AAC&U Conference on AI and Higher Education · October 28, 2026

Software on Demand

A replicable model for purpose-fit AI in higher education.

Roy Magnuson · Director, Adaptive Edge Institute · Illinois State University

Follow along at aei-web.apps.paas02-t.ilstu.edu/aei-software-on-demand-story.html
Press → or click to begin
The design question

Work backwards from the person.

Not “What can AI do?”

“What is this person trying to accomplish, and what’s in their way?”

The tool serves the vision. The vision serves the human.

Every piece is built backwards from the people who’ll play it and the room they’ll play it in.

Distributing licenses is a starting point.

Calling it an AI strategy is a category error.

The failure mode

Solving an organizational design problem as a procurement problem.

The people who want to do this work are already at your institution.

The infrastructure you need is probably already there.

The question is permission and alignment, not acquisition.

The model

Five moves, in order.

  1. Start with a real, specific problem.
  2. Use infrastructure you already have or can procure cleanly.
  3. Find the staff who want to do this. They exist.
  4. Build tight scope.
  5. Evaluate against the original problem, not the platform.
Case 01 · Teaching

By Design

A course-specific interactive study guide. One course. One instructor.

Not a generic chatbot. A tool shaped by specific course content, pedagogical intent, and instructor judgment.

9,000+ messages before the first exam.

The infrastructure is replicable. The judgment that shaped it was human.

Case 02 · Operations

Untitled PDF Tool

Accessibility remediation.

One accessibility burden. Hours of repetitive lift.

Upload. Review & edit. Download.

One person freed to make decisions that require judgment.

AI proposes. Humans decide.

The tool removes the tedium; the expertise stays in the room.

Case 03 · Governance

Policy, made queryable.

CBA. ASPT. Workload Policy.

Documents that govern everything. Navigated by everyone, understood by almost no one at speed.

AI doesn’t read policy for you.

It gives the right people faster access to what they already needed.

Who got that time back? What did they do instead?

That’s the question.

The pattern

Same model. Three domains.

What’s replicable is the method, not the artifact.

In each case: what was the actual problem, and who owned solving it?

  • By Design. Generic AI doesn’t know the course. Build a tool that does.
  • PDF Tool. Accessibility labor is a human sink. Remove the repetitive; keep the expertise.
  • Policy Querying. Policy governs everything. Few can navigate it at speed.
What it requires · 1 of 3

A question before a solution.

Risk appetite. Leadership that can say “we’ll try something that might not work.” That permission doesn’t need to be large. It just needs to be real.

Requirement 2 of 3

People.

Staff who want to do this are already at your institution, doing adjacent work informally. You don’t need to hire. You need to point.

Requirement 3 of 3

Infrastructure.

If you have an AWS, Microsoft, or Google relationship, start there. You probably don’t need to buy the repackaged version.

Discussion

Let’s talk.

Question 1 of 3

What’s one process at your institution that everyone knows is broken, and why hasn’t AI touched it?

Question 2 of 3

Has your institution purchased an AI tool in the last 18 months that repackaged capabilities you already had through an existing contract?

Question 3 of 3

Where does resistance to AI implementation actually live at your institution, and is it really about the technology?

Adaptive Edge Institute · Illinois State University

Roy Magnuson, Director · rdmagnu@ilstu.edu · adaptiveedge.illinoisstate.edu