The pattern that AI has produced in student life by 2026 is consistent across surveys, classroom observations, and what students will tell you in conversation. Artefacts have improved. The thinking habits the artefacts were supposed to develop have not. Grades are still happening; the learning that grades used to represent is harder to pin down.
This post is for students who want to use AI without slipping into the version of student life where the AI does the work and nothing gets learned. The published evidence is on the cautious-pragmatist side, not on either of the extremes.
Why this matters — the published evidence
The 2025 EDUCAUSE Review piece on the "productivity paradox of AI" is the cleanest single reference. It names the pattern: AI-assisted students produce better artefacts and develop weaker thinking habits in parallel. The artefact-and-thinking gap is the load-bearing observation.
The Conversation's 2023 piece on ChatGPT and student writing reported the same direction early in the AI-in-classroom timeline. AI-assisted students often produced more polished work and developed weaker motivation to think for themselves. The qualitative read foreshadowed what later surveys would find.
The 2024 Springer overview of AI overreliance in education (Springer Nature) covers the broader transfer. The mechanism is the same as in the general cognitive-offloading literature.
The 2025 MIT Media Lab preprint by Kosmyna et al. (arXiv 2506.08872) is the cleanest recent empirical reference on the writing side. It measured EEG, recall, and self-reported ownership across essay writers using brain-only, search-engine, and LLM-assisted conditions. The LLM group showed the weakest neural connectivity during composition, the lowest sense of ownership, and the worst recall of their own arguments. The Stanković 2026 commentary is a preprint critique of the methods — both are preprints, neither is the final word — and they are best read together. The direction is consistent with the older cognitive-offloading literature, and the broader argument does not depend on any one paper's exact numbers.
The deliberate-practice literature (Ericsson, Krampe & Tesch-Römer, 1993) is the constructive frame. Specific cognitive skills respond to daily, effortful, on-purpose engagement. For students, the relevant skills are the ones the curriculum is trying to develop.
The artefact-and-thinking paradox
The simplest way to describe what AI has changed for students: the artefact you submit and the thinking you did to produce it have come apart. A polished essay used to be reasonably good evidence that some writing-and-thinking had happened. In 2026, it is evidence that some artefact-production happened — which may or may not have involved the thinking the artefact was designed to develop.
This is not a moral story about cheating. Most students using AI on coursework in 2026 are not cheating in any cynical sense. They are using a tool that produces a better artefact than they would have produced alone. The artefact is what gets assessed. The grade rewards the artefact. The cognitive act the artefact was designed to develop is the thing that quietly goes missing.
The honest framing: it is not the AI that does the harm. It is the workflow where the AI substitutes for the originating cognitive act. If you do the originating act and the AI handles the polish, the learning happens. If the AI does the originating act and you handle the prompt, it does not.
A study-habits playbook
The playbook below is built for actual student life — a real workload, AI tools available, social pressure to use them. None of it requires you to stop using AI. All of it requires you to keep the originating cognitive acts on your study timetable.
1. Draft first. AI joins later. Open a blank doc. Write what you actually think before asking AI for help. The first version does not have to be good. The cognitive act is what counts. AI then enters as editor, structurer, length-adjuster, alternative-phrasing generator.
2. Read one full source per assignment without an AI summary. Not every source. One. The act of reading the source unmediated — sitting with the actual text, noticing what is and is not there — is the cognitive act AI summaries flatten out.
3. Reason out an answer before asking AI to check it. When you have a problem to solve, articulate your reasoning before opening AI. Two or three sentences. Even if your answer is wrong, the originating act has happened. AI is the check, not the originator.
4. Make AI explain rather than answer. When you do use AI, ask it to explain the concept or critique your reasoning rather than produce the answer. The cognitive load stays on you. The AI becomes the tutor, not the homework-doer.
5. Re-read your work a week later. Once a week, look at something you wrote or solved a week ago. Can you still follow the reasoning? Could you do it again without AI? If the answers are no, the AI integration has gone too far for the learning to stick.
6. Protect one no-AI study block a week. Two hours, on the calendar, with no AI tools open. Read, reason, write, calculate — whatever your week needs — without the tool in the loop. The block does not need to produce a shippable assignment. The unmediated study has to exist somewhere on the week.
The shape that works for most students is one block, one obvious time slot, repeated until it is the default. A Sunday afternoon. A Tuesday morning before the first class. The specific time matters less than the predictability — what is on the calendar at the same time every week tends to survive the busy week; what floats around tends to get displaced. Pair the block with a single source you actually want to read end-to-end, or one problem set you want to work through unmediated, and the block has a shape rather than being two hours of vague intention.
What this is not
A few honest disclaimers, because the AI-and-students category is louder than the data.
This is not a case for refusing AI. The published evidence does not support an abstinence position and most students will use AI tools regardless. The recommendation is calibration — keep the originating cognitive acts on the timetable, use AI for everything else.
It is not a claim that AI use causes lasting cognitive harm to students. The published evidence does not support that claim and Senwitt does not make it. The narrower claim — that cognitive acts which get fewer reps tend to weaken, and that AI in student workflows shifts which acts get reps — is what the evidence supports.
It is not a substitute for school-level or course-level policy. Whether your course permits AI, how assessment is structured, and what your university's honour code says are all separate questions. The playbook is what an individual student can do regardless of how those questions are answered.
How Senwitt fits
Senwitt is a daily seven-minute Set across six Senwitt Skills — Writing, Math, Code, Memory, Reading, and Reasoning. For students it sits alongside actual coursework. It is not a study aid; it is a daily practice on the underlying cognitive surfaces.
The for students page lays out the case in more detail. The research/your-brain-on-chatgpt page covers what the cognitive-debt research actually supports with the Stanković critique alongside the Kosmyna preprint.
