"Productive struggle" has become an educational buzzword — invoked frequently in pedagogical discussion, understood imprecisely, and abandoned quickly when it produces the friction, confusion, or low performance metrics that it is supposed to produce. We have built Aethel around it anyway. Not because we are indifferent to user experience, and not because we believe difficulty is inherently virtuous. But because we have understood "productive struggle" more carefully than most, and because the understanding leads to specific design choices that we want to explain.
I want to begin with what the phrase does not mean, because the misreading is common and has real consequences.
Productive struggle does not mean confusion without direction. It does not mean being stuck without recourse, frustrated without feedback, or wrong without any means of knowing how to become right. It does not mean that struggle is productive by virtue of being struggle — that the harder something is to do, the more learning it produces, and therefore the best educational design is the most difficult one. This version of the idea is both intellectually unjustified and experientially harmful, and it is, in various subtle forms, what most implementations of "productive struggle" actually deliver.
The phrase was introduced into mathematics education research primarily through the work of James Hiebert and Douglas Grouws, who used it to describe a specific condition: the state of working on a problem that is within reach but not yet solved, that requires the learner to engage with mathematical concepts at a level slightly beyond what they can currently execute automatically. The word productive is doing real work here. The struggle is productive not because it is hard, but because it is hard in the right way, at the right time, with the right support. Take any one of those conditions away and the struggle becomes unproductive — demoralising rather than developing, confusing rather than clarifying.
The Neurological Foundation
Before we discuss what Aethel does, it is worth being clear about why productive struggle works at the neurological level, because the mechanism informs the design.
Memory and understanding are not products of passive exposure to information. They are products of active neural processing — of the cognitive operations that occur when the mind is required to do something with information rather than receive it. The hippocampus, which plays a central role in the formation of new memories, is not activated equally by all cognitive states. It is activated preferentially by states of uncertainty, prediction error, and effortful processing — by precisely the conditions that characterise productive struggle.
When you are working on a problem you cannot immediately solve, your brain is generating predictions, checking them against what you know, identifying where the predictions fail, and searching for alternative approaches. This process of prediction, failure, and search is the process that drives hippocampal encoding. The struggle is not an obstacle to learning; it is the condition that activates the neural processes by which learning occurs.
Conversely, when you receive a correct answer to a question you have not yet attempted, the prediction error that drives encoding does not occur. The information is processed — you understand that it is correct, you follow the logic — but the depth of encoding is substantially lower than it would have been if the prediction process had preceded the answer. The neuroscience explains what the behavioural research documents: the effort of trying precedes and enables the learning of the answer.
This is the foundation on which Aethel is built, and it is not flexible. You cannot improve on this mechanism with better explanation, more comprehensive answers, or more sophisticated AI. The mechanism is in the trying, not in the receiving, and no enhancement of the received content changes this.
What Makes Struggle Productive
The difference between productive and unproductive struggle is not a matter of degree — it is a matter of kind. Three conditions are necessary.
The first is calibration: the task must be within the learner's Zone of Proximal Development — beyond what they can do independently but within what they can do with appropriate support. Struggle on a task that is too far beyond current competence is not productive; it is demoralising and produces no useful prediction error because the learner has no basis from which to generate useful predictions. Struggle on a task that is within current independent competence is not productive either; it is tedious rather than challenging. The calibration is everything, and it is also extremely difficult to get right.
The second is feedback. Productive struggle requires the learner to know, at some point, whether the thinking they did was on track and where it fell short. Without this feedback, struggle produces confusion without direction: the learner knows they have not yet succeeded but does not know how to revise their approach. Feedback in productive struggle is not the provision of the answer; it is the provision of information about the thinking — not "the answer is X" but "the approach you are using runs into this difficulty" or "there is a distinction you have not yet made that matters here."
The third is support at the right threshold. When genuine impasse occurs — when the learner is not making any progress and does not have the resources to do so — some form of intervention is necessary to restore productive struggle rather than prolong unproductive frustration. The intervention should be the minimum that restores forward movement: a question that redirects attention, a hint that opens one path without closing others, a partial observation that gives the learner something new to work with. The support is not designed to resolve the struggle; it is designed to keep the struggle productive.
