Hosted by Jackrabbit Learning Experience (JLX) in October 2025, the AI Forum for Learning Professionals is a series of webinars that brings together educators, instructional designers, technologists, and learning leaders to explore how artificial intelligence is reshaping the way we design, deliver, and imagine learning.
Visit our AI Forum page to recap other brilliant sessions. To continue these conversations, join us in the AI Forum Slack community.
If you wish to receive updates about our next AI forum, please subscribe to our newsletter now.
Session 5: AI Founders on the Future of Our Work
April 30th, 2026
In this terrific session, we had the great opportunity to talk to two founders in ed-tech with two very different products, but one shared conviction: AI is only as good as the human system it serves.
If you missed our previous forum recaps, you can find them below.
Who Was in the Room
Diane Weaver is the co-founder of Baryons, an AI mentorship platform that gives every employee a personalized AI mentor, available by phone, in the flow of their day. You call in, you talk, and the system remembers. It can send you a summary afterward, help you identify your top priority for the day, or walk you through a reflective checkout at the end of a shift.
Louis NeJame is the founder of Bevel, an AI workspace for instructional design teams in higher education. Bevel plugs into Canvas, Blackboard, Moodle, and D2L, letting teams audit courses against quality frameworks, run accessibility updates, and make large-scale revisions while keeping humans in control of every change. Louis also leads AI product development at McGraw-Hill.
Jason Gorman facilitated, with time for questions from the full group.
What Users Are Actually Doing with AI
When Bevel looked at what people were really using the tool for, the answer was not content generation. It was the boring stuff. Fixing header hierarchies for screen readers. Running accessibility passes. Mapping against accreditation standards.
“Boring is a term of endearment. It’s important, but tedious, operational, administrative work.” — Louis NeJame
Louis was blunt about the alternative: generating a full course from a single prompt with no context is a losing game right now. The models pull toward the mean. But for precise, well-defined tasks? The tech is already working very well.
Diane saw a similar pattern from a different angle. Users were talking to their Baryons mentor while driving, walking the dog, running errands. The phone-based interface unlocked something: people using moments of physical movement to decompress and think out loud with a system designed to help them do it well.
The Real Killer Feature Is the Guardrail
Louis told a story that reframed the fear around AI and jobs. One of Bevel’s first customers was an institution with about eight instructional designers and 2,000 online courses in rough shape, many rushed online during COVID. They had an accreditation visit in two months and the math simply did not work. Bevel helped them review 1,800 courses against their standards. Ten months later, the team is hiring. The tool revealed so many issues that faculty demand for instructional design support surged. One attendee responded: “So the tool actually created more work for the humans, not less?” Exactly, Louis said.
That led into a moment that stuck. A colleague at a conference asked what Bevel’s killer feature was, expecting “generate a course in a minute.” What Louis pulled up was the diff view: a side-by-side comparison showing exactly what the AI wants to change before any edit goes through.
“He was just so befuddled that it was these very guardrail-heavy functionalities that were really resonating with people, as opposed to the turbocharge Ferrari features.” — Louis NeJame
In education, in healthcare, in any high-stakes context, a small weird AI change inside a course is not a small deal. The review step is the line you do not automate.
Diane built on this with this framing: human-in-the-loop does not just happen. It has to be intentionally designed. And when it is, it serves two purposes at once:
-
-
- It provides the regulatory oversight AI needs
- It sharpens the humans doing the work, reinforcing expertise and keeping critical thinking skills active
-
“If our human-in-the-loop is designed for sharpening our humans and for regulatory oversight, that’s when the magic is going to happen.” — Diane Weaver
From Technology Users to Technology Stewards
The expectation that AI would “save us time” has largely not played out. There is no extra time. There is just more work.
Diane introduced a reframe: we are shifting from technology users to technology stewards. That changes the role, the responsibility, and the kind of capacity we need to build. Working through different perspectives with an AI mentor before a meeting means you arrive having already stress-tested your thinking. If an entire team does that, the conversation in the room changes.
The Skills That Will Survive
When the conversation turned to what will matter most going forward, two themes came through clearly.
Louis on the relational and the creative: The hardest part of doing a quality review with a faculty member was never the Bloom’s alignment. It was reading the room. Noticing discomfort. Building trust. And then there is the creative signature: the unexpected simulation, the interaction nobody else would have designed. Those are hard to replicate with next-token prediction.
