Learning is being renovated. AI sits in the middle of it, capable and unsettling in equal measure. The conversation about what to do swings between enthusiasm and prohibition, but the load-bearing walls have held through every renovation before this one.
day 75 of 2026·7 min read
The dust will not settle in our time. And when it does, some great roaring machine will come and whirl it all sky-high again.
— Samuel Beckett
I've been attempting to write a longer piece about how children use and relate to AI for a while now. It became Tales from the Synthetic City, which is mostly about the way these systems perform understanding without possessing it, and the strange pull of talking to something that seems to listen. But that piece deals with the relational danger. This one is about the structural question: what happens to learning itself when AI can produce the appearance of competence on demand?
If you have ever renovated a house while living in it, you know the particular disorientation of walking through rooms that are half-familiar and half-new. The kitchen bench is gone, but you still reach for where it was. Dust gets into everything. Someone assures you it will be better when it is finished, but right now you are eating dinner at a camping table in the lounge room, wondering whether anyone has a plan.
Learning is being renovated. Classwork, homework and assessment are changing. AI sits somewhere in the middle of it all, capable and unsettling in equal measure. Yet the conversation around technology in education tends to swing between breathless enthusiasm and blunt prohibition, as though the only choices are to knock down every wall or refuse to touch the house at all.
The load-bearing walls
Any sensible renovation starts by identifying what is structural. In learning, the load-bearing walls are the things no tool can replace: students thinking carefully, explaining their reasoning, working through difficulty, building understanding. A student who can articulate how they arrived at an answer, who can revise their thinking when challenged, who can transfer what they know to an unfamiliar problem; that student has learned something. The evidence is in their reasoning, not in the polish of what they hand in.
AI can produce polished work quickly. It can write a coherent essay, draft a cover letter, generate an authoritative summary. If schools assessed only the finished product, they would have a problem. But the shift towards process, towards drafts, conversations, and discipline-specific thinking, was underway well before ChatGPT became mainstream. AI has made the shift more urgent, not more radical.
Plain prompt-and-response work is passé.
AI is already moving well beyond the simple act of prompting a chatbot for a response. I could send a single message from my phone to an AI agent I have built, or more accurately, previously asked AI to build, asking it to draft five articles from the work I have been doing this term. That agent could then pass each draft to a researcher, a fact-checker, a citation agent, an anti-synthetic prose agent, and finally an editor, leaving five polished pieces waiting for me by the time I finish my coffee. That kind of workflow scarcely existed 24 months ago. In another 18 months, it may already seem primitive.
Perhaps I will not be writing text prompts at all. Even now, I could shift to a combination of voice, vision, and interactive control. It is already possible, and not much harder than relying on these human hands to hammer keys, keystroke by keystroke. I might simply point my phone camera at parts of a website, or just open it on my phone, and say, 'Change the header. Update the interface to reflect current design trends and accessibility standards. Make the site more interactive.' What feels impressive now will soon look routine to anyone paying attention, and I suspect that text prompting will not remain the primary interface for much longer.
This is why teaching students a specific set of AI skills, such as how to write a prompt or use a particular tool, can be useful in the short term but short-sighted if we treat it as the destination. The platforms a student learns on this year may not exist in three years, and I am interested to see whether their interactions with devices, and with what is online, change again. The history of technology suggests they will, and the increasing pace of development encourages me to believe that these new ways of consuming, interacting, communicating, and creating will evolve alongside them. What will not change is their need, and our desire for them, to think critically, evaluate the information with which they are constantly bombarded, and develop the understanding required to judge whether an answer, claim, or argument is worth its salt.
What this looks like in practice
A history teacher asks a student to explain why they selected certain sources and what they left out. A mathematics teacher watches a student work through a problem and asks them to describe where they got stuck. These are not workarounds for AI. They are what good assessment has always looked like.
Leon Furze argues that assessment should build trustworthy evidence of learning over time, valuing process, reasoning and professional judgement rather than relying on single high-stakes products (Furze, 2025). Research from TEQSA reinforces the same point: assessing what students can actually demonstrate is more reliable than trying to detect what they might have offloaded (Lodge et al., 2025).
AI use also looks different across disciplines. The way a science student evaluates an AI-generated claim is not the same as how a music student uses AI to experiment with composition. Nick Potkalitsky's work on disciplinary AI literacy is worth reading here, particularly the idea that each subject area needs its own conversation about what AI can and cannot do within its particular way of knowing (Potkalitsky, 2025). A generic 'AI skills' program will date as fast as the tools themselves. The conversations that last are the ones that live inside each discipline, because that is where they matter and where they last.
The question families keep asking
Families are working through their own version of the renovation, trying to understand what is allowed, what is helpful and how to have useful conversations about it at home. I think the most useful thing a parent can do does not require understanding any school's AI policy in detail. When your child finishes a piece of work, instead of asking 'Did you use AI?', try asking 'How did you figure that out?' or 'What was the hardest part?' Those questions signal that you value the thinking, not the product. They open a conversation rather than an interrogation. And they mirror what the best classrooms are already doing.
The dust may never settle
I have struggled with the idea that the dust may never settle. During any renovation I have done, there is a moment when you put the tools away and admire the finished room. I am not sure that moment is coming. It is not only the pace of change but the sheer number of directions it is branching into. AI is not one development moving in one direction. It is a dozen developments moving simultaneously, each with its own implications.
But I am coming to terms with it. The load-bearing walls have held through every renovation before this one. We have moved from slate boards to exercise books to word processors to whatever comes next, and the things that matter most in learning have stayed. Students who can think, reason, question and explain their way through a problem will be well served regardless of which tools surround them.
The renovation is not finished. It may never be. But the structure is sound.
References
Beckett, S. (1957). All That Fall. Faber and Faber.
Furze, L. (2025). Five principles for rethinking assessment with Gen AI. Leon Furze. https://leonfurze.com
Lodge, J. M., Bearman, M., Dawson, P., Gniel, H., Harper, R., Liu, D., McLean, J., Ucnik, L. & Associates. (2025). Enacting assessment reform in a time of artificial intelligence. Tertiary Education Quality and Standards Agency, Australian Government.
Potkalitsky, N. (2025). Six territories for disciplinary AI literacy. Disciplinary Specific AI Literacy [Substack newsletter].