跳到主要內容

All or Nothing: The Real Impact of AI

This is a short conclusion as of February 2026.

In the AI era, senior people (those with ability and manpower) do not need to worry. Because human ability and experience are combined skills that AI still cannot replace.

I still say the same thing:

“Engineers will definitely still have work. Frontend will become even more important. People who build applications will never be unemployed. The only question is who you build for. Demand will only increase.”

Now I will share three key points I see.


Changes in Company Structure

I believe there are several direct impacts and results.


All or Nothing

All

If a company does not have people with this new ability, it will go backward. Such companies will first declare that “agents are useless.” The company structure and staff will still exist. People will still receive salary. Everything looks normal.

But you will realize: The company making money has nothing to do with you.

Nothing

Entire departments may be removed.

Simply put: Now one person can function as a whole team.

There may no longer be a marketing department. Instead, one marketing leader manages many marketing agents.

One person can work like ten people. Ten people can equal what used to be a 100-person company. Revenue can reach hundreds of millions. Everyone is happy.


The “Efficiency First” Myth

“Efficiency” alone becomes a fake topic.

Increasing speed is not enough. What matters is results.

Before, people analyzed data, ran ads, handled digital tasks. Now AI agents can do all of that — and do it well.

If a human’s only role is to “take the blame,” then that is all they will end up doing.


Willing to Believe

You must believe first to move forward. You must believe before you can see.

Unless you have already seen it.

When belief is strong enough, making money and expanding becomes natural.

In the past, expansion meant growing the team. Now expansion means who can efficiently use tokens and burn tokens.


Antifragile Thinking

Being antifragile means acting now causes less loss than doing nothing.

Back to the main point:

“Whoever can continuously burn enough tokens efficiently will win.”

While you hesitate over spending a few hundred or a few thousand dollars to register something, many companies are already quietly burning tokens.

They are training their digital consultants, digital assets, and digital systems. They are feeding agents as digital nutrients to every key person — unless there is an extremely serious data leak risk.

Doing something loses less than doing nothing.

This game must be top-down.

You can add many agents, add AI, build teams — but execution speed always gets stuck at people.

So AI adoption must be top-down.

Maybe 1% can succeed bottom-up, but that is too slow.

As people say: The ones responsible decide the speed of execution.

Leaders must personally do it with the team.


“Digital Work” Becomes Result-Based

All digital work — writing, media, design, research — will be judged by results.

More directly: Vertical focus is better.

Focus deeply. Focus enough to build your moat.

Like what we are doing with cymkube.com — vertical, real, manufacturing. AI cannot replace this quickly.

We believe in:

“Software defines manufacturing.”

These processes are not about influence, reach, or views. Those are only support metrics.

AI can collect and organize those easily.

The only true metric is:

Results.

Workers will not disappear, but their roles will shift.

Fewer people are needed in digital processes, but a small number of responsible humans (who take the blame) are still required.


One-Person Companies and Super Individuals

Next, more people — especially in aging societies — will move toward one-person work models.

As I said earlier:

We lack people with expertise, experience, and networks.

These people are senior. They will not easily work full-time for one company. Startups cannot afford them full-time.

Work will shift to:

• One-person companies • Outsourcing models

Before, these experts worked full-time inside companies. Now one person can serve multiple companies.

Their ability becomes distributed. Phase-based. Spread across multiple businesses.

This trend is already visible.

The rise of the super individual.


Conclusion

There is no final conclusion.

AI is still evolving.

留言

這個網誌中的熱門文章

Vibe Coding:為什麼 Junior 更快上手?Senior 要如何追趕?

現象層面(市場觀察) 最近有篇文章討論 junior & senior 開發者在 AI 時代的角色轉變,非常熱門。 身為 Cympack 產品開發團隊 ,我們也一直關注這個議題,在閱讀這篇文章時觀察到一些有趣的現象,對我們來說,這正好反映出 AI 正在改變開發生態,junior 借力 AI 快速成長、senior 則需要在 「架構思維」 與 「多 agent 協作」 中找到新定位,其中有些啟發(insight) 可以跟大家分享。 為什麼 Junior 更容易上手 vibe coding? 心智負擔低 → Junior 沒有太多傳統 code workflow 的框架包袱 敢於嘗鮮 → Gen Z / 年輕工程師天生習慣用 prompt-based 工具、跟 LLM 互動 少「優雅程式設計」的束縛 → 不太糾結「這樣寫會不會不夠優雅」,反而 embrace 快速迭代、快速出成果 反觀 Senior: 熟悉大型系統設計 有豐富的「工程正統流程」知識(架構設計、測試策略、效能優化、設計模式) 對 AI 生成 code 的品質 / 維護性通常比較保留 部分 10+ 年資深工程師,對 prompt engineering 沒那麼熟練,還在觀望 技能面(未來的關鍵能力) Vibe coding 本質上 = prompt engineering + AI co-pilot 管理能力 能力項目 誰目前比較有優勢? Prompt 撰寫 / AI 互動 Junior 較強(熟悉 chat-based 流程) 系統設計 / 架構把關 Senior 較強 AI 生成 code 驗證 / Bug 察覺能力 Senior 較強(能看出潛在問題) 快速疊代 / Hackathon 式開發 Junior 較強 長期維護性 / 穩定性 Senior 較強 總結 Junior 確實更快適應 vibe coding,並且更習慣以 「chat-based coding」 的工作流開發。 Senior 擁有驗證 AI 產物與系統設計的深度能力,但若不主動練習 vibe coding,長期會逐漸落後於新一波開發潮流。 就如同在 GAI 技術年會分享,希望帶給各位的感受, 『與 AI 協...

