 vibe coding 能做什麼、不能做什麼，動手做一個小原型，認識資料與資安風險，以及如何評估導入。交付講義與導入評估表。

② 工程團隊實作班

對象：開發者。時數：1 天或 2 天，每天約 6 小時，20 人以內。以練習 repo 或你指定的非機密專案實作，交付練習 repo、團隊設定範本與 PR 審查清單。

③ 導入陪跑

對象：上完實作班的團隊。時數：4 週起，每週一次工作坊加程式碼審查。用團隊的真實任務練習，一起訂出團隊 AI 使用政策，並回顧 PR 品質與成效。

④ AI 代理使用者訓練

對象：一般員工。時數：約 2 小時。說明公司的 AI 助理怎麼用、哪些資料不能貼、遇到異常怎麼回報，交付一頁式使用守則。這堂課也是AI Agent 導入的一部分。

工程團隊實作班內容

  • 工具比較與選擇：終端機代理、AI 編輯器與雲端代理各適合什麼工作。
  • 專案說明檔：用 AGENTS.md、CLAUDE.md 讓 AI 了解專案慣例與禁止事項。
  • MCP 串接：讓 AI 安全地讀取文件、議題與資料庫結構。
  • 測試先行與小步提交：先寫會失敗的測試，再讓 AI 實作。
  • AI 程式碼審查：審查清單、常見錯誤模式，以及人要負責的部分。
  • 密鑰與權限：避免把密鑰、個資貼給 AI，設定工具的權限與資料保留。

給主管與非工程同仁的 vibe coding 工作坊

「vibe coding」是用自然語言請 AI 寫程式的做法，很適合做原型與驗證想法，但不等於可以直接上線的產品。工作坊讓主管與非工程同仁親手體驗，理解它的能力與界線，回到工作上能和工程團隊用同一種語言討論 AI 導入。

資安與程式碼品質把關

每一堂工程課都包含資安與品質：AI 產生的程式碼要和人寫的一樣經過測試與審查，密鑰不進版本控制，也不貼進對話；課程也會說明各工具企業方案的資料保留與訓練設定，協助團隊依公司政策選擇。

到府授課與線上課程

台北、新北、桃園、基隆可到公司上課，不另收車馬費；其他地區或分散各地的團隊可以線上進行，價格相同。課綱會依團隊規模、技術棧與已訂閱的工具調整。

政府訓練補助怎麼看

勞動部有部分企業訓練補助方案由企業提出申請、可聘請外部講師。我們可以協助評估是否適用勞動部企業訓練補助，並提供課綱與講師資料；是否核定由主管機關決定。相關資源可參考中小企業數位轉型：政府資源與補助。

進行方式

  1. 需求訪談：團隊規模、技術棧、已訂閱或打算採用的工具。
  2. 客製課綱：確認目標、時數與練習專案。
  3. 上課：到府或線上。
  4. 課後兩週問答：上課後遇到的問題都可以提出。
  5. 導入陪跑（選購）：用真實任務持續練習與審查。

適合誰

  • 已經在用 AI 寫程式，但品質與安全沒有共同標準的工程團隊。
  • 想導入 Claude Code、Codex、Cursor 或 Copilot，希望一開始就建立好習慣的公司。
  • 想了解 AI 能做什麼，好評估導入與預算的主管與非工程同仁。

想先知道課程怎麼安排，歡迎預約免費 30 分鐘諮詢；需要導入 AI 代理的團隊，也可以參考 AI Agent 導入與AI 流程導入。

Why teams need more than tool tutorials

Claude Code, Codex, Cursor and GitHub Copilot take an afternoon to learn; using them well as a team is the hard part. Google’s DORA 2025 report describes AI as an amplifier: healthy teams get better and struggling teams see their problems grow. In METR’s 2025 randomised study, experienced developers were about 19% slower with AI tools while believing they were faster. Veracode found security flaws in AI-generated code in about 45% of the tasks it tested.

So beyond the tools, our training focuses on breaking work down, testing, code review, permissions and secrets, so AI actually helps instead of creating more code to clean up.

Formats and audiences

① Workshop for managers and non-engineers

For: owners, PMs, designers and operations. Length: half a day, about 3 hours, up to 30 people. What AI agents and vibe coding can and cannot do, building a small prototype by hand, data and security risks, and how to assess adoption. Includes handouts and an assessment worksheet.

② Hands-on course for developers

For: developers. Length: one or two days of about 6 hours, up to 20 people. Practice on a training repo or a non-confidential project you choose; you keep the repo, team configuration templates and a PR review checklist.

③ Adoption coaching

For: teams that have taken the hands-on course. Length: from 4 weeks, with a weekly workshop and code review. Practice on real tasks, write the team AI policy together and review PR quality and results.

