I have spent this series writing about AI in education as an outside observer — analysing trends, citing research, drawing conclusions. But I am not an outside observer. I run product at Wall Street English, a global language learning business operating across 70+ countries.
We deploy AI. Daily. At scale. And the lessons from doing it are different from the lessons you read about in vendor brochures.
The Context
Wall Street English is a hybrid model. Structured curriculum delivered digitally with personalised learning paths, 350+ centres filled with human educators for conversation and motivation, and technology that holds it all together. That combination gives us a specific vantage point: we want to optimise but don’t pursue full automation, and we value the human educator most of all. The interface between them is where most of the interesting learning happens.
What Is Working
Authentic Conversation Practice
The feature that has worked best is also the most genuinely novel: AI-powered conversation practice that simulates real-world scenarios.
Students practice the kind of conversations they actually need — job interviews, business meetings, everyday situations — with an AI partner that responds naturally and adapts to their level. The result is practice that is personalised, available on demand, and free of the social anxiety that often inhibits adult learners in group settings.
Students rate these sessions dramatically higher than the previous equivalent exercises. The shift is not marginal. When something triples in preference, that is not an incremental improvement — that is a fundamentally different experience.
CoachBot
We have deployed a conversational interface that lets students interact with the platform in natural language. Ask when your next class is. Get the definition of a word. Find out what you need to do next. Submit a long-form writing exercise and receive feedback.
The functionality is not new. What is new is the interface. A dashboard that shows you the same information requires you to know what to look for and where to find it. A conversation requires only that you know what you need. For adult learners balancing language study with work and life, that difference matters more than it might appear.
CoachBot is currently student-facing. The plan is to extend it to educators as well.
Instant Assessment — The Vision We Can Now Build Toward
We are designing an AI-powered assessment layer that would act as a teaching assistant in every classroom — documenting what happens during a session, surfacing relevant information for teachers, and distributing useful summaries to students.
This does not exist yet. But for the first time, it is imaginable — and that matters. The cost and complexity that previously made this kind of ambient, intelligent classroom presence impossible are falling. We are building toward something that, two years ago, we would not have put on a roadmap because it would have felt like science fiction.
What We Have Learned
The Educator Relationship Is the Product
There was never a serious conversation at Wall Street English about replacing teachers with AI conversational agents. Not because it was technically impossible — but because it would have missed what students are actually choosing. What they value most.
Adult language learners are not buying access to content. They are buying progress, accountability, and human connection. The educator is not a delivery mechanism for the curriculum. The educator is the value. AI that ignores this will fail regardless of how sophisticated it is.
AI-Generated Curriculum Needs Expert Architecture
AI can generate lesson content. We have explored this. The gap is not in the generation — it is in whether the output reflects any real understanding of how people learn.
The model that works is not AI replacing curriculum designers. It is AI working within a framework that curriculum experts have built: codified learning principles, defined quality criteria, clear guardrails. The expert’s role shifts from writing every lesson to creating the environment in which AI can write lessons well. High-level control, not line-by-line approval. That is a meaningful change in how educational expertise gets applied — and it is a more interesting job, not a lesser one.
Product Design Fundamentals Do Not Go Away — They Become More Important
Before AI, we built tools for teachers that should have worked and did not. Rich dashboards with detailed student data. Feedback systems with extensive logging. The failure modes were consistent: heavy data entry that people were unwilling to sustain, and complex views full of information that was not easy to scan or sufficiently actionable.
These are not AI problems. They are product design problems. And they predate AI entirely.
What AI changes is the scale at which you can make these mistakes. When creation is cheap, the temptation to add more — more data, more features, more change — is harder to resist. The functional, social, and emotional dimensions of why people use tools do not become less important when you can build faster. They become more important. The cost of violating them just gets easier to incur.
What Surprised Us
Students Chose Conversation Over Everything Else
I expected AI practice tools to be used primarily for grammar drills and structured exercises. Students chose conversation simulation at a rate that surprised me. The authentic conversation feature consistently outperforms previous equivalents by a wide margin — and the preference is not close.
The lesson is one that product people know but regularly forget: watch what people do, not what you think they will do.
The Franchisees Became the Loudest Advocates
Wall Street English operates on a franchise model. The people who run the schools are my customers. I expected the strongest enthusiasm for AI innovation to come from educators or students. It came from franchisees.
The reason is commercial, as well as pedagogical. AI gives them a better story to tell: a more personalised product, a more modern method, a clearer demonstration of value. When the technology improves what you sell, the people selling it notice first.
The Ambition Unlock
The biggest surprise was not a product outcome. It was an organisational one.
When the cost of building something innovative drops significantly, the business starts imagining things it had stopped imagining. Features and experiences that were shelved as too expensive, too complex, or too risky are back on the table. The energy that generates — across product teams, educators, business owners and the Senior Leadership Team — is real and it is significant.
We are not just building differently. We are dreaming differently. And after years of working within the constraints of what was affordable, that is a more profound change than any individual feature.
The Honest Summary
Two years in, here is what I actually know: one AI feature is working exceptionally well, one is live and useful in a more modest way, and we are designing toward several things that are now imaginable for the first time.
That is enough. The organisations that are winning with AI in education are not the ones that automated everything. They are the ones that found the right first problems — and learned enough from solving them to know what to build next.
The question is never AI or teachers? It is which tasks belong to AI, and which belong to humans? Getting that division right, and revisiting it constantly, is the work.
Actionable Takeaways
- Start with the relationship — understand what students are actually buying before deciding what AI can replace or enhance.
- Codify before you generate — AI-generated content is only as good as the expert framework it operates within.
- Watch what people do — your assumptions about use cases will be wrong. Look for it in behaviour, not intention.
Leave a Reply