The Assignment Dial
Stop writing one AI rule for everything. Set a value on one real task, in four sittings, using paper you already have. About an hour and three quarters total.
Why this kit exists
Most AI policies are one page long and apply to every task at once. That is why nobody follows them. A single sentence cannot govern a lab report and a reflective essay without either permitting everything or breaking one of them.
Two universities published the fix this year. Neither of them wrote a better one page rule. Both of them stopped writing one rule at all.
The evidence, in three lines.
Kaśka Porayska-Pomsta, director of the UCL Knowledge Lab, in Times Higher Education, July 30, 2026: universities have “moved too quickly and at the same time not quickly enough,” adopting technology “that’s not really backed with evidence and best practices,” and creating implicit pressure “that everyone should use generative AI regardless of context.”
University of Manchester, Senate approved April 2026: four values applied per assessment, AI Prohibited, AI Minimal, AI Permitted, AI Required. Dr Mark Carrigan on what it replaced: “students are effectively left to figure it out for themselves when presented with inconsistent guidance.”
Stellenbosch University declined a single institution wide rule, because “a programming task, laboratory report, literature review, design project or professional placement may each require a different approach.” Academics decide against four principles: authenticity, fairness, accountability, transparency.
A dial has settings. A switch has two, and most policies are wired as a switch. This kit turns one task into a dial setting you can defend, and then makes the reasoning portable so the next task takes ten minutes instead of a semester.
How to use this kit. Work one phase per sitting. Type directly into the fields. They save nothing and send nothing, so use Copy everything at the bottom when you are done and paste it wherever you keep your notes.
Name what the task proves
Nobody can set a value on a task whose purpose is unwritten. This is where most policy work skips a step and never recovers. Before you decide what AI may do here, say out loud what the finished work is supposed to demonstrate about the person who made it.
Stellenbosch’s test, in one question: does the finished work still amount to “a valid representation of what the student knows, understands and can do.” Swap in employee, analyst, or yourself. The question holds.
Set the value
Four values, one choice. The discipline is not the picking. It is writing the reason next to the pick, because a reason you can write is a reason you can defend when somebody pushes on it, and a reason you cannot write means you chose from habit.
Watch for the reflex. If every task in your head lands on Prohibited, you have a switch, not a dial. Manchester built a scale that can say yes on purpose. That is what makes the no credible.
Write the line they actually read
Your policy document is not where the question gets asked. The question gets asked at eleven at night by somebody staring at the task itself. Whatever you write here is the only version of the rule that will ever be consulted, so it has to survive being read once, quickly, by a tired person.
Read fields 8 through 11 aloud. If you stumble on a sentence, so will they. Rewrite the sentence, not the person.
Do a second task, then hand it over
One setting is an opinion. Two settings side by side is where your own reasoning becomes visible, and where a colleague can pick it up and use it. This is the sitting that turns an afternoon of thinking into something your department can run without you in the room.
Field 14 is the deliverable. Manchester gave unit leads a four value scale and let them classify their own work. You just did the same thing on a smaller field, and you have the reasoning written down instead of carried around.
Bring your field 14 to the community.
The decision rule you surfaced is the interesting part, and it is the part nobody publishes. Post it, read three others, and watch how fast your next task takes ten minutes instead of an afternoon.
Harvest Kernel · Plant Ideas. Cultivate Skills. Harvest Results. · Read the article this came from · All resources
Sources: Times Higher Education, July 30, 2026 · EdTech Innovation Hub on the University of Manchester and on Stellenbosch University.