You think you're training AI, but you're actually drafting an authorizing constitution.

You think you're training AI, but you're actually drafting an authorizing constitution.

by | 4 month 6, 2026 | Blog

In 2025, a marketing team of a mid-sized e-commerce company in Taiwan had five people, taking turns managing three social media accounts. Passwords were stored in a shared Google Sheet, accessible to everyone. No one saw anything wrong with this until they decided to implement an AI agent to automate posting. Their first instinct was: "Just give the account and password to the agent." This is an instinct almost every company starting to use AI has. And it's almost always wrong. Before training the agent, ask a question. Most people talk about AI agents in terms of capabilities: Can it search? Can it write reports? What tools can it connect to? These questions are all correct, but they are secondary questions. The primary question is: This...
The lessons learned from the lobster craze, and what it didn't tell you.

The lessons learned from the lobster craze, and what it didn't tell you.

by | 2 month 27, 2026 | News

I. The Walking USB Drive is Quietly Leaving the Company. There's a type of employee every company has, but few realize how important they are. They're not engineers, nor are they sales personnel. They're the ones who "know a little about every system." They investigate ERP problems, export CRM data, piece together reports the boss needs using Excel, and even reconcile factory MES data with the financial system's figures. This person has a less-than-flattering name: the Walking USB Drive. Their value lies not in their intelligence, but in being the sole bridge between various systems. ERP has its language, CRM has its logic, HRM, POS, MES...
You think you're lacking [tools], but actually you're lacking [architecture].

You think you're lacking [tools], but actually you're lacking [architecture].

by | 2 month 27, 2026 | Blog

The wishing well has truly appeared. In the old days of software development, there was a joke among engineers: "If only there were a wishing well where requirements could be fulfilled spontaneously." Every new requirement required communication with the PM, evaluation by a systems analyst, scheduling in a sprint, waiting three months, revisions, and more waiting. Every "addition/modification/deletion" was a war of attrition. Now, the wishing well has truly arrived. AI is shortening the distance between "stating a requirement" and "implementing a requirement" from three months to three days. Vibe Coding, AI Agents, and automated workflows are doing something unprecedented: lowering the barrier to logical expression to almost zero. You just need to clearly state what you want, and AI can help you achieve it. ...
SaaS is not dead, but it is no longer "reliable".

SaaS is not dead, but it is no longer "reliable".

by | 2 month 27, 2026 | Blog

Last week's sharp decline in US stocks was partly due to the release of Claude Opus 4.6 and OpenAI Codex-5.3. These weren't just model upgrades; they made many people realize for the first time that a large number of "functional SaaS" services are losing their competitive advantage. This doesn't mean all SaaS will die immediately, but rather that a certain type of SaaS has entered a high-risk zone. Services like Jotform, DocuSign, Monday, and Ragic, which focus on "processes, forms, approvals, and management interfaces," essentially encapsulate requirements → rules → processes → UI → data into a usable product. Now, AI is doing this, and the time...
When AI turns you into a "one-person company," your team is quietly crumbling.

When AI turns you into a "one-person company," your team is quietly crumbling.

by | 12 month 9, 2025 | Blog

Last week at a gathering, a friend who works in trade told me, "Vincent, we have a young salesperson in our company who uses AI to organize customer data, write outreach emails, and create quotations—he's three times more efficient than everyone else. But now he hardly communicates with his colleagues anymore, saying, 'Anyway, asking AI is faster than asking people.'" He then chuckled bitterly, "What's worse, seeing him like this, some other salespeople are starting to get anxious and want to learn AI too, while others are simply giving up, saying, 'AI is unreliable and will make mistakes! It often gives nonsensical answers with a straight face, so just trust it.' Now in meetings, it feels like everyone's on a different wavelength." This isn't an isolated case; it's a collective phenomenon happening now. The market is flooded with courses teaching you: "Learn AI, transform into a one-person company!", "Use ChatGPT..."