The scarcity in the LLM era

三月 20, 2026
Tags: #AI #writing #publishing

I'm not a designer. This does not change the fact that I can build professional-looking interactive prototypes in minutes using an AI agent.

Google has come up with a new vibe-designing tool called Stitch. From the demo, it seems to render even my half-ass "design" process antique and dumb. Why am I still generating prototypes page by page, while I can just tell Stitch what kind of app I am building and it instantly creates all the pages one can expect in such an app, along with complete design system documentation that my coding agents can use?

Yet this does not change the fact that I know very little about design, either. I guess the real problem is that it makes even less sense for me to know more about design now.

The same is happening to everyone who knows Gemini is not just the constellation and Claude is not only the name of their French neighbour. There was no way in the past that a programmer also knew like the back of their hands how a hedge fund manager worked, or vice versa. But this all-trade knowledge is expected of everyone now. At least from where I stand, the job market expects many candidates know everything from programming to product management to how specific industries (aviation, health care, robotics, logistics, etc.) operate.

I guess the rationale behind this is "at least the programming part is taken care of". Now, design skills are added to the "taken care of" list.

Now, as everyone claims they can do this and that (with a tongue-in-cheek "or at least I know what to ask ChatGPT"), it becomes really difficult for anyone who is not in a school, college or university to learn how or why a thing is.

And know-how and know-why are going to be in short supply when all human knowledge are stored, coded and embedded in all sorts of models.

A hedge fund manager is now able to use an agent to make an agent that embodies their knowledge in the financial market, but no one can learn anything by watching this agent work. And the agent would not be able to change itself to come up with new solutions when the market environment changes. To do that, one would have to know how various stakeholders and market components interact. A large part of that know-how is not possible to be encoded in prose or program. Does your LLM agent know who to call when the market melts down, or when a panic sell crashes your asset price?

This can happen in any sector or industry. At best, what you have is a text generator that mimics what a professional would say when presented with an expected problem. It is not impossible that such a generated answer ignites a eureka moment that helps you come up with a genuine solution, but that scenario is not guaranteed, and even impossible if you don't understand the hows and whys in your domain.

In times of crisis, you need someone who can think out of the box and also know what's at stake. Neither is possible for generative LLM agents. And would not be possible if a real person has had no opportunity to know how and why things are done in the way they are done.

Even if not in such a dire situation, not knowing hows and whys stymies innovation. Yes, vibe-coding can make you an app in no time, but don't expect it to come up with genius touches like Quake III's fast inverse sqrt operation. And Stitch (to a even bigger extent, Figma Make) cannot churn out something that is not routine, bland, soulless, although admittedly still pleasing to the eyes.

It is tempting for organizations and managers to vertically merge positions in expectation of a fullstack everybody, given what LLM agents can do today. Well, don't, unless you honestly think the job can be done by a robot. For any person working on such a position, they will find it difficult to know anything profoundly beyond processes and routines. Your product quality won't improve over time and cannot adapt when the market changes.

Corollary: for individuals especially those who desperately need a job in today's labor market, don't heed to "opportunities" that require you to do work that used to be what we call professional work and you don't grasp enough how or why, including but not limited to coding, designing, law services, finance, and HR. You won't be able to hone your existing skills and would not know more about those professions.