We built this AI curriculum for working HBO-ICT professionals. But once we
looked hard at what is actually in those courses, something stood out:
little of it is ICT-specific. And our way of teaching is not ICT-specific at
all. What is here is yours to use.
why we rebuilt it
The hybrid world of human and agent
Under the influence of AI — and of agentic AI above all — the role of the
employee is changing. Someone who used to do the work now directs a crew of
agents: setting the task, weighing the result, stepping in where it goes
wrong, and staying accountable for what comes out. That is not an existing
job with a new tool bolted on. It is a different job.
A programme that merely tweaks its courses in response is already behind. So
we did not add a course. We rethought the whole chain — from who we send out
into the world, through what we must teach them and how our own lecturers
make that shift themselves, to the material they stand in front of a class
with. Each of those four steps has its own method.
step 1 · the profiles
Orgith
The name comes from the Greek organon — tool, instrument. The
word Aristotle used for his own instrument of thought. The only one of
the four that isn't Latin, and rightly so: where Nascith and Crescith
describe what happens, Orgith names the instrument. Orgith is the
organising of people, work and agents into one coherent system.
Using the Orgith method we redefined the programme's exit profiles,
explicitly for this agentic reality. Not: which tasks will a graduate
perform. But: which tasks does she still do herself, which does she hand
to agents, and where is the line at which she must be able to intervene.
A different picture of the profession follows from that distinction — and
so does a different profile.
The name comes from the Latin nascī: to be born, to arise, to
grow, to become. It is a deponent verb — passive in form, active in
meaning — and that fits: becoming does not happen to a programme, a
programme does it. Nascith is about that emergence — not education
adjusted at the edges, but education born again.
With Nascith the curriculum was then redesigned around those new profiles.
Deliberately, this was not a clear-cut. Every course was traced back to its
core value — what it genuinely contributes to the profile — and that core
was kept. What had accumulated around it out of habit and history was let
go. The curriculum shifts without the programme losing its foundation.
From the Latin crescere: to grow, to increase, to gather
strength. The same word that still sounds in the waxing moon, the
crescent — and that is exactly the movement: not larger all at
once, but steadily increasing. Crescith is growth through change — the
capacity of people and organisations to grow along with agentic AI.
A redesigned curriculum is worthless if the lecturers do not carry it.
Crescith is how we determined the shape of the change process for them —
the management of change. Who has to learn what, in what order, with what
support, and how you deal with the lecturer who watches his subject change
under his hands. This is the part that usually gets skipped.
The name comes from the Latin scribere: to write. Originally a
rough word — to scratch, to incise, to cut lines into wax or stone. Writing
was an act first and a product only much later. That fits: Scribith is not
about texts that are finished, but about carving out again and again what a
subject has to convey at this moment. Alone among the four it names
not a movement but a craft.
What a lecturer uses in class — the reader, the case, the exam — was written
before the agentic reality. Not wrong, but dated. With Scribith the lecturer
rewrites that material himself: the existing reader and sources go in first,
the structure stays intact, and the AI guards what a person loses over a
hundred pages — consistency, build-up, terminology. What comes out is not a
file but a source: reader, presentation or workbook, in whatever form the
lesson needs. Otherwise the whole change stays on paper.
"Not one more course. A different profession to prepare people for."
the courses
AI has stopped being an ICT question
The nurse who wants to know whether an AI triage model can be trusted, the
logistics planner watching a forecast drift, the communications adviser
caught out by a hallucinated source — those questions resemble each other far
more than they resemble anything technical. They are about data that does not
hold up, models you cannot fully see into, and who carries responsibility
when it goes wrong.
That is exactly what our courses are about. They were written for people with
a working practice, not for people with a computer science background. You can
put them in front of another programme's students without stripping them down.
A single elective
Take one course — data literacy, say, or AI ethics — and offer it as an
elective. The smallest step, and one your students see immediately.
A minor
Combine a handful of courses into a coherent minor for your own domain.
We will think along with you about which courses suit which professional
practice.
A full learning track
Build AI into your curriculum structurally, from first encounter to
graduation project. This is what we did — and where we learned the most.
Here AI is not just the subject — it is the medium
This is the part we most want to pass on, because it is the part most
programmes still have ahead of them. With us, AI is not only what the lesson
is about; it is also what the lesson is taught with. In every class students
work with bots: bots that explain, give feedback, push back, and drill with
them until it sticks.
This is not a gadget bolted onto existing teaching. It changes what fits into
an hour. And it has one effect any programme director grasps immediately: the
rate at which students build skills goes up — not by a little, but by an order
of magnitude.
