How we think about AI in learning
Nearly every organisation in education has an AI story now. Most of them are the same story, told with different logos. Ours is narrower and, we think, more useful.
AI is good at some parts of this work and bad at others
The difference is not subtle once you have done the work. AI is very good at the repetitive, rule-bound and unglamorous parts of building learning, and actively bad at the parts that need judgement about people.
We do not sell AI as a separate service, because we do not think it is one. It is an accelerant across strategy, design, build, delivery and platforms. A thing that makes the work faster and more thorough, not a thing you buy on its own.
Work that is repetitive and rule-bound
Checking hundreds of units against a quality standard. Finding the accessibility problems. Spotting where a stated learning outcome has no assessment behind it. Machines do not get tired on the four hundredth unit.
Work that unblocks a person
A first pass at a structure, a draft set of scenarios, a rough map of curriculum overlap. Not the finished thing. The thing that gets a designer past the blank page and into editing, which is where their expertise lives.
Work at a scale nobody would fund otherwise
Reviewing every course rather than a sample. Analysing feedback across a whole faculty instead of one program. Work that was always technically possible and never economically sensible.
Deciding what is worth learning
A judgement about a discipline, a profession and the people entering it. It is contested, it is political, and it should be made by humans who will answer for it.
Understanding the learner in front of you
A model can tell you what worked for a population. It cannot tell you why this cohort is disengaged this semester, and the answer is usually something no data set captured.
Anything where being confidently wrong is expensive
Clinical education, safety-critical training, regulated professions. The failure mode is fluent, plausible error, which is the worst possible one when someone will act on the output.
Buying AI licences isn't an AI strategy
Most institutions can now say how many AI seats they've bought. Far fewer can say whether anyone got better at anything. Access and capability are different purchases, and only one of them is on most budgets.
The gap isn't prompting, which people work out quickly. It's judgement: knowing what the tool is good for, where it fails quietly, and what you still have to decide yourself. That's learned, not licensed.
There's an equity problem coming too. The best tools sit behind subscriptions that well-resourced institutions can buy and smaller providers often can't. If AI capability follows the money, the technology most often called a leveller will widen the gap instead.
What to invest in next
A small group who get genuinely good
Give a few people time to get properly good, then have them teach the rest. Peer credibility beats a vendor webinar, and it stays in the building.
A clear line on where AI doesn't go
Staff hesitate most when nobody has said what's out of bounds, especially in clinical, safety-critical and regulated work. Publish the line so people can argue with it. Ours is on this page.
Follow-through, not a launch event
Six weeks after the training, someone should be helping with the specific thing that didn't work. Capability comes from ongoing support, not an onboarding session.
If AI can produce a passable version of your course in an afternoon, that says something about the course
A lot of learning content was always mediocre. It was assembled rather than designed, and it survived because producing it was slow enough that nobody compared it to an alternative. That protection is gone. We think this is good: it raises the floor and forces the interesting question, which was never how quickly can we build this, but whether it teaches anyone anything.
The commitments behind our own AI use
A person signs off everything
Nothing we produce with AI goes to a client or a learner until a qualified person has reviewed, checked and approved it. An AI draft is a starting point, not finished work.
We tell you up front
If we plan to use AI on your work, we say so when we agree the engagement, not in a footnote at the end.
Your information doesn't train anyone's model
We only use tools whose published terms rule out training on what we enter, and personal or confidential information stays out of AI tools unless you've given written approval.
Approved tools only
Every AI tool we use is classified by risk and passes a security review before anyone uses it for client work.
A trail you can inspect
Work of lasting value is kept in our own secure records system, with a history from AI-assisted draft to approved output. We can show you how something was made and who approved it.
Checked for bias
We check AI output for bias, especially in learning design and assessment. When in doubt, human judgement wins.
Judged on whether practitioners got better feedback, not on whether it used a model
The most useful AI work we have delivered was not a content generator. We built a platform for the University of Melbourne combining educational resources with AI feedback for practitioners in the disability sector, where Professor Keith McVilly described us as moving from consultants to an integral part of the national project team.
That project is a fair picture of how we work with AI: embedded in something with a clear purpose. It runs inside our products too. Scaffold Advisor checks course quality before publication rather than after a complaint, supporting our designers rather than replacing them.
This position runs through the rest of what we do: strategy, design, build, delivery and platforms. AI is threaded through all of it, not a separate offer.
What people usually ask
Do you sell AI as a service?
Will AI write our courses?
Where will you not use it?
How do you know it is actually helping?
Do you put our information into AI tools?
Will you tell us when you've used AI?
Read more on AI in learning
What we actually think about AI in learning design sets out the argument behind this page in full. The tool drafts. The human directs.
More of our writing on learning, technology and capability is in Insights.