Teachers have been asking the same thing for three years: what are they supposed to teach now?
Since 2023 we have worked on AI adoption with teaching staff and students at 15+ universities.
In most cases it is worth starting from the same place — and it tends to die in the same place. This page describes both.
We start with a teacher ready to change their course.
The courses where AI worked shared one thing: the teacher wanted to change something that was failing in their teaching today.
That is why most institutions begin with a workshop. It surfaces the few teachers who already have a course they would change.
The first weeks establish where AI can help. The necessary starting point is the teacher's intent.
An administrative process
Where the same question comes back every week.
An integration
Something that is manual copying between two systems today.
What makes a good first use case
A good first use case centres on a problem worth solving in its own right. AI makes that solution feasible.
What matters is how much AI takes away from the thing being learned.
The same technology does something different in each discipline. The situation becomes acute in proportion to how much of the work of learning the model now does in the student's place.
This is the first thing we establish, because it decides where to begin — and which faculty still has time.
Programming and computer science
This is where it is sharpest. The thing the student is supposed to practise is exactly what the model does for them, correctly and instantly. Teachers here are asking what the weekly assignment is still for.
Law and text-based disciplines
What teachers tell us is that students no longer read the material. The reading was the learning, and the summary is good. The input to learning itself is at risk here.
Engineering and lab-based disciplines
Concrete structures, lab practicals, field work: these disciplines still have time to prepare for the change ahead. That is the rarer position.
Why we start with a faculty diagnosis
All three situations exist inside the same university at once. A single policy or a central tool rollout treats them identically, when one of them needs an answer this semester and another has two years.
We look first at what happened to the learning, faculty by faculty. The institutional strategy follows from that diagnosis.
A fixed decision point keeps adoption moving.
An initiative stops when the day never comes for someone to review what became of it.
A date, a measure and a reporting obligation give the initiative its next decision point. Without them, it goes quiet by the next semester.
In practice
We fix the measurement point and the owner in the first week.
Every pilot starts with a measurement point because it prevents a supported initiative from dying quietly.
First we look at what happens today.
Discovery leads to the technology decision.
What discovery settles
- what the task is, and who does it today
- what counts as a good answer, and who decides that
- which materials and systems are involved
- who can grant access, and to what
- when, and by whom, success will be reviewed
Result
A prioritised use case and a pilot plan with an owner, a measurement point and a date.
A pilot runs with real users.
A pilot with real users reveals how the solution is actually used.
What a pilot contains
- a working solution ready for use
- real members of its intended user group
- permissions from day one
- one measurement point and one date
- a channel where people answer back
The measurement point is part of the pilot from day one.
What is worth watching depends on the use case, but on the fixed day the decision is always the same three: carry it on, reshape it, or close it.
The evidence determines whether the pilot continues, changes or closes. Closure is a result too, and cheaper than maintaining something for a year that nobody uses.
What we look at
- whether people use it unprompted
- what they actually use it for, and how that differs from the intended use
- whether it was worth what it cost
The current limit of measurement
We can see what happens during the semester. One teacher moved the weekly assignments far enough that students hand in good work, and the logs show how they got there.
The relationship between behaviour during the semester and exam results remains an open question. The evidence so far shows no demonstrable link between the two.
Scaling connects solutions that already work.
What works, we widen: more courses, more faculties, a shared skill library, SSO, further integrations.
Separate solutions should connect where that creates a real advantage.
Principle
An institutional AI capability grows from a few working solutions that know about each other.
People, preparation and the missing pieces.
Workshops, training, advisory, integration and development are described on the Solutions page. What matters here is when each one joins the path.
Control is part of the design.
Who can access what, what the AI may do on its own, where human approval is required, what gets logged. These are settled in the pilot's plan from the start.
From one intent to institutional practice.
Intent
A teacher wants to change something: students ask the same questions every week, and it has been like this for years.
Pilot
An AI tutor is built for one course from the teacher's own materials, with a measurement point and a date.
Use
We look at how students use it, what they ask and where the system gets stuck.
Decision
On the fixed day it is settled: carry it on, reshape it, or close it.
Spread
Further courses join, using a template that already exists. This is when SSO and the LMS connection arrive.
A first step that works is what you can learn the most from.
Do you have a teacher who would change something in their own subject? We start with them.
The first conversation establishes how we can begin together.