30 June 2026 · 1 min read
Assessment was already broken
Generative AI did not break the take-home essay. It made visible a set of design problems that markers had been absorbing quietly for two decades.
By Nick McIntosh
The most common framing of the last three years is that assessment has been disrupted. A more honest framing is that assessment was already under strain, and a very cheap text generator removed the last of the slack.
What actually changed
The cost of producing plausible prose fell to roughly zero. That is the whole change. Every downstream problem — unverifiable authorship, inflated marks, students learning less than their transcript claims — follows from tasks that only ever measured the output.
Three things that still work
- Assessment where the process is the evidence: drafts, annotated decisions, supervised iteration.
- Assessment tied to a specific local context that a model has no access to.
- Assessment conducted live, where a person can ask a follow-up question.
All three cost more staff time than a marked essay. That is the real conversation, and it is a resourcing conversation, not a technology one.