AI lowers the barrier to building an app—not the standard for building a good productAI 降低了做 App 的门槛，却没有降低做好产品的门槛

After hearing Duolingo’s CEO discuss AI, competition, and defensibility, I reconsidered the problem ThaiCue should truly solve.听完多邻国 CEO 谈 AI、竞争与护城河，我重新想了想 ThaiCue 真正要解决的问题。

Interview excerpt · 20:19–22:57访谈片段 · 20:19–22:57

Lately, I have been using AI to build ThaiCue—a website for learning Thai through songs.

When Duolingo’s CEO answered the question, “If anyone can build an app in 2026, should Duolingo be worried?”, one idea resonated with me: AI can help far more people make an app, but building a genuinely good language-learning product remains extremely difficult.

The moat is not merely code

Duolingo’s moat is not a particular interface or the ability to put exercises into an app. It comes from the learning data generated by hundreds of millions of users, more than a billion exercise responses, and years of accumulated knowledge about which exercises work, which feedback keeps people going, and what brings someone back tomorrow.

There may once have been two or three thousand language-learning apps. After vibe coding, there may be twenty thousand. More products, however, do not automatically mean proportionally more good products.

AI helps me turn ideas into features

Recently, I added lyric-sequence learning, word-by-word explanations, bilingual prompts, adaptive practice, draggable sentences, and a daily-learning homepage to ThaiCue. AI helped these ideas become working features much faster.

Yet once the features are real, the difficult questions begin:

Why would someone choose to open it?
How should practice be designed so that something is truly remembered?
How can learning be enjoyable enough that someone wants to return tomorrow?

AI shortened the distance between an idea and a feature. It did not answer these questions for me.

ThaiCue does not need to copy Duolingo

This made me reconsider ThaiCue’s direction. It does not need to become another Duolingo covering every language and every learning path. It can begin with a narrower and more genuine question:

How can people who love Thai music and Thai GL begin with a lyric they love—and truly learn one sentence of Thai each day?

Loving a song is already a reason to learn. ThaiCue should not replace that affection with an abstract curriculum. It should shorten the journey from hearing a line, to understanding it, to being able to sing it.

A lower barrier creates a higher standard

AI enables one person to begin building a product, but it also raises expectations. Products should feel smarter, more capabilities should be free, and improvements should arrive faster. When platforms shift, today’s leader is not guaranteed to remain tomorrow’s winner.

The worthwhile accumulation, then, is not the number of features I have shipped. It is whether I increasingly understand whom I am helping, which experiences genuinely support learning, and what makes someone want to return.

Code is becoming easier to write. Understanding users, designing a learning experience, and making a product good over time remain difficult—and that is precisely why this is worth continuing.

最近，我一直在用人工智能做 ThaiCue——一个通过泰语歌曲学习泰语的网站。

听多邻国首席执行官回答“到了 2026 年，任何人都可以做自己的应用，多邻国会担心吗？”这个问题时，有一句话让我很有共鸣：人工智能的确能帮更多人做出应用，但做出一个真正好的语言学习产品，依然非常难。

护城河不只是代码

多邻国的护城河不是某一个界面，也不是把练习题写进应用的能力。它来自数亿用户每天产生的学习数据、超过十亿次练习反馈，以及多年积累下来的经验：什么样的练习有效，什么样的反馈能让人继续，怎样让一个人明天还愿意回来。

以前可能有两三千个语言学习应用。氛围编程兴起之后，也许会出现两万个。但产品数量增加，并不等于好产品也会按同样的比例增加。

人工智能帮我把想法做出来

最近，我为 ThaiCue 做了歌词顺序学习、逐词拆解、双语提示、智能练习、句子拖拽和每日学习首页。人工智能让这些想法能够更快地变成真实、可用的功能。

但当功能真的摆在眼前，困难的问题才开始出现：

用户为什么愿意打开它？
怎样设计练习，才能让人真正记住？
怎样让学习过程有趣到愿意明天再来？

人工智能缩短了从想法到功能的距离，却没有替我回答这些问题。

不必复制另一个多邻国

这也让我重新思考 ThaiCue 的方向。它不需要成为另一个覆盖所有语言、所有学习路径的多邻国。它可以从一个更具体、也更真实的问题出发：

怎样让喜欢泰国音乐和泰国女性爱情题材作品的人，从喜欢的一句歌词开始，每天真正学会一句泰语？

喜欢一首歌，本身就是学习的理由。ThaiCue 要做的，不是用一套抽象课程替代这种喜欢，而是把“听见一句”“理解一句”“能唱一句”之间的距离变短。

门槛降低之后，标准反而更高

人工智能让一个人也能开始做产品，同时也抬高了用户的期待：产品应该更智能，更多能力应该免费，更新也应该更快。平台变化以后，今天的领先者未必仍然是明天的赢家。

所以，真正值得积累的不是“我又做出了多少功能”，而是我是否越来越理解：我在为谁解决什么问题，什么体验真的帮助了学习，以及什么会让人愿意回来。

代码越来越容易写。理解用户、设计学习体验、长期把产品做好，仍然很难——也正因为这样，这件事才值得继续做。