Fundamentals
Two teachers, the same subject, the same textbook, the same group of teenagers. One class understands it and the other does not. Whatever accounts for that gap is not the content — the content was identical. It is the thing this page is about.
The word
In English the word sounds either academic or faintly insulting — “didactic” has picked up the sense of preachy. In German, Didaktik is an ordinary word for a serious field: the study of how something is opened up so that a particular person can actually understand it. Not what to teach, which is the curriculum. Not who is being taught, though that matters enormously. How the thing is made learnable. The tradition even has a standard picture for it — teacher, learner and content as three corners of a triangle, with didactics living in the relationships between them rather than at any corner.
Content tells you what there is to learn. Didactics decides whether anyone does.
Why it is treated as a discipline
Worth knowing, because it is the clearest argument against treating teaching method as a feature to be added later.
In the seventeenth century the Czech educator Jan Amos Comenius wrote the Didactica Magna, the Great Didactic, proposing something startling for its time: that teaching well was a subject that could be studied and improved, rather than a personal knack some people had. Much of what he argued for — start from the concrete, move in an order that respects how understanding builds, do not make children memorise what they do not understand — would not be out of place in a teacher-training course now.
The German-speaking tradition carried that on and made it its own field. In the twentieth century Wolfgang Klafki put a question at its centre that still reads as uncomfortably demanding: what is it about this particular content that makes it worth a person’s time, and what does understanding it actually open up for them? Not how to cover the syllabus. What the material is for.
That is the background we work in. It is also why we find it strange when teaching method is treated as something to bolt onto a content platform once the video hosting works.
What the craft consists of
None of these are about knowing the subject better. They are about what you do with a person in front of the subject — and every one has been studied, named and tested.
Support that never comes away produces someone who can only work with support. The researchers Wood, Bruner and Ross gave this the name scaffolding in 1976, borrowing the builder’s word deliberately: scaffolding is temporary, and taking it down is part of the job. Most courses are good at putting it up and have no plan for removing it.
Beginners learn more from studying a fully worked example than from attempting a problem with no idea how to start — a well-replicated finding from cognitive load research. But that advantage reverses as people improve: keep showing complete solutions to someone who is ready to try, and you are now getting in the way. So the steps come out one at a time until the learner is doing all of them.
Sometimes the right move is the opposite: hand someone a problem before they have been taught how to solve it and let the attempt fail. Manu Kapur’s research on productive failure finds this can beat being taught first, because the failed attempt shows the learner exactly what they are missing, and the explanation then lands in a gap they can feel. The failure has to be followed by the teaching, though. On its own it is just failure.
Ask what will happen and why, before demonstrating it. Predict, observe, explain — a sequence set out by White and Gunstone for science teaching. A prediction you have committed to makes you notice the result in a way that watching passively never does, and the prediction is not marked. Being wrong is the mechanism, not a failure of it.
Practising one type of problem twenty times feels efficient and produces less durable learning than mixing types together. Mixing forces you to work out which method applies before applying it — and deciding which tool to reach for is the skill that actually transfers out of the classroom. It feels worse while you do it. It works better afterwards.
The idea most people are never told
Look again at the five moves above. Failing at a problem before the explanation, and failing at it after, use exactly the same pieces — a hard problem and an explanation — and are not the same teaching at all. One produces a learner who now knows what they were missing. The other produces a learner who feels caught out. Predicting before you see the demonstration, rather than after, is the entire difference between the prediction doing something and the prediction being a formality. Order is not the packaging around a lesson. A great deal of the teaching lives in it, which is why a course that has all the right components in the wrong sequence can be genuinely bad while looking complete.
Same pieces. Different order. Different lesson.
The most useful moment
This is where teaching is either done or skipped, and it is the moment most learning technology handles worst — usually with a red cross and the correct answer, which is the one response guaranteed to teach nothing.
Not a verdict. A learner who thinks a 20% rise followed by a 20% fall returns you to the start has a specific, nameable belief — and it is a reasonable one until someone shows otherwise. Marking it incorrect addresses none of that.
Experienced teachers know where their subject trips people, and they aim at those places deliberately. That knowledge is some of the most valuable a school or company holds, and it is almost never written down anywhere.
Being told the answer ends the thinking. A hint that gets someone unstuck while leaving them to finish is harder to write and worth considerably more than the answer was.
Black and Wiliam’s work on formative assessment made the case plainly: checks that adjust what happens next help learning, while checks that only record a result mostly do not. Same question, different purpose, very different effect.
How we approach it
Everything above is the field, not us. Here is what we think follows from it for anyone building software that teaches.
Software can carry out a didactic decision. Choosing what kind of lesson someone needs should follow from reasoning a person can read and disagree with, rather than from a system’s unexplainable judgement about them.
What gets taught comes from the institution’s own material, approved by the people responsible for it. We are trying to give teachers more reach, not to write their subject for them.
A learner can ask why a lesson was built the way it was and get an answer. A teaching choice nobody can name is a teaching choice nobody can improve.
There is one line underneath all three. A language model can write the words of a lesson well. It should not also be the thing deciding what kind of lesson you need — that decision shapes how a person comes to see themselves as a learner, and it ought to rest on reasoning that can be shown, questioned and improved by the educators responsible for it.
Where this sits
What is there to teach, how do people actually learn it, and how should this be taught to this person. None of the three answers the other two, and most learning technology has an answer to only the first.
The other two: knowledge graphs, on holding a subject as a structure rather than a pile of documents, and adaptive learning science, on what is known about teaching people differently. All of it runs on Atlas and reaches people through Levio, Rise and Stage.
For teachers and training leads
Every subject has one. It is a far better conversation than a demo, and it tells you more about whether we know what we are talking about.