On this page
I saw a discussion online recently: “Too many students come in with unrealistic expectations. They think one class will make them master a skill.” Plenty of instructors agreed in the comments.
I want to offer the other angle: both sides are at fault.
I’ve taught GMAT for 15 years, and I’ve recently started teaching Claude Code. I understand the “students want to magically improve overnight but don’t see the concept-building and accumulation underneath” complaint. But I’ve also taken a lot of online classes myself, and I’ll say this responsibly: a lot of instructors genuinely don’t know how to teach.
Want AI working in your own job, not just in demos? See the 3-month 1-on-1 program →
The basic bar for teaching quality
A few basics:
- Are you reading off the slides, or can you riff based on what students are showing you in the room?
- Do chapters connect logically? Is there designed practice time so students can keep up?
- When students hit the obvious follow-up question, can you catch it? How do you handle the off-script ones?
A lot of “instructors” fail these. But online they’re marketed as amazing. My test for a teacher’s actual depth is the Q&A session. Do they have real substance or not, and that’s when it shows. A teacher with real accumulation can handle questions across skill levels and explain in a structured, logical way that helps the asker understand.
Plenty of teachers can’t do this. And they compensate by selling faith.
Course marketing also needs expectation management
If students shouldn’t have unrealistic expectations, why do instructors overpromise in course marketing? “This class will make you master XXX.”
Setting boundaries and managing expectations. Isn’t that also the course producer’s job? But a lot of course marketing will say anything for reach these days.
“This class will make you master Claude Code” vs. “This class gets you started with Claude Code. The rest is on you and your practice.” Sounds similar, but the integrity gap is huge.
Responsibility > Ability
A teacher’s most basic responsibility is to think from the learner’s angle: what do they need?
I’ve recently started teaching Claude Code, and I know my depth isn’t where I want it yet. But I still do learner surveys, and I design lesson plans and slides with each person’s needs in mind. I try to give every student customized suggestions.
This has nothing to do with ability. It’s pure responsibility. I’ve seen plenty of capable people with no responsibility, and learners who are still building their capability but have strong responsibility. Students feel the difference clearly.
With the teaching mindset covered, back to the tools: plenty of cram school peers have also been asking me which AI they should use to assist their teaching.
Cram school AI transition: the one-word answer
I’ve been a cram school teacher for 15 years, and I’ve also consulted for humanities/sciences cram schools on their AI transition.
One-word answer: Claude.
Not Gemini, not GPT, not some locally-hosted small model. The reason deserves a longer post, but in practice: for a cram school teacher’s workflow (lesson prep, question authoring, grading, personalized student suggestions, admin paperwork), Claude’s answer quality and agent reliability are the only options that hold up at this stage.
I’ve tested the others. Choosing Claude isn’t faith.
At the end of the day, whether it’s a teacher teaching or a child learning on their own, the biggest trap is turning a tool or a theory into an article of faith.
Protecting Curiosity in Learning
A few days ago I made a decision: I threw out my kid’s copy of “Attention Is All You Need.”
The kid cried. But I hope this was the right call. I hope he rediscovers his curiosity for learning across the board, instead of treating some specific book as a “must master” article of faith. In the AI era, for a beginner, understanding architecture matters less than keeping your curiosity alive. Papers are always findable; curiosity, once lost, is much harder to recover.
Hope it’s not too late.