All three of these conditions are hard to maintain. Getting them right requires judgement rather than algorithm, responsiveness to the specific learner in the specific moment rather than application of a general rule. It is among the most difficult things that good teachers do. And it is what we have been working to build Aethel to do.
How Most Tools Fail at This
The failure mode of most educational tools — and most AI learning tools specifically — is straightforward to describe and difficult to avoid: they optimise for the resolution of struggle rather than its maintenance.
When a user is confused, the most natural response for a well-designed conversational system is to resolve the confusion: to explain the concept more clearly, to provide an example, to give the answer and then check whether the user understands. This response is intuitively correct, technically achievable, and educationally counterproductive at scale. The confusion that the tool resolves is the confusion that would have driven productive engagement if it had been held open rather than closed. The explanation that seems helpful is, frequently, the substitution of the tool's understanding for the understanding the user was in the process of developing.
There is a related failure mode that is more subtle: providing scaffolding that never withdraws. A tool that is always available to assist, always ready to simplify, always responsive to the signal that the user is struggling will, over time, train the user to lean on the scaffolding rather than develop the capacity the scaffolding was supposed to support. The scaffolding is still there; the building that should be standing independently is not.
We have worked hard to avoid both of these failure modes, with mixed success and ongoing revision. The first — resolving struggle prematurely — we address by responding to confusion with questions rather than explanations. When a user signals that they do not understand, our first move is to ask what specifically they do not understand, what they have tried, what their current model of the problem is. This response often frustrates users who wanted an explanation. It also keeps the struggle going rather than ending it, which is the point.
The second — scaffolding that never withdraws — we address through a progressive reduction in support as the user demonstrates competence. Concepts and problems that a user has engaged with successfully receive less support in subsequent encounters; the tool's assistance reduces as the user's capacity grows. This is harder to implement well than to describe, and we are candid that our current implementation is imperfect. It is, however, the right direction, and we are committed to moving further in it.
The Product Decisions This Has Forced
Building around productive struggle has forced a set of product decisions that are worth making explicit, because they are decisions to not do things that most users would like us to do and that our competitors do or will do.
We do not provide comprehensive explanations on request. We provide partial explanations, directed questions, and observations that open paths rather than close them. A user who comes to Aethel and asks to be explained a concept will receive a response that begins with their own understanding rather than with ours.
We do not produce complete worked examples. We produce partial worked examples with deliberate gaps — places where the learner is expected to fill in the next step, identify the error, or articulate the principle being applied. The gaps are not accidents; they are the mechanism.
We do not allow users to skip the struggle. When a user indicates they want to move past a concept they have not yet engaged with — to get to the answer, to skip the problem, to just be told what to do — Aethel does not comply. It redirects. This is the most-complained-about feature of the product. It is also the feature we are least willing to change.
We do not optimise for session satisfaction metrics. This is perhaps the most consequential decision, and the one that puts us most directly in conflict with conventional product wisdom. Satisfaction metrics — star ratings, willingness to return, immediate post-session feedback — will always favour tools that resolve struggle over tools that maintain it. The session that ends with a clear explanation feels better than the session that ends with an ongoing productive question. The learning that follows from the second session is substantially better. We have chosen to optimise for the second.
The Research Behind the Principle
We want to be specific about the research that underpins our design, because "productive struggle" as a phrase has accumulated enough vague endorsement that the specific findings behind it are sometimes lost.
The most directly relevant body of work is Manu Kapur's research on productive failure, conducted primarily at the National Institute of Education in Singapore and later at ETH Zürich. Kapur's studies consistently found that students who were asked to attempt to solve problems before receiving formal instruction — students who were therefore guaranteed to fail, or to produce imperfect solutions — performed worse on immediate tests but significantly better on delayed tests and transfer problems than students who received instruction before attempting the problems.
The explanation Kapur offers is specific: the failed attempt at the problem activates and organises relevant prior knowledge, creates cognitive structures that the subsequent instruction can attach to, and produces awareness of the specific gaps that the instruction needs to fill. The instruction that follows a genuine attempt at the problem lands differently than instruction that precedes any attempt. It has somewhere to go.