One bot per learning task
Not a single do-everything assistant, but twenty-five types that each do
one thing — twenty for the student, five for the lecturer. Theory bots that
explain the material at your level. Socratic bots that pointedly withhold
the answer. Feedforward and rubric bots. Role-play and oral exam bots.
Reflection and team bots. Build the bot around the learning task rather than
the other way round, and you get teaching that works.
Skills finally get time to practise
This is where the real acceleration sits. Presenting, facing down a
sceptical regulator, convincing a board member — those were always the
things a student got to attempt twice a year, because a practice audience
costs people. Role-play and voice bots make repetition free. What used to
be two attempts is now twenty.
Feedback before submission, not after
Feedforward bots point out what could be better while the student can still
act on it. Every draft becomes a moment of learning instead of a hand-in.
That is the engine behind the portfolio: many small corrections, and no
single decisive exam.
The lecturer is freed up to judge
What a bot can do — explain, repeat, drill, check the form — the lecturer
no longer has to. What remains is exactly where she is irreplaceable:
weighing whether it is correct, whether it is sound, and whether it is good
enough for the practice she still works in herself. That does ask something
of lecturers, and we set up a separate track for it.
There is a second layer underneath. Our graduates will have to direct agents —
set the task, weigh the result, intervene. They learn that by working with
agents themselves for three years. The form of the teaching is already the
profession.
"The curriculum can be copied. The acceleration is in how you teach."
For us a bot is not a pick from a list but a combination of three independent
axes: what it does pedagogically, how you talk to it, and which course data it
reads. The axes multiply — which saves dozens of near-identical types and keeps
the whole thing explainable to a lecturer.
Axis
Answers
Values
Bot type
What does the bot do pedagogically?
25 types — 20 for students, 5 for lecturers
Modality
How do you communicate with it?
textspeechboth
Source
Which course data does it read?
CanvasBooksHBO-ISharePoint several at once is possible
Example. A spoken mock oral exam for Research Skills that also
consults the HBO-I competency matrix is an oral exam bot with modality
speech and source HBO-I — not a separate type. Every bot hangs
off a single course offering, has its own system prompt that the lecturer revises
and versions, and a moment at which it opens up to students.
Twenty types for the student
Ordered by the moment in the learning process. Each type has an explicit
boundary: what the bot deliberately does not do. That
boundary is pedagogical, not technical — it is there so the student keeps doing
the learning.
These five are not meant for students; they support the lecturer's design work.
The boundary always sits at the same place: the bot delivers a draft, the
lecturer stays responsible.
Design assessment items that match a learning outcome and the intended Bloom level.
Boundary: delivers draft items; the lecturer remains responsible for validity, standard setting and deployment.
How you use it
You give the learning outcome, the intended Bloom level and the assessment format; the bot returns draft items, each with the reasoning for why it measures at that level. It knows the learning outcomes and the material of your own course offering, so the items are about the content you actually taught. What you end up with is a working stock to pick from, rewrite and discard — not a ready-made exam.
What that sounds like
You: “Three multiple-choice questions for learning outcome 3, at application level.”
Bot: “Here are three drafts. Watch question 2: it really asks for recall — the distractors can be ruled out without applying the rule. Shall I tighten it?”
Why this works
Most assessment work is not in inventing questions but in guarding that they measure what the learning outcome promises. A bot that spells out that link per item forces the conversation at the moment you can still change course.
More about this bot type →
Build a rubric from the learning outcomes, with criteria and distinguishing level descriptors.
Boundary: sets no standard or cut score and does not decide which level counts as a pass.
How you use it
You enter the learning outcomes and the type of product — report, design, code, presentation — and the bot proposes criteria with level descriptors that genuinely exclude one another. Where two levels come down to the same thing in practice, it says so rather than inventing a difference. Where the cut score sits is yours to decide.
What that sounds like
You: “Make a rubric for the research report.”
Bot: “The criterion ‘use of sources’ is in four levels. Between ‘good’ and ‘excellent’ I found no distinction a second marker would also see. Do you want the distinction on the number of sources, or on how critically they are weighed?”
Why this works
Rubrics fall over on level descriptors that differ only in the head of whoever wrote them. Anyone who cannot write the distinction down cannot mark it consistently either — and two markers then arrive at two grades.
More about this bot type →
Check constructive alignment: do the learning outcome, the teaching format and the assessment format line up?
Boundary: flags misalignment and explains why, but decides nothing and rewrites nothing.
How you use it
You put the block design in front of it: learning outcomes, teaching formats, assessment formats. The bot checks whether they line up and reports where it chafes — a learning outcome at creation level closed off with a multiple-choice test, a teaching format that practises nothing of what is assessed, a learning outcome that no longer returns anywhere. You get signals with the reasoning attached; the rewriting stays yours.
