Yesterday I recorded an episode for the Zao Wu Matrix solo creator podcast.

We mainly talked about three things: the evolution of md2wechat, my view on Agent tools, and how AI can enter concrete workflows in traditional industries.

I already wrote the md2wechat origin story yesterday, so I will not repeat that timeline here.

This article is only about the new questions that came out of the conversation: who is this tool really for? Why do technical users understand it but hesitate to pay, while non-technical users need it but cannot use it? Is a course really a course, or is it a way to discover To B leads? Does the future of a WeChat content tool even stop at WeChat?

§The User Mismatch

People who find md2wechat through GitHub, X, OpenClaw, or AI search are often technical users.

They understand CLI, can install the environment, and know what an Agent Skill is.

But they are not always willing to pay because they can modify the tool, fork it, or rebuild part of it themselves.

The other group is the opposite.

Content operators, content factories, matrix teams, and internal enterprise content teams are closer to the real need. They write, format, publish, track data, and maintain brand consistency every day.

But they often get stuck at the first step: Claude Code, Codex, terminal commands, installation, and CLI mental models.

That creates a painful split:

The people who understand it may not buy.

The people who may buy cannot easily use it.

The next stage is not about adding more themes, modules, or commands. It is about lowering the barrier.

§A Course Is Not Just a Course

During the podcast, the host asked whether a future course would teach beginners how to use the WeChat creation tool.

That question clarified something for me.

If a course only teaches people how to use a tool, its value is limited. Tools change. Platforms change. Tutorials expire quickly.

But if the course brings together people with real content production needs, and helps me observe what they are actually trying to accomplish, then it becomes a different thing.

Someone says they want to write WeChat articles. That is only the surface.

One layer deeper, they may want customer acquisition.

Another layer deeper, they may be an industry consultant trying to influence potential clients through content.

Another layer deeper, they may have research, sales, delivery, review, and knowledge management workflows that can be automated.

In that sense, To C education may not be the destination.

It may be the entry point for discovering deeper To B needs.

§Industry Agents Are Not About Search

The second half of the podcast moved into industry examples.

One friend in retail supply chain consulting uses AI search tools, but the output is often weak.

The problem is not that the model cannot search. The problem is that it does not know how an industry expert judges.

For example, to understand how a chain brand is expanding in a city, an expert may look at company registration data, hiring positions, map heat, and store distribution instead of only reading public articles.

Travel planning has the same issue.

A general Agent can list attractions and produce a full itinerary.

A real travel planner first checks distance, altitude, cross-day constraints, and whether two locations can be reasonably grouped together.

This is not just information retrieval.

It is judgment path extraction.

The value of many future industry Agents will not come from searching harder. It will come from encoding how experts actually make decisions.

§Demand Discovery Can Be Engineered

Near the end, we talked about how someone should start building a small product.

My answer was practical:

Start with your own repeated pain. Is there something you have done more than three times and still find annoying?

Then look at people around you. Are friends, colleagues, or community members repeatedly complaining about the same problem?

Then look at public demand surfaces: Xiaohongshu comments, GitHub Issues, App Store negative reviews, product forums, and user group chats.

These places contain problems that people have already expressed.

In the past, demand discovery meant manually reading comments.

Now it can be engineered.

You can scan GitHub Issues, classify repeated complaints, analyze App Store reviews, and extract unmet needs from social comments.

As AI lowers the cost of building, the bottleneck moves from “can you build it” to “can you judge what is worth building.”

§The Real Shift

Yesterday's article was about how md2wechat grew.

This podcast pushed me into the next layer:

I should not only build the tool. I need to place it into the right user group, the right business workflow, and the right industry context.

That is harder than writing code.

It is also more important than adding another feature.

Yesterday was about where md2wechat came from.

Today is about where it should go next.