The Value of “Useless” Things—and Some Thoughts on AI

A few years ago, during a rather boring stretch,

I casually bought a Python course from an unfamiliar provider advertised on WeChat,

just to play around with it. I think it cost a little over RMB 1,000.

At the time,

first, I wanted to see whether I could learn web scraping to collect data, since I often need to do analysis;

second, I wanted to understand how programmers think. (I often work with technical people, and wanted to understand their thinking so we could communicate more easily .)

I bought two subjects, but finished only the basic language course. Later, work got busy and I stopped studying.

The company I bought the course from later became a star startup that raised hundreds of millions of yuan—风变科技 (Fengbian Technology, the course provider). I had a pretty good eye, didn’t I?

Recently, while studying agent workflows and RPA,

I was delighted to discover that their underlying logic is much like Python’s!

It was so easy to understand and get started. I was thrilled! Hahaha

Because I had never expected an offhand investment years ago to suddenly become so useful years later, hahaha.

But actually, this wasn’t the first time I had tasted its benefits.

I remember one occasion

when I was discussing a very specific scenario involving trucks entering and leaving a factory with people in the technical department,

and I actually used programming logic and mathematical problem-solving to propose a solution,

even thinking through how the code might be written .

When I saw the surprise on the other person’s face,

I knew this stuff was useful! Ha!

On another occasion,

I quite boldly “challenged” the algorithm team because the application scenario seemed unreasonable.

Fortunately, a manager at the time was very open-minded and supported me. They ultimately helped me get an answer from the algorithm team.

That experience left a deep impression on me, and I am still grateful!

One small incident helped me understand

that sometimes a slight loss of efficiency or logic at the level of individual units allows a larger architecture to accommodate and coordinate more units, so that together they achieve an optimal system-level result.

This resonated strongly with me, and I agreed with it. Coming from an engineering background, I immediately understood its significance.

Putting myself in their shoes, I knew I would have made a similar choice.

We were looking at different “chessboards.”

(P.S. This is such a fascinating point. I really want to write a separate article someday about “maximizing system-wide value” .)

Looking back,

both incidents connect to my having taken a coding course that seemed useless to me—a person working in marketing who would never become a programmer.

It brought me many interesting exchanges and discoveries.

Years later,

I may no longer remember the details of Python,

or exactly how to use those functions,

but I deeply understand and remember its logic.

Afterward, I could roughly judge how a requirement might be implemented and how complex it would be;

and roughly tell whether project managers or programmers were misleading me when they estimated that a requirement would take a month ;

I could also roughly distinguish good programmers from poor ones;

and recognize the strengths of good system architecture and an architect’s way of thinking.

These are all hidden benefits it gave me!

Yesterday, I even recognized something else:

the essence of an agent is a large language model plus a workflow.

Why do we need this combination? Because today’s large language models are not fully controllable.

They cannot yet carry out certain tasks in an orderly, step-by-step manner like people,

whereas workflows can constrain the models to execute as people require, making them more usable and better suited to the task.

In other words, they have greater commercial value.

Before true AGI arrives, agents are the best expression of AI commercialization and application.

And when we talk about using agents, or the foundational abilities needed to build good ones, we come back to the same enduring abilities—

1. The ability to break down problems

Holding both a large-scale architectural perspective and a view of individual functional units, knowing how to take things apart and put them together.

2. The ability to make logical judgments

Judging and constructing relationships between sequences of events, modules, and scenarios.

3. Mathematical and computational ability.

Deciding which mathematical approach to use to solve and calculate a particular event, module, or scenario.

Suppose true AGI arrives in the future,

and the barriers and tools that now prevent many people from using AI disappear, with AI powerful enough to execute our instructions seamlessly.

What really matters?

What truly distinguishes people will still be those things above:

our own foundational abilities.

So we should return to human value and consciously cultivate these abilities.

AI isn’t mysterious or frightening. You yourself are the most impressive AI.

I meant to talk about the value of apparently useless things, but once again I couldn’t stop. These are also some of my recent thoughts and feelings about AI.

You can judge whether they are useful. Either way, I’m very happy!

But the core hasn’t changed. What I wanted to say is,

not everything in life needs to serve a practical purpose.

Sometimes doing something apparently useless simply because it makes you happy has meaning in itself. You never know when it might bring an unexpected return.

So allow yourself to do some “useless” things!