The "involution" of East Asian education is being re-exported by AI.

"It's all over."

Shortly after the launch of the first iteration of Perspeak AI, a user completed a multi-party discussion training session, saw the system-generated score — 60 points, and blurted out these three words immediately.

Courtney, the founder of Perspeak AI, was leading the team to review user feedback at that time. This message served as a sharp wake-up call: if AI only retranslates "communication competence" into a cold, rigid number, what it replicates will never be the driving force for self-improvement, but the most daunting part of East Asian education — an evaluation mechanism that defines individual value solely through scores and rankings, a verdict that declares "you are not good enough".

This specific reaction from one individual user eventually overturned the entire evaluation logic of Perspeak AI: the system shifted its core from "what score did you get" to "what did you do, and what more can you do to improve".

This is not an isolated accident. Over the past two years, a group of Chinese entrepreneurs have unexpectedly run into the exact same wall. What they wanted to do sounds rather straightforward: reorganize the most effective training methods of East Asian education — goal setting, task decomposition, repeated practice, and timely error correction — with AI, and sell this solution to learners all over the world. However, the problem exposed by the "60-point incident" is only the tip of the iceberg: while the training methodology itself may be replicable, the evaluation system, cultural context and learning motivation behind the training are far from being easily transplanted elsewhere.

This business is now being validated by both capital and real market demands. Data released by HolonIQ in July this year shows that global edtech venture investment in the first half of 2026 was approximately 1 billion US dollars, down 26% year on year; yet against the overall cooling trend, the number of transactions in the East Asian region rose by 37% against the market downturn. Grand View Research estimates that the global AI education market will reach 11.4 billion US dollars in 2026. In the top 50 global consumer AI mobile app list released by a16z in March this year, education products including Gauth, Brainly, Photomath and Learna AI still hold their positions — AI education is no longer just a concept in the financing market, but a real practice embedded in the daily learning routines of students.

The temperature gap between capital expectations and actual market demands is splitting the AI education track into two parallel forces: large tech companies quickly capture general scenarios with their traffic, capital and content ecosystems, while more lightweight teams avoid head-on competition, and dive into the unsolved tiny pain points in the learning process. Shuta AI, co-founded by Pmis, focuses on courseware reading scenarios; Inspired AI founded by Jia Zijian cuts into the language listening and speaking track; Perspeak AI built by Courtney targets the communication pressure that international students face in discussions and presentations. Behind these three cases, a similar path has emerged: combining the underlying training logic of East Asian education with AI, and bringing this solution to the global market.

And the problems left by this small "60-point incident" will keep popping up in different forms for each of these entrepreneurs.

What East Asian "involution" brings out is far more than scores

When talking about East Asian education, "involution" is almost an unavoidable keyword: doing exercises, taking exams, competing for rankings, facing pressure, and chasing standard answers. But beneath these highly controversial appearances, there still exists a complete set of methodologies for how competence is formed — setting goals, breaking down tasks, practicing repeatedly, and making continuous adjustments through feedback.

What AI education is reactivating right now is exactly this underlying logic.

The most typical manifestation of this logic is the ability to solve problems around clear goals. In a highly result-oriented environment, students are accustomed to confirming the standards they need to meet first, then finding the path to achieve the goals, and completing rounds of training within a limited time.

Pmis is very familiar with this set of logic: she entered primary school at the age of four and a half, took the national college entrance examination at 16, and completed her high school training in a key middle school known as "Hengshui High School of Liaoning Province". This experience helped her form a relatively pragmatic view of efficiency: pressure cannot be completely avoided, but clear goals and long-term commitment can cultivate concentration, learning ability and the spirit of in-depth research.

She has also carried this mindset into her current learning. When facing unfamiliar courses, she usually sets clear goals first, then figures out the path to complete them. When using AI to read hundreds of pages of foreign books, she compresses the time spent on information processing, and devotes the saved energy to more knowledge fields. In this scenario, AI does not replace learning, but improves learning efficiency under established goals.

Deeper than the awareness of goals is the belief in the value of training.

East Asian education is built on a simple premise: competence does not entirely depend on talent, and repetition, error correction and continuous investment can lead to progress. This belief is often overshadowed by the negative narrative of "involution", but for some learners, "involution" does not entirely come from external competition, and can also be a proactive process of self-iteration.

