I just read Mark House’s recent post about education and felt the tingle of recognition. I worked with Mark, and lived through some of those years alongside him. Forty years of incremental change. Exam scripts now scanned instead of posted. Markers working on screens instead of paper. Results landing in inboxes instead of brown envelopes. The bits have changed. The atoms — the underlying shape of what learning actually is — have barely moved.
Mark is writing about assessment. But the same diagnosis applies to the whole of language education. And I think about this constantly, because at Wall Street English I’m responsible for the product and technology that hundreds of thousands of learners use every day.
We did something unusual at Wall Street English. We changed some atoms decades before AI arrived. The core of our method is a flipped model: learners engage with structured content independently before coming to a live session with a teacher. The teacher doesn’t deliver a lesson. They create the conditions for language to happen, skill to develop, and confidence to grow. It’s a different role, with different skills, and it produces different results.
That was the original atom change. But here’s the thing — we’ve been running that model with fixed atoms ever since. The method works. The question now is what else can shift around it.
The personalisation atom
The most obvious candidate is curriculum. Not personalisation in the weak sense — you prefer videos, so here’s a video. Real personalisation: adaptive sequencing, dynamic difficulty, vocabulary tracking that distinguishes what a learner consistently misses from what they’ve genuinely internalised. An AI system can build and maintain a learner model that no human teacher has the bandwidth to hold across a full caseload.
That changes the shape of the path each learner walks through the method — without changing the method itself.
The structure atom
A 45-minute session made sense when scheduling was the constraint. But if a learner is available for 20 minutes at lunch and another 30 on the commute home, the constraint isn’t the learner — it’s the format. AI can restructure content around when and how a person learns, not around when a classroom is available.
The same applies to format. Some learners internalise grammar through explicit explanation. Others need it to emerge from exposure. Some need heavy scaffolding on pronunciation before they’ll take risks in conversation. An AI system that understands a learner’s profile doesn’t have to serve the same UX to everyone. The experience of learning can be different — not just the content within it.
The teacher atom
This is where the conversation about AI in education most often goes wrong. The assumption tends to be that AI replaces the human or, at best, supplements them — takes the marking, frees up time for more human work. That’s the bits argument again.
The atoms argument is different. If AI handles the adaptive curriculum, the progress tracking, the content assembly — what it hands back to the teacher isn’t just time. It’s leverage. A teacher who deeply understands a learner’s profile, with AI surfacing where that learner is stuck and why, is a different kind of professional. Their expertise becomes a superpower rather than a constraint.
At Wall Street English, this is the direction we’re building toward. Teachers are the core of the method. They always will be. The academic foundations don’t change. The relationships that make a learner commit to their practice don’t change. What AI changes is a teacher’s ability to act on what they know. They stop being the person who also has to track fifteen learners’ vocabulary gaps manually. They become the person who can respond because those gaps are already visible.
What actually changes, and what doesn’t
Mark ends his post with a line I keep returning to: after four decades, we finally have tools that could change more than just the stationery — if we choose to use them that way.
The word if is doing a lot of work there.
Most organisations will use AI to optimise the model they already have. Faster feedback. Lower cost per interaction. Better efficiency on the same assessments. That’s not nothing. But it isn’t the change education has needed.
The institutions that matter in ten years will be the ones that use AI to ask a harder question: which of our atoms are actually load-bearing?
In language learning, the answer is clear. The method is load-bearing — the human interaction at the core, the applied linguistics underpinning the design, the relationship between teacher and learner. Those atoms hold, and they should.
Everything else is a candidate.
Inspired by Mark House on LinkedIn.
Leave a Reply