What that sounds like
You: “Is the alignment of block 3 sound?”
Bot: “Learning outcome 4 asks students to ‘design a data pipeline’. Designing does not appear in the teaching formats — there are three exercises analysing an existing design, and the assessment is a case analysis. As it stands, designing is promised but neither practised nor assessed.”
Why this works
Misalignment creeps in: a learning outcome is adjusted, the assessment stays put. It rarely shows while writing and almost always while marking. A systematic check up front costs half an hour; afterwards it costs a resit round.
More about this bot type →
Design teaching formats and lesson plans per class or block, fitted to the learning outcome and the contact time.
Boundary: accounts for the contact time and invents no sources, literature or tools.
How you use it
You give the learning outcome, the contact time and the group size; the bot proposes a lesson design with teaching formats, timing and what students need to do beforehand. It does the arithmetic on time — a format that does not fit the hour does not go in. This is the only type allowed to search the web, and only to check that a source it wants to cite actually exists.
What that sounds like
You: “A 90-minute seminar on feature engineering, 24 students.”
Bot: “Proposal: 15 minutes plenary intro, 40 minutes in pairs on their own dataset, 20 minutes exchanging in groups of four, 15 minutes plenary close. With 24 students the exchange costs more transition time than you expect; if it gets tight, drop that step and not the pair work — that is where the learning sits.”
Why this works
A lesson plan almost always dies on time, not on content. By handing over the contact time as a hard constraint, the conversation goes straight to what does fit and what you give up for it.
More about this bot type →
Summarise frequently asked questions from chat logs into a teaching signal for the lecturer.
Boundary: works only in aggregate and never reports anything traceable to an individual student.
How you use it
After a teaching week you look at what students asked the bots of your course — in aggregate, never per student. You see which concepts keep coming back, where questions stay stuck and which part of the material clearly did not land. That is material for the next class, not for a grade.
What that sounds like
You: “Where did week 2 get stuck?”
Bot: “Forty per cent of the questions were about the difference between validation and test set, spread across the whole week rather than one moment. Notably, almost nobody asked anything about the assignment itself. That looks like a comprehension problem, not an unclear brief.”
Why this works
What students do not understand, they usually do not ask in class. It does show up in what they ask a bot on a Thursday evening. In aggregate that is the most honest signal you can get about your own explanation — and the anonymisation is precisely why those questions get asked at all.
More about this bot type →
File upload — nine types accept a submitted file, because they
discuss the student's work: feedforward, rubric, peer review, writing, code
review and defence bots, and on the lecturer side rubric design, curriculum and
analytics bots. Web search — only lesson design bots search the web, so they can
check a source before naming it.
"Every bot has a boundary. That boundary is pedagogical, not technical."
the toolset
The Applied AI Workbench
A bot per learning task sounds appealing, but it stands or falls on who
builds those bots and who keeps them alive. That is why there is a complete
toolset underneath our approach: the Applied AI Workbench. It is where you
build the bots and agents for every course — from teaching bot to
role-play bot — and where you keep them running afterwards.
A lecturer builds a bot there without programming, tests it before any
student sees it, and puts it in the classroom in a single move. Not loose
prompts in a chat window that nobody can find again, but bots with an
owner, a version, and a place in the curriculum.
Build it per course
Every bot is tied to a course and its learning outcomes, fed with the
programme data and course material that already exist. It answers
within the frame of that course.
Test before the class
A bot only enters the classroom once it has been tried out. In the test
environment you put the hard questions to it and adjust — not after
thirty students already got a strange answer.
Lifecycle management
Courses change, so bots change with them. Versions, ownership and
retirement are built into the tool. A bot that no longer holds up
disappears, instead of drifting around for years.
Deployment
From built to available-in-class is one step. The bot shows up where
the student already works, with the right permissions, without a
lecturer having to install anything.
Hooked into your LMS
The Workbench integrates fully with an LMS such as Canvas. Courses,
modules and course material come from there, so when the material
changes the bot changes with it. No second place to keep everything up
to date.
The Applied AI Workbench: programme data, building bots, testing, deploying and managing them — in one environment. Build. Test. Orchestrate. Deliver impact.
"A bot per learning task only survives if there is a workshop underneath it."
contact
Write to Mark or Wiemer
The two of us started this programme and we teach on it ourselves. If you want
to take over a course, set up a minor, or simply think out loud about getting
AI into your curriculum — write to us. A first conversation does not come with
a quote attached.
