This week many of our colleagues in education gathered in Denver, Colorado for the EDUCAUSE annual conference. Regulars at this long-running industry conference know the week really hosts two parallel events at once. One happens in the formal sessions, the other happens over coffee or lunch, before anyone's badge scanner comes out.
This year, that second conference began with a CIO breakfast Instructure hosted on Wednesday morning. My colleagues Armin Molavi and Zach Pendleton joined me, along with Dr. Humberto Ariza, who directs augmented intelligence operations at Bay Path University, and Matt Leger, an industry analyst who spends his career across the table from CIOs working through technology budgets. We spent the hour on a question nobody has fully answered yet: what is AI actually buying higher education, and how do we prove it?
The hour flew by and surfaced a theme that's been circling our own strategy conversations all year. The number worth defending to a board isn't the dollars a tool saves. It's the capacity the tool frees up, and what an institution chooses to do with that capacity next.
The case for return on augmentation.
Dr. Ariza made the sharpest argument on the panel. Return on investment, he said, is the wrong lens for AI in an academic setting. His alternative: return on augmentation.
"There is time saved, but there is also time reallocated," he told us. At Bay Path, that reallocation is deliberate and tracked. Time faculty once spent grading now shows up as more office hours, more outreach to students who've stopped logging in, and more advising touches before a student disappears for good.
A few minutes later, he borrowed a line from a product designer he'd talked with: "I don't look at AI as an end-to-end solution. I look at it as a middle-to-middle solution," with humanity on both ends. I told him I love that he leads a team with "augmented" in the title, not "artificial." The win isn't the automation. It's what a person does with the room the automation creates.
Why chasing AI for its own sake often backfires.
Matt helps CIOs make exactly these calls, and he kept steering us back to the J-curve. Every technology transition dips before it climbs. You spend money and effort learning a new tool before you see any return, and higher ed sits near the bottom of that curve for AI right now.
His advice was direct. Quantify the actual problem before you reach for AI as the fix. AI solutionism is its own trap, because a lot of campus friction is a process problem, and a better workflow can solve it without any AI at all. His example: students juggling 20 different logins just to register for a class and pay a bill. That's a measurable cost before AI ever enters the room. It's also a much easier case to bring to a budget committee.
Zach offered the builder's version of the same caution. Inside Canvas by Instructure, his team starts every AI decision with the problem it solves. The feature that gets demoed comes second. He calls the result point-and-click AI: a button that gets a rubric started instead of leaving you with a blank page. His line to every campus he visits is that his AI roadmap shouldn't be anyone else's AI strategy, because no two institutions know their students and faculty the same way.
He also flagged a backlash starting to show up in the data. People feel comfortable with AI in low-stakes moments like tutoring. They feel uneasy the moment it replaces a human connection they expected to have. A student who learns an instructor is using AI to grade every assignment feels like they lost something. A student whose instructor adds a new AI-assisted checkpoint that wasn't possible before feels the opposite. Same technology, opposite reaction. The difference is whether it extended a relationship or stood in for one.
What actually earns trust on campus.
The best evidence at the table came in two stories.
First, Dr. Ariza described building a course in 18 minutes with IgniteAI, a feature within Canvas, for a case study he was presenting later that day. A colleague built one in 30. He's taught long enough to know what a course normally costs in hours, and that gap is what got his board to approve the tool. A slide about efficiency didn't.
Second, he described a course accessibility and remediation tool that started to soften the most committed AI skeptics on his faculty. These were tenured professors who'd happily call themselves anti-AI, until they saw what the tool did for alt text and layout, on work nobody wanted to do by hand anyway, in service of reaching more of their own students.
Notice what's missing from both stories: usage dashboards. The CIOs at the table kept asking for what Dr. Ariza calls "the qualitative gates." What did a student say? What did a faculty member say? That kind of testimony is something a prompt count can never capture.
His change-management model runs on three words: awareness, access, and adoption. It's built on research suggesting it takes about 11 hours of structured exposure before someone trusts a new tool enough to use it well. Much of that trust gets built away from leadership entirely, in collaboration spaces where staff can admit how they're really using AI without anyone from the top of the org chart in the room.
The fatigue nobody's measuring yet.
At one point, a CIO told us she's now doing the work of five people, because AI handles pieces of everything. Then she asked, honestly, why that doesn't feel like freedom.
The answers have taken up residence in my head. Removing the small, mindless tasks also removes the brain's only break in a workday. The complex work that fills the space is exhausting in a way busywork never was. Someone raised the reverse-centaur risk, where the tool starts running you instead of the other way around. Someone else offered a simple countermeasure: keep a literal not-to-do list, and automate whatever lands on it the moment you have room.
Advisors came up as the clearest version of the problem. Students who can handle things on their own start routing to AI. The students left for a human advisor are disproportionately the hardest cases, and that takes a different skill set than the job used to require.
We also spent a few minutes on the headlines, because they shape so many conversations before we get the chance to. Someone had recently heard a tech executive predict that 90% of white-collar jobs would be automated within 18 months, said out loud to a room full of people whose jobs that would include. Most of what gets called an AI efficiency gain in a layoff announcement is better described as a layoff with an AI label on it. Separating the two has quietly become part of the job.
Mark Watkins at the University of Mississippi has a line that feels right for where higher ed actually sits: "AI is not inevitable, but it is unavoidable."
Why these conversations matter.
Nobody at that breakfast walked away with a formula. We didn't solve the measurement problem. We didn't agree on a single definition of return. And that's exactly the point.
The future of higher education won't get decided in a product launch or a keynote. It'll get decided in rooms like this one, where a CIO, a faculty leader, an analyst, and a few people who build the tools can say what's working, what isn't, and what they're afraid of. Those conversations are where an institution figures out whether AI will stand in for its people or give them more room to show up. They're also where the questions of AI literacy, privacy, security, and accessibility stop being abstract and start being budget lines and policy choices.
So here's my ask. Don't wait for the formal session. Pull your CIO, your provost, your faculty, and your students into the same room, and ask the uncomfortable question out loud: what are we going to do with the time AI gives back? Then write down the answer, and hold yourselves to it.
That choice belongs to us. Let's make it on purpose.