Diane on knowing what you don’t know: With AI so eager to affirm your thinking, it is dangerously easy to go down a rabbit hole of “wrong.” The people who will thrive are those who recognize when they have hit the limits of AI and reach out to a human expert.
“I think of it less like a sandwich and more like making a croissant. You’re taking the dough, putting the butter in, doing multiple turns to get that perfectly flaky result.” — Diane Weaver on moving between AI and human collaborators
A practical rule of thumb she offered: if you have been working with your AI for more than 30 minutes straight, stop. Do something else. Call someone. Come back with fresh eyes.
What Is a Fool’s Errand
An AI Forum attendee cut straight to it: “While AI isn’t quite the wild west it was two years ago, it’s still relatively undefined in education. Where are you seeing meaningful progress, and what is a waste of time and effort?”
The answers were sharp:
-
-
- Fool’s errand: Generating full courses from a single prompt with no context. The quality is not there yet.
- Fool’s errand: Using AI as a punitive “gotcha” tool to catch students or employees. If we position ourselves as stewards of AI, we cannot simultaneously use it to diminish people.
- Real progress: Stacking context (outlines, OER, interviews, standards) and letting AI work within those boundaries.
- Real progress: Precise, well-defined operational tasks like accessibility compliance and bulk formatting changes.
-
Who Is Driving Adoption
An attendee posed the question directly: “Is this being driven organically from the ground up, or is it coming from leadership? Who’s making these decisions?”
The answer from both founders: both, but the two groups want different things.
Instructional designers who find Bevel first get excited about never doing tedious accessibility work again. Administrators looking at it from above want cross-program evaluation and quality metrics at scale. Louis learned that the admin-level view closes the deal, while the designer-level experience makes people love the product.
As for whether leadership has a clear AI strategy: the honest answer from both founders was usually no. The strategy is rarely fully baked. There are assumptions, urgency, and a willingness to fail fast. But the clarity most teams need is still being built in real time.
“Even in very modern workforces where they move very quickly, the AI strategy is not fully baked.” — Diane Weaver
Looking Ahead: What You Can Do Now
The conversation after the formal session moved into genuinely forward-looking territory. One participant raised a question that got everyone leaning in: “At what point does this stop being a series of one-offs and become an actual system? How do we get to a headless content architecture where universities can store, remix, and redeploy their IP at scale?”
That question opened up a rich thread. Once you build a deep knowledge base of source material, creating different versions and formats becomes almost infinitely possible. Louis described Bevel’s near-term path: taking a 10-week course and condensing it into a 3-week microcredential, AI handling the first draft while a designer provides the judgment. Diane brought it back to the human layer: in workforce learning, people forget 80% of their training within 30 days. The answer is not more content. It is more relevant, more personalized, more continuous support.
If there is a through-line from everything discussed in this session, it is that the learning professionals who will thrive are the ones making deliberate choices right now about how they work with AI. Here is what that looks like in practice:
-
-
- Use AI for the tedious, well-defined work first. Accessibility fixes, formatting passes, compliance mapping. That is where the tech is strongest today, and where it frees you up for the work that actually requires your expertise.
- Stack your context before you prompt. A single prompt with no source material produces mediocre output. An outline, some OER, a few interviews, and a clear framework? That is where AI starts earning its keep.
- Design the human-in-the-loop, do not assume it. Every AI workflow should have a deliberate review step that keeps the human sharp, not just checking a box.
- Set a 30-minute rule. If you have been going back and forth with an AI for more than half an hour, stop. Call a colleague. Get a different perspective. The rabbit holes are real.
- Know what you do not know. AI will happily affirm your thinking all the way down the wrong path. Build the habit of reaching out to human experts, especially when the AI is telling you exactly what you want to hear.
- Think about your institutional IP differently. What are the evergreen pillars of content your organization owns? Once you identify those, the possibilities for remixing, repurposing, and personalizing multiply fast.
-
The session closed the way the best AI Forum conversations do: not with tidy answers, but with shared direction and something concrete to try on Monday morning.
The AI Forum for Learning Professionals meets monthly. These sessions are not recorded so participants can speak freely. If you want to be in the room for the next one, you can join us on Luma. And if the conversations between sessions matter to you as much as the sessions themselves, join us in the AI Forum Slack community.