Today, I want to talk about focus.

In today’s information society, focus is actually a very rare quality.

There are endless short videos on Tiktok, nonstop articles pushed by influencer, and never-ending posts on X.com . The feed never stops. There is always something new competing for your attention.

Many times, even when I am doing important work, I still cannot resist getting distracted and picking up our phone to scroll through social updates.

Many creators online have already explained this phenomenon from the perspective of neuroscience and dopamine. In the book Cognitive Awakening, the author mentions that focus is, in a sense, against human nature. To focus means your brain has to keep thinking, and thinking is an extremely energy-intensive activity.

As biological beings, we naturally resist high-energy behaviors.

By contrast, entertainment and zoning out are low-energy activities. (Don’t tell me you are deeply thinking while scrolling Tiktok.)

So we naturally gravitate toward things that are easier, lighter, and bring immediate pleasure. There is nothing shameful about that. It is simply human instinct.

Recently, vibe coding has become very popular, and in some ways, it is also a form of distraction.

For a while, I became obsessed with rebuilding all the tools I use every day with AI through vibe coding.

Watching files get created one after another, seeing an app slowly come to life, and watching an interface appear that looked surprisingly decent, it all felt great. It felt like I was building a lot of things in a very short time.

But after a while, I started to experience AI fatigue.

I had five projects running at the same time, constantly switching contexts between them, and my brain stayed in a permanent state of overload.

When I looked back at those vibe-coded projects, I suddenly realized:

I could barely understand them anymore.

Even worse, I had no idea what to do next.

In theory, I had built five new applications that could all run.

But looking back, I had barely learned anything.

I had not done much systematic thinking. Most of the time, I was just mechanically typing in requirements that could not have been simpler, then waiting for AI to generate the result.

To be fair, vibe coding does have value.

It can validate ideas quickly and create small tools quickly.

For people without a technical background, it also means they can build applications to automate repetitive work simply by stating what they want.

In a certain sense, AI really has lowered the barrier to software development.

But that is also exactly where the problem begins.

Because vibe coding makes hands-on practice extremely cheap, it breaks a cycle that used to be very important for the first time:

Learning knowledge -> forming thoughts -> practicing by building

In the traditional learning path, practice is built on top of understanding and thinking.

But in the vibe coding model, the process often gets compressed into this:

Idea -> AI generation

And the most important step in the middle, thinking, is very easy to skip.

The absence of that thinking process is actually fatal.

In the process of learning, thinking is often the highest-leverage step.

Without it, you may feel like you are constantly “practicing,” but in reality, a lot of your time may just be spent generating code inefficiently.

The same thing is also happening across the software industry.

AI has brought a kind of democratization of technical productivity. As long as someone has an idea, with the help of agents they can build an app or a website.

The number of app submissions to the Apple Store keeps increasing, and GitHub Trending has new projects every day.

But are these things actually useful?

Do we really need N+1 todo list apps?

A lot of projects, whether at companies or done by individuals, face the same issue:

they are too focused on the outcome of “shipping something,” while neglecting the thinking that should happen before building.

As a result, what gets produced is often just garbage.

At the same time, the software industry really is standing at a historical turning point.

From the ChatGPT moment, to prompt engineering, to agentic engineering, from MCP to Skills, technology is moving faster than at any point in the past twenty years.

And this time, the progress is leap-based rather than incremental.

You may just have learned a few prompt engineering techniques, and a few months later, an agent with Skills can already do the same thing with ease.

The multi-agent collaboration platform you built may be replaced a few days later by a newer tool. Tools like Claude Code, for example, are already starting to support native agent team collaboration.

In this kind of environment, it is easy to develop a feeling that:

any investment in technology no longer seems meaningful.

In the past, learning one language or one framework might support your work for years.

Now, a system you spend several days building may be replaced a few days later by a feature that a new tool supports natively.

In the Agent era, what really matters is not what you want to build, but first being clear about what you should not build.

You can direct a whole group of agents, open five terminal windows, make one hundred commits in a single day, and build a system that looks extremely complex and “complete.”

But you can also choose another path.

Focus on one specific scenario. Solve one problem. Polish the experience and the interaction to the highest possible level.

The Unix philosophy is still true today:

Do one thing, and do one thing well.

There is also no need to worry too much about AI replacing programmers.

Change will definitely happen, and it will be a major change.

But as long as you can stay focused, you will often be among the first to sense the shift and embrace it.