Reading a Wall Street Journal article on the AI-driven always-on economy, I stopped and thought for a long while. Not because the article was so well written, but because it made me realize something: we have been using the wrong framework to understand this change.
The article’s narrative goes like this — AI lets the economy operate uninterrupted around the clock, and this is a brand-new era.
But is it really “brand-new”?
The always-on economy was never new
Convenience stores in Taipei were already open 24 hours twenty years ago. Global financial markets trade in relay across time zones, from Tokyo to London to New York; the sun never sets on the exchanges. Semiconductor fabs run their production lines 365 days a year without stopping, because the cost of restarting a furnace is higher than letting it keep running.
We have long lived inside an always-on economy.
The difference is that, in the past, the always-on ran on human lives.
I remember when I first started my company, there was a period when we were running a cross-time-zone project. Taiwan handled the Asian side during the day, and at night we took on the demands of American clients. It sounds very international, but in reality it was just me and a few people on the team taking turns staying up all night, never daring to put our phones on silent. Once, at three in the morning, I was woken up to handle a system alert; I investigated for two hours, only to find it was a false alarm. But you don’t dare not to check, because “what if it’s real?”
The cost of those days wasn’t the overtime on the books. It was the team’s eyes growing wearier, the brainstorming meetings growing quieter, everyone conserving energy and dealing only with whatever was most urgent in front of them.
This is the truth of the old version of the always-on economy: using the physiological limits of human beings to brute-force a system that exceeds physiological design.
AI has restructured the cost logic of the always-on economy
After AI enters the picture, around-the-clock operation itself hasn’t changed, but the cost structure of maintaining it is fundamentally different.
Here’s a concrete example. My team now uses AI agents to handle frontline customer-request triage and system monitoring. In the past, a false alarm at three in the morning required someone to get up and verify it; now the AI first judges the severity, and a notification is triggered only when the situation truly requires human intervention. The result? Genuinely urgent incidents get faster responses, because the person on duty is at full energy rather than already drained by five prior false alarms.
Compared to stories like “30% efficiency improvement,” this is a qualitative change.
When you push the marginal cost of maintaining around-the-clock operations close to zero, a lot of things that previously didn’t make sense suddenly do. Monitoring the status of every supply chain in real time? In the past that took an entire department; now a single AI agent can do it. Real-time risk assessment on every transaction? In the past only large financial institutions could afford it; now small and medium-sized enterprises can too.
But there’s a trap here: many people think AI simply means turning what humans used to do into something machines do. This is the most superficial understanding, and the most dangerous strategy.
Three-layer restructuring: not an upgrade, but a renovation
In observing how different companies adopt AI, I’ve noticed a pattern: the ones that succeed are not those that “use AI to replace human labor,” but those willing to rethink their entire operating logic.
This rethinking has three layers.
The first layer is process restructuring. Traditional business processes are designed around “human working hours.” Meetings on Monday, reports due Wednesday, review on Friday. This rhythm made sense in the human-driven era, because people need time to digest information, form judgments, and coordinate action. But when AI can process and analyze in real time, this rhythm becomes an artificial bottleneck. True process restructuring isn’t about “changing the weekly meeting frequency to daily,” but about asking: “Do we really still need meetings at a fixed frequency? Or can we switch to event-driven — gather when there’s a situation, and otherwise each push forward on our own?”
The second layer is restructuring the human-machine division of labor. This layer is the easiest to misunderstand. Many people draw a diagram: AI handles A, B, C; humans handle X, Y, Z; and here are the handoff points. This kind of static division-of-labor diagram simply doesn’t work, because the boundary of AI’s capabilities shifts every three months. My approach in my own company is to design a “dynamic authorization framework” — the AI has basic autonomous authority, but any decision exceeding a certain level of complexity or risk is automatically escalated to a human. This threshold isn’t fixed; it’s continuously adjusted according to the AI’s performance and the team’s level of trust. It’s like mentoring a new hire: at first you check everything, after six months you only review key decisions, and after a year you only look at the results.
