For most of the history of formal schooling, the scarce resource was the answer. Knowledge lived in a small number of places — a teacher, a textbook, a library — and the job of education was to move it into a student’s head and then verify that the move had happened.
That model has been under pressure for thirty years. Generative AI has not created the problem; it has removed the last excuse for pretending the problem does not exist. A student with a phone can now produce a competent-looking answer to almost any question a school is likely to ask them, in seconds, in whatever register the assignment demands.
Schools have three available responses. They can try to police it, which fails. They can ignore it, which fails more slowly. Or they can change what they ask students to produce.
The assessment problem is really a curriculum problem
Most of the current anxiety in schools is framed as an assessment problem: how do we know the student wrote this? But assessment is downstream. If the only thing a student is asked to produce is a text that demonstrates they understood something, then any technology that produces plausible text will break the assessment. The problem is not the detection technology. It is that the artefact being requested was always a weak proxy for capability.
Project work does not have this weakness in the same way. A student who has built a working sensor network can be asked why they chose that sampling interval. A student who trained a model can be asked what it gets wrong and why. A student who deployed an application can be asked what happened the first time somebody else used it. These questions have no plausible-sounding shortcut, because the answers are specific to a process that actually happened.
This is not an argument that AI should be kept out of student work. It is an argument that the artefact should be substantial enough that using AI on it is a skill rather than a substitution.
What “project-based” has to mean to be worth anything
The phrase has been diluted by a decade of marketing. A worksheet with a coloured border is not a project. Neither is a demonstration the teacher already knows the outcome of, nor a build that comes in a box with instructions.
For project work to carry the weight now being placed on it, four things have to be true.
The outcome must be genuinely uncertain at the start. If everyone in the class will produce approximately the same thing, the students are executing, not designing. Uncertainty is what creates the decisions, and decisions are what create learning.
The thing must actually work, or visibly fail. Software that runs. A mechanism that moves. A model that classifies, correctly or otherwise. Reality is the most patient and least negotiable assessor a school has access to, and it is free.
The problem should be locatable. Abstract problems produce abstract effort. A student who is measuring the air quality in their own corridor, or modelling the traffic at their own school gate, is working on something whose answer they actually want to know.
There must be a second version. Most of the learning in any technical work happens between the first attempt and the second. Curricula that only budget time for one pass systematically skip the part where understanding forms.
What this asks of a school
The honest answer is: more than a worksheet does.
Project work takes longer per unit of syllabus covered. It requires teachers who can hold uncertainty in a room rather than resolve it immediately. It needs a clearer definition of assessment, because “did it work” is necessary but not sufficient — a student can produce a working artefact with poor reasoning, and a student can reason excellently about something that never quite ran.
It also requires curriculum design that treats projects as the spine rather than the reward. A programme that teaches theory for eight weeks and then runs a project in week nine has not adopted project-based learning; it has appended it. In that structure the project is always the first thing sacrificed when the term runs short.
The part that does not change
There is a version of this argument that overshoots into claiming knowledge no longer matters — that because a machine can retrieve a fact, students no longer need to hold any. That is wrong, and the evidence from cognitive science has been consistent for decades: you cannot reason well about a domain you know nothing about. Working memory is limited, and expertise works by having enough in long-term memory that the working memory is free to think.
Project work is not an alternative to knowing things. It is the reason for knowing them. A student who needs to make a robot follow a line has a reason to understand a control loop that no amount of exposition will supply. The sequence matters: the problem creates the appetite, the instruction satisfies it, and the build consolidates it.
Where this leaves the argument
The case for project-based learning in a technology curriculum was always strong. What has changed is that the alternative has stopped working. When the production of plausible text was hard, asking for plausible text was a reasonable proxy for capability. It is not any more.
That leaves schools with a question that is easier to state than to answer: at the end of this year, what will a student be able to show us that they made?
If there is no good answer to that question, the technology programme is not measuring what it thinks it is measuring — and it was probably not measuring it before AI arrived either.