By Professor David Schmidtchen
When we think of ‘craft’, we call to mind objects of aesthetic beauty. They are beautiful because the craftsperson has matched the materials used with the object’s purpose, its environment, and how it will be viewed and used.
But craft is more than a metaphor for polished outputs. It describes a relationship between a person and their work, expressed in what they produce. For public servants, this is the relationship with values, ethics, internalised patterns of behaviour, the accumulation of contextual judgement, and an instinct for navigating the contradictions of public service.
Friction is the teacher
Craft knowledge cannot be acquired or absorbed by reading about it. As Richard Sennett shows in his study of skilled practice, craft is inseparable from working through resistance. It is the iterations between hand and mind, and between attempt and revision, that deepen understanding.
The initial, flawed draft and imperfect problem analysis are essential for learning and improving our skills. Often, we view these as frictions in the system that we seek to eliminate through technology's efficiency. However, this overlooks the value of friction itself.
Instead of removing friction, we should recognise its role in learning. While some friction, such as difficulty searching disorganised systems, causes frustration and adds little value, other forms - such as early drafts, testing assumptions, letting go of weak arguments, and defending decisions - are crucial to the development of craft. These constructive frictions foster growth and mastery.
Researchers call this the effort paradox. People avoid effort wherever possible, even though effort is precisely what gives work meaning and builds capability. When effort connects to outcomes, industriousness and persistence become rewarding in themselves. When effort seems disconnected from outcomes, learned helplessness follows. Neither is inherent to individuals; both are shaped by the work environment and the tools we use.
This brings us to AI
Novices and experts are not the same user.
An expert and a novice prompting the same AI system can produce the same output, but they are doing different things. Experts carry internal models against which to test the machine’s output, and they can recognise where a generated draft is weak because they have written a hundred such drafts. For the expert, AI is a lever.
The novice doesn’t have these models. What appears efficient isn’t. AI resolves difficulties without troubling the novice, yet these difficulties have been the foundation of the expert’s pattern library.
The novice and the expert produce comparable documents, but their relationship to the output and to their developing capabilities differs. AI disrupts the craft learning loop before it has formed.
A 2026 series of Random Control Trials (RCTs) with 1,222 participants studied AI's effect on mathematic reasoning and reading comprehension. Brief AI support improved immediate results but reduced independent skills and persistence after help was withdrawn. Similarly, among doctors detecting polyps, regular AI use led to a decline in detection skills without AI, raising concerns about potential ‘deskilling’ if AI isn't carefully managed.
The question for leaders
The question for APS leaders is not whether the APS should use AI. It is where AI should accelerate work, support learning, and where it should be deliberately delayed until a novice has wrestled with the task unaided.
The recipe is straightforward: use AI to remove administrative friction while preserving the experience of developmental friction. Design AI use so junior staff still practise the hard parts of public service judgement before the machine completes them. An APS that offloads its apprenticeship will look productive on every metric until, years later, there are no experts left, only machine operators.
When we over-rely on AI, we surrender more than we realise. We gradually forget that the struggle of learning was ever important. The novice and the expert may produce similar outputs using AI, but they are qualitatively different kinds of public servants.
APS craft is forged through the uncomfortable friction of learning. The task of leadership is to ensure AI is used to build craft rather than undermine it.