The third layer is value restructuring. This is the deepest layer, and the one fewest people are talking about. When AI pushes execution efficiency to its limit, competition between organizations shifts to domains AI cannot easily replicate — understanding cultural context, the ability to build trust with people, and the courage to make responsible judgments in gray areas. I discussed a similar point in Thinking in the Post-Code Era: When Taste Becomes Humanity’s Key Competitive Edge: when the cost of execution approaches zero, judgment becomes the only differentiating factor. At the organizational level, this “judgment” is the company’s culture, the quality of its decisions, and its resilience in the face of uncertainty.
From power socket to barrier: the metaphor of MCP
There’s one technical development worth discussing specifically: MCP (Model Context Protocol).
MCP is a standard protocol that lets different AI models and tools communicate with one another. It sounds very technical, but its significance goes far beyond the technology itself.
Imagine that before MCP appeared, every AI tool was an island. Your customer-service AI didn’t know what your warehouse AI was thinking; your analytics AI couldn’t see your marketing AI’s data. To make them collaborate, you had to write a pile of custom integration code, and the maintenance cost was extremely high.
What MCP does is like how the computer industry once standardized the USB interface. Before USB, every manufacturer had its own connector, and buying a printer meant praying the connector matched your computer. After USB standardized things, you no longer had to think about connectors — you could focus on what you wanted to print.
MCP’s impact on the AI ecosystem is similar. When AI agents can communicate seamlessly, the always-on economy is no longer “a bunch of AIs each going their own way, working in different corners,” but an organic, instantly coordinating intelligent network.
I call this a “barrier” — an operating space with AI as its underlying infrastructure, in which information and decisions can flow without being constrained by human schedules. Once the barrier takes shape, the efficiency inside it and the efficiency outside it are no longer in the same order of magnitude.
The real challenges for Taiwanese companies
After all this talk, let’s return to the reality in Taiwan.
Taiwanese companies face a structural difficulty in transitioning to the always-on economy: our organizational culture relies far too much on “people.”
This isn’t a bad thing — the flexibility of Taiwan’s small and medium-sized enterprises, their trust networks, and the culture of the boss rolling up his sleeves to work alongside everyone, are the reasons we hold a place in the global supply chain. But it also means that many of our processes “follow the person” rather than “follow the system.” The boss remembers every customer’s preferences, the veteran salesperson judges the authenticity of an order by intuition, the factory supervisor adjusts production-line parameters by experience.
This “tacit knowledge” becomes a double-edged sword in the AI era. On one hand, it is a valuable asset that AI cannot easily replace. On the other, it is also a barrier to transformation, because this knowledge hasn’t been systematized, and AI cannot take it over.
In my own experience, the first step in transformation is not adopting some AI tool, but spending time “translating” the team’s tacit knowledge into a form a system can understand. This process is painful, slow, and often meets resistance — “this kind of thing just can’t be written into rules” is the line I’ve heard most often. But once it’s done, AI can finally be put to genuine use, rather than being just an expensive toy.
Are we ready? The question itself is wrong
Every time AI-driven transformation comes up, someone eventually asks: “Are we ready?”
I think there’s something wrong with the question itself. It assumes there is a state of being “ready,” as though you can perfect your swimming stroke on the shore before getting into the water. But the reality is that the water is already up to your ankles.
A more practical question than “Are we ready?” is: Do you have an architecture that can keep adjusting amid change? Is your organization flexible enough to keep adapting at the pace of AI capabilities upgrading every three months? Does your team have people who understand both technology and business, who can translate between the two?
With AI’s support, the always-on economy is shifting from “a norm sustained by human endurance” to “an instinct driven by intelligence.” But instinct isn’t innate — for an organization, it is designed.
Those who start redesigning now won’t necessarily win. But those still waiting until they’re “ready” are probably already too late.
💬 Comments
Loading...