Courtney belongs to this latter group. She once named her social account "Courtney the Involution Champion", but later changed the name. As an English teacher, she does not deny the value of practice. In her view, even in Western education that emphasizes interest and autonomy, basic competence still requires repeated training. The real problem of East Asian education is not that there is too much practice, but that the training often stays at the level of chasing standard answers, and fails to extend to real-world scenarios.

"Mute English" is the most typical manifestation of this breakpoint. Many East Asian students master a large vocabulary and grammatical rules, and can achieve good scores in reading and exams, but they can barely speak naturally in real communication scenarios.

It is not that they lack a basic foundation, but that they lack the training to turn knowledge into action. When language shifts from being an exam subject to a communication tool, learners not only need to say correct sentences, but also face uncertain responses, judge when to join a discussion, how to handle disagreements, and how to form their own opinions when there is no standard answer at all.

This means that the training methods accumulated by East Asian education are still effective, but the end point of training needs to be changed: from getting the correct answer, to completing tasks in real scenarios.

At the same time, the awareness of results also makes this group of entrepreneurs focus more on educational effects, rather than simply chasing the novelty of technology. When Jia Zijian worked at TAL, an internal saying left a deep impression on him: Failing to teach students well is equivalent to stealing and robbing money.

In his view, this sentence defines the bottom line of the education business. Education can be charged for, and can be scaled up, but it cannot be separated from real learning effects. A product may acquire users through advertising investment, but it is difficult to build word-of-mouth and achieve long-term operation without tangible learning outcomes.

Being responsible for results also means accepting that both education and entrepreneurship require long-term accumulation. Jia Zijian describes himself as "slightly involuted", but it is more about "pushing myself to improve rather than competing against others". In his view, growth rarely comes from a one-off explosion, but from constantly identifying knowledge gaps and filling them one by one; entrepreneurship is also a marathon, and what is more important than short-term input intensity is continuous learning and repeated polishing of key links.

These entrepreneurs have different understandings of East Asian education, but they all point to a similar underlying cognition: complex abilities can be disassembled, basic skills require practice, errors should be fed back in a timely manner, and all inputs need to be verified by the final results. This cognition cannot represent the full picture of East Asian education, but it constitutes the more universally applicable training logic of this system.

The emergence of AI has further amplified the value of this training capability. Explaining a problem, translating a piece of material, and generating a model essay have all become increasingly easy. Knowledge has not lost its value, but the cost of obtaining standard answers is dropping rapidly. The new competition is no longer about who owns more answers, but about who knows better how to organize these answers into effective training, and help users turn knowledge into practical competence.

However, a set of training methods that works well in East Asian classrooms will not automatically become applicable globally just by integrating with AI. What can truly go global is the underlying structure of the training, rather than the exam goals, classroom order and cultural habits attached to it. How to separate these two parts has also become an important issue that Chinese AI education entrepreneurs must face.

Methodologies can go global, but classrooms cannot be copied directly

The easiest part of AI education product globalization is language adaptation, while the most underestimated part is the real life behind the language.

Pmis maintains a relatively cautious attitude towards this. "Shuta AI currently mainly serves overseas Chinese communities. If we expand to the local user market later, we will prioritize East Asia and Southeast Asian regions that are culturally closer to us". The reason is that the curriculum settings, courseware formats and learning paces in different regions are not the same. The demands that hold true for Chinese international students cannot be directly inferred to be accepted by local students in Europe and America.

Shuta AI product design

User feedback is also very different from her initial assumptions. Pmis originally envisioned Shuta AI as a learning space covering the full "preview-learn-review" workflow, but after the product was actually launched, the favorite feature of users is neither summarizing key points nor AI Q&A, but a function that does not seem to have high technical content — reading English courseware while viewing the corresponding Chinese translation side by side. "This function actually has little to do with AI, but everyone loves it." Her own explanation is that other tools such as Youdao Translate and Doubao require manual clicks and screenshots every time to translate, while Shuta AI supports "seamless translation while browsing", and this smooth experience retains users earlier and more directly than any other advanced in-depth functions.

Real, vivid user feedback is the most fundamental driving force for product optimization and iteration.

Courtney mentioned that many international student users told her that overseas, due to language differences, insufficient competence, plus their identity as "Asians, Chinese", they often feel discriminated against, and have even been bullied and framed by classmates. It is difficult for them to make local friends, integrate into local social circles, or get decent scores on collaborative assignments. These international students said that if they could practice in a safe environment and experience what the atmosphere of group discussions is really like before stepping into the campus, they would be very willing to try. This is the real starting point for Perspeak AI to develop "AI high-pressure communication training": it is not about whether to be involuted or not, but that these students already come to overseas campuses with unspeakable pressure and a sense of isolation, and the product can first help them cope with this real awkward situation.

