Before 2025, most of my work was related to design, digital fabrication workflows, and robotic arms.

I helped companies set up robotic fabrication labs, introduced new manufacturing methods and materials, and thought carefully about how to integrate these processes into existing factory operations.

During that time, what I learned was not only technical skills. More importantly, I learned many implicit system-level ideas.

The real difficulty was never the technology itself, but understanding:

• how the original process actually worked
• which parts would be affected first when a new system was introduced
• whether a blockage in one step would impact the entire workflow

These questions are usually not written in specifications. But if they are not thought through early, problems almost always appear later.

By 2024, I began to notice some ideas that were still not fully formed.

I realized that I was no longer only interested in whether a robotic arm could do something, but more curious about how an entire organization actually operates.

During a gap period in my career, I had opportunities to participate in technology-related discussions and collaborations in different contexts. Sometimes it was product design discussions, sometimes helping with development decisions, and sometimes writing code myself.

What surprised me during that time was that many of the process principles I learned on manufacturing floors could be applied to completely different industries.

More accurately, they shared similar underlying logic.

Later, as I worked with organizations at different levels of maturity, I noticed a common pattern across many companies I encountered:

they rely heavily on multiple software tools, processes, and human judgment, while many operational workflows still depend on Excel, manual consolidation, and experience-based knowledge transfer.

At first, I was not sure why I was drawn to this kind of environment. But through working and talking with them, I slowly realized one thing:

the "process requirements" users describe are often not what they truly need.

Instead of only listening to how they explain their processes, I became more interested in observing how they actually work, how they interact with existing workflows, and how they use the minimum viable models (MVPs) I build.

This approach strongly connects to what I learned from manufacturing systems in the past.

Because of this, throughout 2025, I spent a lot of time repeatedly asking myself these questions:

• Does this process really need automation?
• If we automate it now, will it actually make things more complex?
• Are there parts that should remain human-driven for now, instead of being handed to a system too early?

At the same time, starting in 2024, I also experimented with integrating AI into both hardware and software automation workflows.

However, after a full year of hands-on experience in 2025, I noticed something slightly counterintuitive.

For many small and medium-sized businesses, their existing processes are not yet systematic enough.

In this situation, introducing AI often brings limited benefits and can even create more doubt and distrust inside the organization.

What these companies truly need is often not "AI first," but a stable and understandable base process.

Only after that process has run for some time, and has been understood and trusted, can AI become a helpful support — rather than an extra burden.

在 2025 年之前，我的工作內容大多和設計、數位製造流程、機械手臂有關。

我會幫公司建立機器人製造實驗室，引入新的製造方法與材料，思考怎麼把這些流程放進工廠既有的運作方式裡。

這段時間，我學到的其實不只是技術。比較多是一些很隱性的系統觀念：

困難的地方，從來不在於技術本身，而是在於理解

• 原本的流程是怎麼運作的
• 新的系統加進來之後，哪裡會先被影響？
• 如果其中一個環節卡住，會不會牽動整條流程？

這些問題，通常不會寫在規格書裡，但如果一開始沒想清楚，後面幾乎一定會出問題。

到了 2024 年，我開始察覺一些尚未成形的想法。我發現自己已經不只是對「機械手臂能不能做」有興趣，而是對「整個組織是怎麼運作的」這件事更好奇。

在職涯的一個 gap 期間，我有機會在不同情境中參與一些科技相關的討論與協作。有時是一起討論產品設計，有時是協助開發決策，也會實際寫一些程式。

那段時間讓我有點意外的是，以前在製造現場學到的流程觀念，其實可以套用到完全不同的產業。或者更準確地說，它們背後有相似的底層邏輯。

後來我接觸到更多不同成熟度的組織，讓我意識到，很多我接觸過的公司都有類似的狀況：

高度依賴多種軟體、流程和人工判斷，很多營運流程仍然停留在 Excel、手動整理、經驗傳承。

一開始我也不確定自己為什麼會對這樣的環境感興趣。但在實際和他們共事、談話的過程中，我慢慢發現一件事：使用者告訴我的「流程需求」，很多時候並不是他們真正需要的東西。

與其一直聽他們描述流程，我反而更想觀察他們怎麼實際操作、怎麼和流程互動，以及怎麼使用我建立的最小可行模型（MVP）。這件事，和我過去在製造系統中學到的觀念，有很強的連結。

因此，整個 2025 年，我花了很多時間反覆思考這些問題：

• 這個流程，真的需要自動化嗎？
• 如果現在就做，會不會反而讓事情變得更複雜？
• 有些地方，是否應該先讓人來處理，而不是急著交給系統？

同一時間，從 2024 年開始，我也嘗試把 AI 放進硬體與軟體的自動化流程中。但經過 2025 一整年的實際經驗，我發現了一件有點反直覺的事。

對很多中小型企業來說，現有流程本身還不夠有系統。在這種情況下導入 AI，不但效益有限，還可能在組織內部引發更多懷疑與不信任。

這些公司真正需要的，往往不是「先有 AI」，而是先建立一個能穩定運作的基礎流程。等這個流程跑過一段時間、被理解、被信任之後，AI 才有可能成為有幫助的輔助，而不是額外的負擔。