How Teachers Can Embrace AI Without Getting Replaced

I was among the first test-prep teachers in Taiwan to research end-user AI applications, integrate AI into my courses, and advocate for students to use AI. Over two years ago, people were flip-flopping between “AI will replace humans any day now” and “AI sucks, why would I bother.” My thinking has stayed the same from then to now.

Want AI working in your own job, not just in demos? See the 3-month 1-on-1 program →

It’s Not AI Replacing Teachers — It’s Those Who Use It Replacing Those Who Don’t

Thirty years ago when the internet first appeared, plenty of people thought computers would replace humans. What actually happened: people who could use computers replaced people who couldn’t. Five years ago when AlphaGo burst onto the scene, some thought professional Go players were finished. Today, every top-tier Go player trains with AI, and the competitive level of the game has risen higher than ever.

The generative AI wave is unstoppable. It will become ubiquitous — a native skill for the next generation. That part can’t be changed. What can be changed is how we collaborate with it to create even greater value.

From Burnout to Full Revival

A few years back, I was just a teaching grunt. My daily explanations to students were on autopilot, I was on the verge of losing all passion for teaching, trapped in the “trading time for money” career dead-end.

AI changed everything. The thing I least expected was this — my professional passion got reignited. My know-how had calcified, but after AI appeared, I saw new inspiration and new challenges again. It’s like Go meeting AlphaGo: a game with thousands of years of history had entered a plateau, with all the standard openings and patterns locked in. Then AI shattered those constraints, and a new generation of players used AI to sharpen their skills, sparking a whole new era of growth.

14 years in, I’d almost lost the fire. AI brought me back to life.

Don’t Think Full Automation — Think Collaboration

The first question a lot of teachers ask is: can AI fully automate my teaching?

My answer: you’re asking the wrong question.

Most people think “can AI do everything so I don’t have to lift a finger?” What I think is “how can AI help me analyze things? Can it identify my weak spots?”

Full automation means replacing the teacher, but the value of your teaching experience shouldn’t be replaced by AI. Students can tell when something is AI-speak — all those emojis, that relentlessly positive but ice-cold tone, lifeless bullet-point lists. If everything you put out reeks of AI, and you use that to interact with students, how can you expect good results?

The right approach: let AI do what it’s good at (24/7 Q&A, instant feedback, knowledge retrieval), and you do what you’re good at (deep guidance, emotional support, experience-based judgment).

Build Your AI Clone

Every teacher should have their own AI clone. The process is simpler than you think:

1. Collect all your teaching materials — videos, articles, course notes — dump everything into one folder.
2. Write a script with the Gemini API to automatically transcribe all your teaching materials and do a rough categorization by content.
3. Once the categories look right, start distilling. Strip away the filler, keep only the teaching essence. My own 800,000 characters across 350+ files distilled down to 120,000 characters across 50-something files.
4. Go back and compare against the originals to make sure nothing important got lost in distillation.

One evening to consolidate over a decade of teaching essence, for less than 100 TWD in cost.

The resulting AI clone can work behind the scenes to help iterate on your courses and guide students, and on the front lines to handle after-class tutoring and Q&A. Most importantly, this clone’s output isn’t AI hallucination — it’s grounded in what the teacher actually knows and has taught.

15-20 USD a Day — Is It Worth It?

I’ve asked other teachers: if there were a tool that could fully replicate your thinking and answer all your students’ questions 24/7, and it costs 15-20 USD a day to run, would you pay for it?

A lot of teachers faint when they see the bill. But do the math: how much time does a teacher spend every day answering the same basic questions over and over? If you could free up that time for higher-value deep teaching and mentoring, 15-20 USD is actually dirt cheap.

And efficiency isn’t the only thing you get. Teachers can track students’ daily conversations with the AI in the backend, quickly gauging where each student stands based on what they’re asking. Before, you had to wait until a face-to-face meeting to find out where a student was stuck. Now you can see it every day.

Evangelizing AI to Teachers Takes Persistence

I have a colleague who teaches TOEFL writing. I designed a RAG knowledge-base AI teaching assistant for him. He was stunned: he’d only given me one sample — a one-hour teaching video — and it could already replicate his teaching approach and answer most student questions on his behalf.

But the evangelism process isn’t easy. Getting teachers on board with AI is a water-dripping-on-stone kind of effort — slow and steady. A lot of teachers have a natural resistance to technology, or feel like “I’ve been teaching for years, why should I change?”

What I bring to the table are my own real examples of teacher-AI collaboration that actually worked. I’m not selling a product. I want more teachers to coexist comfortably with AI. More importantly, I want teachers to break free from the “trading time for money” career trap and be liberated from repetitive tasks. There’s a lot more to life worth pursuing.

Open-Sourcing Beats Hoarding

I’ve put nearly all of my previously paid course content up on YouTube and my personal website for free. Surprisingly, the number of students coming to me didn’t decrease — it increased.

Another surprise: going open-source made me more creative than guarding everything ever did. People who are willing to give things away never have to fear being copied. Competitors can only copy your yesterday. They can’t copy your today, and they’ll never catch up to your tomorrow.

So I open-sourced Teaching Copilot too. It’s my first product designed with teachers as the primary users. Fellow educators are welcome to try it out.

Teaching Experiments I’ve Done and Am Doing

For anyone else in education who’s using AI, here’s what I’ve tried so far in bringing AI into my teaching:

1. I turned ten years of my teaching videos into transcripts and, together with my handouts and the teaching articles I’ve shared online, distilled them into a knowledge base of my own teaching system.
2. I plugged that knowledge base into a large model as RAG. At first it was my behind-the-scenes advisor, helping me iterate on handouts and draft answers to questions. Later I put it directly in front of students, so their day-to-day error reviews line up with how I, their teacher, actually think.
3. For the non-knowledge side, the advisory work of personal study planning, I had an agent help me build templates and a planning database. Add the months I spent personally correcting the model’s first drafts, and the planning database now matches about 95% of what I’d plan on the first pass. I only need to polish the last 5%.
4. I ran statistics on hundreds of students’ score reports and reverse-engineered the algorithm to find easier paths to a high score, including time management and question-allocation strategy.

And what I’m working on lately:

5. Using Colab and a large amount of my past teaching material to fine-tune two small open-source models. One is a TTS model so it can produce my voice. The other is an LLM so it can use my tone and style, plug into the RAG, and answer students’ questions as me.

Put those two together, and with streaming interactive voice models sure to improve fast over the next six months, a complete Dustin-the-teacher clone should be ready.

What We Need to Study Isn’t Just What AI Can Do

AI can already do too many things. Anything you can think of, just go ask it — more often than not, the answer exceeds your expectations. What we need to study now is the value of human-to-human connection — what are the core tasks we should keep in our own hands.

If you mistakenly hand that over to AI, you’ll dilute your own uniqueness and competitiveness. Every article, every piece of content you share should be written by hand. Don’t use AI-generated illustrations. At most, use AI to help rearrange your article structure or crunch large datasets. Because what you want is to personally reach your students.

I’m not just a teacher — I’m an architect of educational products. I hope more educators will join me in moving toward this vision.

Meanwhile, where are you as a teacher? Still negative and dismissive about AI? Or curious but stuck, not sure where to start?

My advice: keep the pain and the cost low, and learn as you go.