Courtney also observed that some Chinese students are used to fully forming their opinions first, and waiting for others to finish speaking before they start to talk. In their received education and cultural concepts, this is a sign of politeness; but in overseas seminars or group discussions, speeches are often connected to each other, and few people will formally pass the speaking turn to you. If you keep waiting for "your turn", you may never find an opportunity to join the whole discussion.

Perspeak AI simulates multi-person interaction scenarios

Therefore, AI must not only understand what the user is saying, but also understand what a certain behavior means in the local culture. Whether a direct question means conflict, whether silence means careful listening or withdrawal from the discussion, how to express disagreement clearly without undermining cooperation — these are all problems that cannot be solved simply by translating Chinese courseware into English.

However, cultural differences cannot be simplified into fixed labels either. "Individual differences are far greater than a single cultural label", Courtney emphasizes that truly effective localization requires continuous feedback from local users and real usage scenarios. The same way of expression may lead to completely different effects in different classrooms and cultural environments. AI needs to understand the communication context in specific scenarios, rather than simply relying on regional or cultural labels.

When the training goals and usage scenarios change, the evaluation system also needs to change accordingly. This is also a necessary transformation for East Asian education methods after going global: in schools, exams and further education provide external pressure; in the consumer market, users can quit the product at any time, so the product must rebuild motivation through interest, value and real demands.

Customer acquisition and user feedback have therefore become two practical entry points to test whether localization is successful. A team that can translate the product into more than ten languages does not mean it knows how to acquire users in more than ten countries/regions. Mature teams can rely on multilingual social media, advertising investment, app stores and word-of-mouth to expand their scale; early-stage teams often need to start from campus clubs, campus ambassadors and seed users in one or two schools.

Just like Shuta AI tries both domestic and overseas social media channels and some offline activities; Perspeak AI plans to build real cases in Australian universities first; Inspired AI requires its core team to reply to user emails and feedback from different regions every day. The market cannot be understood through a single globalization decision, but is re-recognized through pieces of feedback, rounds of advertising investment and batches of churned users.

East Asian education excels at correcting learning through continuous feedback, and globalization teams also need to use this method to adjust their strategies. Treating each country or region as a course that requires long-term learning, observing users, getting feedback, and then adjusting content and products, may well be another commercial extension of this set of educational methodologies.

Only by finding training goals that fit the local culture can users participate in the long term; only with continuous user participation can the training advantages of East Asian education be transformed into a viable business by AI.

Is the "AI + East Asian Education" globalization track really a good business?

Today, Jia Zijian's Inspired AI has gone through its most difficult stage. Founded at the end of 2023 when AI applications were still in a boom, the company secured some early investment, and has now achieved profitability, earning hundreds of thousands of US dollars in revenue per month — in his own words, it is "one of the few AI companies that are actually making real profits".

Behind this, it also requires long-term focus and patience. "Education is not a business that can be rushed to mature." Jia Zijian says. He cites three examples that have survived for 20 years — Duolingo, TAL and New Oriental, "they all grew up little by little day after day". Looking back at the education startup teams founded in the same period as them, "almost all of them have switched to other tracks or closed down now", and his team is one of the few that still stays in the game.

Inspired AI has two separate products, TalkMe and ListenLeap, one solves the problem of "dare not speak", the other solves the problem of "cannot understand". Nearly 100 million pieces of user data have been accumulated behind the two products, which have long ranked top 2 in the education category in the Taiwan region of China, and achieved a global user rating of 4.9 points.

TalkMe AI simulates real conversations

Jia Zijian summarizes the customer acquisition logic of this business into three points: first, operate in multiple languages and multiple regions simultaneously from the first day of establishment. "In the past, a product only supported one language in one region, but now our product supports N languages in one region, and we have multiple products at the same time, which forms an N*N*N multiplier effect"; second, the overseas and domestic social media matrix can generate nearly 10 million exposures per month; third, word-of-mouth — "our team has a very good habit: the first thing the core team does every day when arriving at the company is to reply to user feedback, user messages and user emails, and strive to ensure every voice gets a response, no message is left unanswered". He also does not shy away from the necessity of advertising investment: "Anyone who says they never run ads and spend no money on promotion is a liar