← All insights

E · Use · 2026-10-06

An Assistant for Every Teacher: An Education Investment of NT$22 a Day

Illustration · not a service record: Workshop in session: participants working with AI toolsIllustration · not a service record

An Assistant for Every Teacher: An Education Investment of NT$22 a Day

On Teachers’ Day last week, teachers probably heard “Thank you for your hard work, teacher” quite a few times.

But teachers want more than a word about their hard work. They need someone to see what that hard work is actually made up of, and then help them take back a little time.

The OECD’s TALIS 2024 teacher survey shows that full-time junior high school teachers in OECD countries work 41 hours a week, of which only 22.7 hours are spent teaching classes. The rest of the time goes to lesson preparation (7.4 hours), marking homework (4.6 hours) and administrative work (3 hours), plus counseling students, contacting parents, and meetings that never end.

And Taiwan? Early this year, CommonWealth Parenting (親子天下) surveyed 5,412 elementary and junior high school teachers. More than 90 percent were under high pressure, and their sense of well-being scored only 50 points. On World Teachers’ Day, October 5, the demands of the National Federation of Teachers Unions (全教總) also included “making administrative work a dedicated full-time job.”

Some pressures have to be answered by the system. Take the fear of abusive complaints, which in the end leaves teachers no choice but to discipline less, say less and step in less. That is not something any app can solve. But while the system slowly adjusts, I believe more and more:

When AI enters schools, the first thing it should take over is not the teacher’s podium, but the teacher’s paperwork.

▍Many teachers are in fact already paying for it themselves

Drafting meeting minutes, putting together outcome reports, compiling a whole class’s grades into a report, turning one worksheet into three levels of difficulty: Claude, ChatGPT and Gemini can all do these things now. They can also all connect to the cloud drives schools are already using, reading the files directly and producing the output directly.

I also see many dedicated teachers who have long been paying out of their own pockets to subscribe to AI tools. This is money that teachers should never have had to pay themselves.

In my own experience of using Claude, for the same roughly NT$660 a month: last year it was like gaining a senior engineer on call, who could turn a teacher’s ideas directly into small tools and interactive web pages; this year it is like gaining a senior animator as well, who can even make instructional animations with you.

▍This is not a feeling. There is research

The UK’s Education Endowment Foundation (EEF) ran a randomized controlled trial: 259 secondary school teachers, half of whom were given ChatGPT and a usage guide. As a result, lesson preparation time fell by 31%, and the quality of the teaching materials did not get worse.

In the United States, Gallup surveyed more than 2,000 teachers: those who use AI every week estimated that they save 5.9 hours a week, about six weeks over a school year. Where schools have clear AI policies, the time saved is 26% greater still.

▍So I want to propose a new “universal handout”

In recent years we have become very familiar with “universal handout” policies such as universal cash payments and free lunches. I want to propose another kind: a universal handout of tokens to educators.

Taiwan has about 200,000 teachers at the senior high school level and below. At NT$660 per person per month, that is about NT$8,000 a year, or about NT$1.6 billion a year for the whole country.

The number is not small. But worked out, it comes to NT$22 a day per teacher, cheaper than a cup of hand-shaken tea.

▍Countries around the world are already doing it

Estonia has been running “AI Leap” since 2025, training teachers first and then giving senior high school teachers and students free use of tools from OpenAI and Anthropic. One of its goals is precisely to lighten teachers’ administrative burden.

Iceland’s Ministry of Education has had several hundred teachers try out AI: elementary and junior high school teachers use Gemini, senior high and vocational high school teachers use Claude, and practicing teachers serve as the lead of each group.

The United States has a free ChatGPT for Teachers, which American K–12 teachers can use until June 2027, but teachers in Taiwan cannot apply for it. The American Federation of Teachers is also working with AI companies to train 400,000 teachers.

In Singapore, 76% of teachers have taken training related to teaching with AI; the OECD average is only 38%.

There is also a cautionary example. South Korea spent more than one trillion won rolling out “AI digital textbooks,” pushed straight into classrooms from the top down, and nearly 90 percent of teachers said they did not have enough preparation or support. After one semester, the National Assembly downgraded them to ordinary teaching materials.

The difference is this: one side designates a system from the top down; the other lets teachers have the tools, the training and the choice first.

▍This does not conflict with sovereign AI

Some will ask: isn’t Taiwan supposed to be developing sovereign AI? Wouldn’t spending money on foreign services be putting the cart before the horse?

My view is just the opposite. The most critical assets of sovereign AI are not only computing power and models, but also a group of people who are very good at using AI. Put plainly, people who are very good at using AI to solve their own pain points: who know what AI can help with in the classroom, where it goes wrong, and what Taiwan’s teachers really need. Without this group using it heavily and giving feedback heavily, even if a model gets built, no one will know which way to tune it.

So the tokens can be designed as “teachers choose for themselves.” Foreign models are fine, and Taiwan’s own models, as long as they are good enough, go on the menu too. Let quality do the competing, rather than tying teachers to a single system.

▍The precondition: don’t let it turn into yet another outcome report

For a universal handout of tokens to be useful, the supporting measures have to do three things:

1. Pair it with training, and not one-off workshops. Once the government hands the tokens out universally, a training ecosystem of teacher communities, teacher-education universities and civil society groups is sure to grow, and Junyi Academy will be part of it. Drawing on the OECD, we promote “Ask, Use, Manage, Build (問用管造),” and in Taiwan and even abroad we already have the experience and the models for promoting it from elementary school through to adults, so we can certainly make it something teachers really feel.

2. Use education or enterprise editions, so that data is not taken to train models and students’ personal data is not uploaded.

3. Teachers choose their tools freely, with no compulsion and no evaluation of usage rates.

But one caution: what I fear most is that after the tokens are handed out, teachers are then required to submit an “AI usage outcome report.” Please, absolutely not. There are so many ways to track the results.

▍Finally

When teachers use AI to solve their own pain points, this is the investment with the highest leverage: it helps teachers themselves, helps the next generation, and helps the whole public sector learn how to use AI.

I hope the 2027 budgets of the central and local governments can move in this direction.

A question for teachers: if you had an AI assistant tomorrow, what is the first thing you would want to hand off to it?

(Note: Junyi currently supports each full-time team member with three AI tools on average, and those who need more can even ask for more, because they really are that helpful. We also recommend that many innovation-minded schools do the same. But for a breakthrough across the board, an across-the-board policy really is needed.)

Sources are in the first comment