Vibe Coding 協作到自建 Dev Agent?從 Claude / Codex 到 OpenHands

過去一年,越來越多工程師開始 把 AI 真正帶進工作流程 。從一開始用 ChatGPT、Claude 來問語法問題,到後來很多人愛上 Cursor,直接在編輯器裡讓 AI 幫忙改 code、補 test case、甚至自動整理 PR。這樣的開發體驗,已經大大改變了我們寫程式的方式。 更現實的是,在很多企業內部、政府單位、或涉及機密資料的專案裡, 其實根本不能直接用 Cursor 或雲端 LLM 工具。   畢竟這些服務通常會把資料傳到雲端模型做處理,萬一專案裡有未公開的技術、敏感客戶資料,或是受限於法規 (像金融、醫療、政府標案) ,直接用雲端 AI 工具就會踩 紅線 。  因此,許多團隊反而更希望 「自己架一套 Dev Agent」 ,可以在內網執行,資料完全掌握在自己手上,該整合的內部工具、該讀的私有 repo、該串的 CI/CD pipeline,全部客製化、安全可控。 這時候,像 OpenHands 這樣的開源 Dev Agent 框架就特別有價值。它的出發點不是單純的 AI 助手,而是讓你能夠打造出一個真的可以跑在自己環境裡、可以理解整個開發流程的 AI 工程師。從建置到部署,從 CLI 操作到瀏覽器查詢, 從多檔案編輯到自動測試,全部都能自己完成,甚至還能針對不同專案調整專屬的工作流。 對很多開始探索 AI 協作開發的團隊來說,這是一條 從 「AI 幫你寫一段程式」,走向「AI 幫你解決一整個任務」 的進化路徑。而且,還是在可控、可自定義、安全的環境裡完成的。 🧩 主要概述 OpenHands 是由 All‑Hands AI 開發的開源「軟體開發代理人平台」,能模仿人類工程師從建立程式、修改程式碼、執行指令,到瀏覽網頁、呼叫 API……等一整套開發流程 它提供雲端(OpenHands Cloud)與本地 Docker 運行版本,用戶能配置 LLM(如 Claude、OpenAI、Gemini…) 📚 核心特性與怎麼使用 代理人的工具能力 支援代碼編輯、命令行、執行環境、網頁瀏覽、API 呼叫—接近人類開發者完整技能。其中 OpenHands Cloud 版本提供 $50 試用額度讓大家方便使用,又或者如果自己本機有 docker 的話,可以自己Local 版本透過 Docker 自架環境。 ...

Google Gemini 全端 AI Agent 快速入門 - 打造「思考」的 AI 助理

一套從搜尋、反思到輸出的全端 AI 代理人範例,讓你看懂什麼叫 Research Agent 在 AI 工具百家爭鳴的今天,大家都在問一個問題: 「我能不能不只問 AI 答案,而是讓它像一位助理一樣,有流程、有反思、還有出處,真正幫我完成一件事?」 Google 最近釋出了一個相當具有指標意義的開源專案 gemini-fullstack-langgraph-quickstart ,正是為了解這個問題而誕生。 這套系統到底是什麼? 這個範例不是傳統 Chatbot,而是展示一個完整的 AI research agent : 它會根據使用者的提問,自動發想搜尋關鍵字、查資料、整合重點,最後給出答案還附上引用來源。背後的邏輯設計得非常扎實,不只是能跑,更是具備可讀性、可擴展性與可商用性。 它的流程大致如下:  1. 使用者輸入問題(例如:「抖音是否影響台灣選舉?」)  2. Gemini LLM 幫你想出關鍵字(不只是照抄問題)  3. 呼叫 Google Search API 抓資料   4. LangGraph 控制流程 → 判斷資料夠不夠 → 若不足,自動補查  5. 整合最終答案,並產生 citation(來源說明) 你可以想像這就像一位實習助理幫你寫報告, 不只輸出一段內容,而是會 去查、會判斷、會補資料,而且說明「我為什麼這樣說」 。 LangGraph 是什麼角色? LangGraph 就是整個 Agent 背後的控制系統 。 用白話講,它幫你定義 AI 每一步要幹嘛、遇到什麼狀況該走哪條路、要不要反思、要不要再查,甚至可以定義條件邏輯與資料流動。 這就不像寫一個單純的 Chat API,而是比較像「把一個流程圖變成可以跑的程式」。 對工程師來說,它提供了從 prompt 到流程控制的設計彈性;對產品設計來說,它讓 AI 有了 「多步驟任務執行」 的能力。 技術架構與使用方式 這整套系統是 Fullstack 架構,前後端都幫你整好了,技術選型也非常實用:   前端:Vite + React + TailwindCSS + Shadcn UI  後端:FastAPI + LangGraph...