④ AI agent user training

For: all staff. Length: about 2 hours. How to use the company AI assistant, what never to paste into it and how to report problems, with a one-page guide. This session is also part of AI agent adoption.

What the developer course covers

  • Choosing tools: what terminal agents, AI editors and cloud agents are each good for.
  • Project instructions: AGENTS.md and CLAUDE.md files that teach the AI your conventions and limits.
  • MCP: giving AI safe access to docs, issues and database schemas.
  • Test first, small commits: write a failing test, then let the AI implement.
  • Reviewing AI code: checklists, common failure patterns and what a human must own.
  • Secrets and permissions: keeping keys and personal data out of prompts, and setting tool permissions and data retention.

A vibe coding workshop for managers and non-engineers

Vibe coding means asking AI to write software in plain language. It is great for prototypes and testing ideas, but it is not the same as a product ready to launch. The workshop lets managers and non-engineers try it themselves, understand its strengths and limits, and talk about AI adoption with engineers in the same terms.

Security and code quality

Every developer session covers security and quality: AI-written code is tested and reviewed like any other, secrets never go into version control or chats, and we explain the data retention and training settings of each tool’s business plans so your team can choose in line with company policy.

On-site and online

We teach on-site in Taipei, New Taipei, Taoyuan and Keelung at no travel charge; other regions and distributed teams can join online at the same price. The syllabus is adapted to your team size, stack and the tools you already pay for.

Government training subsidies

Some Ministry of Labor training subsidies are applied for by the employer and allow external instructors. We can help you assess whether one applies and provide the syllabus and instructor details; approval is up to the authorities. See also government resources for SME digital transformation.

How it works

  1. Needs interview: team size, stack, and the tools you use or plan to adopt.
  2. Custom syllabus: goals, length and practice project agreed.
  3. Training, on-site or online.
  4. Two weeks of follow-up questions after the course.
  5. Adoption coaching (optional): continued practice and review on real work.

Who it is for

  • Engineering teams already coding with AI but without shared standards for quality and security.
  • Companies adopting Claude Code, Codex, Cursor or Copilot that want good habits from day one.
  • Managers and non-engineers who want to understand what AI can do before planning adoption and budget.

To plan a session, book a free 30-minute consultation. Teams adopting AI agents may also want AI agent adoption and AI workflow integration.

參考價格Reference prices

以下為起價或固定費用，實際金額依範圍與需求報價；免費諮詢後提供逐項報價。

Starting prices or fixed fees; the actual quote depends on scope, itemised after a free consultation.

主管與非工程同仁半日工作坊（3 小時，30 人以內）Half-day workshop for managers and non-engineers (3 hours, up to 30 people)NT$36,000 起From NT$36,000
工程師實作班 1 天（6 小時，20 人以內）Developer course, 1 day (6 hours, up to 20 people)NT$80,000 起From NT$80,000
工程師實作班 2 天（20 人以內）Developer course, 2 days (up to 20 people)NT$150,000 起From NT$150,000
導入陪跑 4 週（每週工作坊加程式碼審查）Adoption coaching, 4 weeks (weekly workshop and code review)NT$120,000 起From NT$120,000
AI 代理使用者訓練（2 小時）AI agent user training (2 hours)NT$20,000 起From NT$20,000

以上未稅。台北、新北、桃園、基隆到府不另收車馬費；線上授課同價。可協助評估是否適用勞動部企業訓練補助。

Prices exclude tax. No travel charge for on-site sessions in Taipei, New Taipei, Taoyuan and Keelung; online sessions cost the same. We can help assess whether a Ministry of Labor training subsidy applies.

自動化、AI 流程導入與 AI Agent 導入怎麼選Automation, AI workflows or AI agents?

三種服務沿著同一條光譜分工：從自動化到 AI 代理，AI 能自己做的越多，權限與人工確認的設計也越重要。三者可以搭配使用。

The three services sit on one spectrum: from automation to AI agents, the AI does more on its own, so permissions and human checks matter more. They can be combined.

