I have written before about sitting in the audience at conferences, watching AI demos compress weeks of work into hours and asking what speed costs. Course design raises the same question. It can be slow, deliberate work, and AI now promises to speed it up. As a learning scientist with expertise in human-machine interaction, I understand the stakes of that promise.
My team and I built Lazuli, an AI-powered course design tool grounded in more than a thousand research-backed learning principles, in collaboration with the Learning Design Alliance. I was one of the learning scientists translating those principles into what the tool would accomplish, rather than reviewing a finished build.
We piloted Lazuli with more than 700 WGU School of Business students. Completion rates held steady. Graduate learners finished about a week faster on average, and students told us the scenarios felt realistic and the pacing respected their schedules.
Then I presented Lazuli at the HACU International Conference, to a room of faculty, staff, and administrators from Hispanic-Serving Institutions (HSIs). The audience pressed back on issues a completion-rate chart cannot show, such as the impact of a student's relationship with the person teaching them. Their questions had me considering a learning science perspective, one where instructional quality depends on both sound design and human relationships. Technology can strengthen the former, but it cannot replace the latter.
The questions from my audience were challenging. They took me back to my time at other conferences, where I was an audience member rather than a presenter. Asking myself hard questions about speed and equity from row six at ASU+GSV was not the same as answering them onstage as a learning scientist, in front of the people who serve these students every day.
The audience handed me the corollary to my earlier argument. Speed is not neutral when agentic AI accelerates institutional decisions. It is also not neutral when we are the ones moving too fast to listen deeply. Listening does not scale. It is not supposed to. And by listening, I mean the sustained attention required to understand learners' lived experiences, not just their outcomes.
Learning scientists know the value of listening to students
My HACU presentation included student personas, which were composites of real learners: (1) Maria, a first-generation paraeducator working toward elementary licensure, and (2) Marcus, a certified nursing assistant pursuing his RN on night shifts while raising a two-year-old. I included these personas to acknowledge that HSIs already listen to their students with this level of specificity.
I also wanted to include underpinnings of what Gina Ann Garcia calls “servingness.” Servingness is not simply enrolling Hispanic students; it means intentionally designing institutions around students’ success, belonging, and strengths.
I sought to honor that HSIs have spent decades doing more than enrolling students like Maria and Marcus. I knew my audience does this active, transformative work daily. Part of that is listening, the slowest part of servingness—the part no tool can replicate.
Why Lazuli needs human expertise
Specifically, my audience asked questions about how humanity shows up inside Lazuli, the kind of humanity students need to feel seen by an educator:
- Can AI-generated content hold the cultural specificity that HSI students need?
- How do students build the relationships with educators that we know are central to their success when a technology is shaping the design of their learning?
- What separates a tool that supports a teacher from a tool that quietly displaces one?
Beneath each question was a shared concern about the limits of technology and the educational responsibilities that remain uniquely human. Each asked about something a tool cannot verify on its own. Lazuli can check whether an assessment aligns to an outcome in minutes. It cannot, however, tell whether a scenario reflects the lived experiences of the students in their classrooms. That’s where faculty and learning science expertise comes in.
How Lazuli prioritizes the expertise that matters
By handing faculty a starting draft, Lazuli gives them more time for the relational work that transforms student outcomes. This is the work of noticing, coaching, adapting, and encouraging students in ways that no workflow automation can accomplish:
- An office-hours conversation with Marcus about why night shifts and RN coursework aren't compatible this semester.
- Time to adapt a draft scenario because Maria's classroom doesn't look like the one suggested by Lazuli.
- The fifteen minutes an under-resourced faculty member did not have before, because a full redesign used to take months they could not spare.
It is up to institutions to choose whether to reinvest saved time in students.
That is the frame I left HACU with: what an educator can do once the mechanical work is off their desk. Educator expertise is not friction to be optimized away. It is the reason the tool exists at all.
Learning scientists belong on the teams building AI tools
In the wrong hands, AI tooling can become a cost-cutting story before it becomes a learning story. The institutions most likely to face that pressure are the ones serving students who have already been failed by efficiency-driven decisions in higher education, HSIs among them. History offers plenty of examples of efficiency becoming a justification for reducing human support rather than improving it.
For my team, resisting that begins with who is doing the building. Our learning designers, cognitive psychologists, and assessment researchers worked directly with engineers to shape Lazuli and ground it in learning design principles, rather than simply handing off requirements and waiting for the build. This kind of access is more possible now than it used to be, and more teams like ours should consider leveraging it so that the people who understand learners best are the ones building for them.
Resisting the pressure to prioritize cost-cutting over learning principles also means designing alongside institutions and learners. This is a posture HSI scholars have articulated for years: sharing rough drafts in front of audiences with the standing to challenge them; saying "we do not know yet" when we do not, and listening for what is and is not said.
During my presentation, the silences carried as much information as the comments spoken aloud. Who was not in the room? Those absences are often where equity challenges first become visible. Whose questions did not get asked? Which concerns were raised privately afterward rather than at the microphone?
How learning science keeps AI grounded
An AI-assisted process like Lazuli can now compress months of course design work into hours. The discipline is deciding what we refuse to compress: the conversations that uncover why a student is struggling, the revisions prompted by faculty who know their communities, and the slow listening that makes good design possible.
That is the case for learning scientists staying close to how these tools get built: not only because proximity makes the tool better-grounded in learning science, but because the most pressing questions in higher ed make it back into the next versions.
Dr. Jenn Killham is a critical media literacy scholar with expertise in human-machine interaction and emerging technologies that are impactful, equitable, and ethical. Her knack for seeing the big picture makes Jenn a go-to voice on early-stage strategic initiatives, particularly around AI adoption and implementation. Read more.