This finding is not confined to mathematics. It appears in language learning, in science education, in professional training contexts, and in studies of adult learning. The generality of the effect suggests that it reflects something fundamental about how instruction and cognitive engagement interact — that the mind that has struggled with a problem, even unsuccessfully, is a different learner for the subsequent instruction than the mind that has not.
Robert Bjork's work on desirable difficulties — the class of learning conditions that appear to impede performance in the short term while enhancing long-term retention and transfer — provides a complementary account. Bjork identifies several conditions that produce this pattern: spacing study sessions, interleaving different types of practice, reducing feedback frequency, and varying the conditions of learning. In each case, the condition that produces better long-term outcomes is the condition that makes the learning harder and slower in the moment. Productive struggle, in Bjork's framework, is not one technique but the general principle that underlies a family of techniques.
A Direct Address to Our Users
We know this is harder than the alternatives. We know there are moments when Aethel's insistence on questioning rather than answering, on holding open what you wanted closed, on requiring effort rather than accepting passivity, is genuinely frustrating in a way that does not feel productive from the inside.
We are asking you to trust the mechanism even when the experience is uncomfortable. Not blindly — the mechanism is well-documented and we have described it explicitly in this piece and elsewhere, so the trust we are asking for is informed rather than faith-based. But trust nonetheless, in the specific sense of choosing the discomfort of productive struggle over the comfort of easy explanation even when easy explanation is available and the discomfort is real.
What we are not asking you to trust is that all discomfort is productive, or that Aethel's calibration is always right, or that the struggle we are putting you through is always at the right level. We are imperfect at this. We miss calibration sometimes; we hold struggle open too long, or not long enough, or in directions that are not useful. We are getting better at this, and we are committed to continuing to get better.
What we are confident of is the principle: that learning is constituted by the effort you make rather than the information you receive, that tools which eliminate effort are eliminating the mechanism rather than the obstacle, and that the correct design response to this is to build a tool that costs something to use — that requires genuine cognitive engagement rather than passive reception — and to be honest about why.
What Productive Struggle Actually Produces
There is a specific experience that users report, sometimes, after a difficult session with Aethel — an experience that is different from the satisfaction of a session that went smoothly and ended with clear answers. It is something like the experience of having genuinely understood something, as distinct from having received an explanation of it. The understanding feels different: more active, more connected to other things known, more available for deployment in new contexts.
This is not an accident. It is the predictable consequence of having engaged in the cognitive operations that produce deep encoding rather than the reception of information that produces surface familiarity. The understanding that productive struggle produces is more durable, more transferable, and more genuinely yours than the understanding that explanation provides.
There is also something that takes longer to describe and that we have observed repeatedly in users who work with Aethel over extended periods: a shift in how they relate to their own confusion. Confusion, which is initially experienced as a signal to seek help — to find an explanation, to receive an answer — gradually becomes something closer to a productive state: a signal that there is something here worth examining rather than resolving, that the confusion is not an obstacle but a direction. This shift is, in our experience, one of the most valuable things a learning tool can produce. It is also one of the least likely to show up in a session satisfaction metric.
The goal we are working toward is not a product that feels good to use, although we care about experience and we are constantly improving it. The goal is a product that produces learners who are more capable than they were before using it — not more capable at using the product, but more capable at the thinking the product was designed to support. Learners who, when Aethel is absent, can do things they could not do before. Learners who have internalised, to whatever degree this is possible, the questioning orientation that the product models. Learners who are, in the old phrase, more their own teachers.
Everything else we have built follows from this conviction. The questions instead of answers. The gaps in the worked examples. The refusal to simply tell you. All of it is in service of the same goal: not to help you in the moment, but to make you more capable in every moment after.
We are building a tool designed to make itself unnecessary. Not by providing everything you need, but by producing in you the capacity to do without it. This is a strange ambition for a product, and we think it is the only honest one for a learning tool. Productive struggle, correctly calibrated, is the path. Everything else is a more comfortable detour.