自動化工作流Workflow automationAI 流程導入AI workflow integrationAI Agent 導入AI agent adoption
核心What it is規則固定的跨系統資料流Rule-based data flows between systems在流程的單一步驟加入 AI 判斷AI judgement inside one step of a workflow能對話、自己決定步驟並使用工具的代理An agent that converses, decides its steps and uses tools
例子Example訂單進來 → 寫入試算表 → 通知負責人New order → added to a spreadsheet → owner notified來信分類、擷取報價單、草擬回覆、知識庫問答Sorting email, extracting quotes, drafting replies, knowledge-base answers同事在 LINE 問訂單狀態，代理查 ERP 回覆並草擬通知A colleague asks about an order in LINE; the agent checks the ERP and drafts a notice
風險與把關Risk and safeguards低：錯誤通知與監控Low: error alerts and monitoring中：重要輸出由人確認Medium: people check important output高：隔離、最小權限、高風險動作人工核准High: isolation, least privilege, human approval for risky actions

  1. 自動化工作流Workflow automation

    核心What it is
    規則固定的跨系統資料流Rule-based data flows between systems
    例子Example
    訂單進來 → 寫入試算表 → 通知負責人New order → added to a spreadsheet → owner notified
    風險與把關Risk and safeguards
    低：錯誤通知與監控Low: error alerts and monitoring

  2. AI 流程導入AI workflow integration

    核心What it is
    在流程的單一步驟加入 AI 判斷AI judgement inside one step of a workflow
    例子Example
    來信分類、擷取報價單、草擬回覆、知識庫問答Sorting email, extracting quotes, drafting replies, knowledge-base answers
    風險與把關Risk and safeguards
    中：重要輸出由人確認Medium: people check important output

  3. AI Agent 導入AI agent adoption

    核心What it is
    能對話、自己決定步驟並使用工具的代理An agent that converses, decides its steps and uses tools
    例子Example
    同事在 LINE 問訂單狀態，代理查 ERP 回覆並草擬通知A colleague asks about an order in LINE; the agent checks the ERP and drafts a notice
    風險與把關Risk and safeguards
    高：隔離、最小權限、高風險動作人工核准High: isolation, least privilege, human approval for risky actions

合作流程How we work

  1. 聊需求Talk it through

    了解你的目標、預算與時程，免費提供初步建議。

    We learn your goals, budget and timeline, and give initial advice for free.

  2. 規劃與設計Plan and design

    確認功能範圍與畫面設計，提供明確報價與時程。

    We agree on scope and screens, then give a clear quote and schedule.

  3. 開發與測試Build and test

    分階段交付，每個階段都看得到實際成果。

    We deliver in stages, so you see real progress at every step.

  4. 上線與維運Launch and support

    協助正式上線，並提供維護、更新與數據追蹤。

    We handle the launch, then keep things maintained, updated and measured.

相關作品Related work

概念作品用於展示設計與開發能力，並非真實客戶案例。

Concept work shows our design and build skills; it is not client work.

常見問題Frequently asked questions

學員需要會寫程式嗎？Do participants need to know how to code?

主管與非工程同仁工作坊、AI 代理使用者訓練不需要；工程團隊實作班與導入陪跑是給開發者的。

Not for the managers’ workshop or the AI agent user training; the hands-on course and adoption coaching are for developers.

教哪一個工具？Which tool do you teach?

以公司已訂閱或打算採用的工具為主，例如 Claude Code、Codex、Cursor 或 GitHub Copilot。重點放在跨工具共通的方法：任務拆解、測試、審查與資安；工具更新很快，課綱會隨版本調整。

Whatever your company already pays for or plans to adopt, such as Claude Code, Codex, Cursor or GitHub Copilot. The focus is on methods that carry across tools: breaking work down, testing, review and security. Tools change fast, so the syllabus is updated with them.

公司程式碼會外流嗎？Will our code leak?

課程會說明各工具企業方案的資料保留設定、如何避免把密鑰與個資貼給 AI，並以練習 repo 或你指定的非機密專案上課。

We cover each tool’s business-plan data retention settings and how to keep secrets and personal data out of prompts, and we teach on a training repo or a non-confidential project you choose.

可以到公司上課嗎？線上可以嗎？Can you teach on-site? Online?

台北、新北、桃園、基隆可到府授課，不另收車馬費；其他地區線上進行，價格相同。

Yes. On-site in Taipei, New Taipei, Taoyuan and Keelung with no travel charge; elsewhere we teach online at the same price.

費用大概多少？可以申請補助嗎？Roughly how much is it? Are subsidies available?

本頁的「參考價格」列出各課程的起價，實際依人數、時數與客製程度報價。部分勞動部企業訓練補助由企業申請、可聘外部講師，我們可以協助評估是否適用並提供課綱，但是否核定由主管機關決定。

The reference prices on this page list starting prices for each format; the quote depends on headcount, hours and customisation. Some Ministry of Labor subsidies are applied for by the employer and allow external instructors; we can help assess eligibility and provide the syllabus, but approval is up to the authorities.

想聊聊你的需求？Want to talk it through?

先約 30 分鐘免費諮詢：聊目標、範圍與時程，我們會給你初步建議。

Start with a free 30-minute call about your goals, scope and timeline, and get our initial advice.

預約免費 30 分鐘諮詢Book a free 30-minute consultation