Twenty stations, with close to 5,000 educators moving through them over the course of a day. At nearly every single stop, the same sudden, audible “oh.”
That's how Dale Leszczynski describes RMIT University's make-and-take session, a room built for one purpose: get educators playing with generative AI instead of talking about it. Some built a game with AI. Others polished up a slide deck. At each station, Leszczynski watched the same two-beat moment repeat itself: first someone saw a colleague try something with the technology and went wide-eyed, then, about five minutes later, tried it themselves and wore the same look. “If we could bottle all those aha moments into one little place and sell it,” he said, “that’s the magic of education right there.”
Leszczynski is head of AI education at RMIT University in Melbourne, Australia, where he leads the Generative AI Lab for Education, known internally as GAILE. Under his direction, GAILE launched a new AI skills framework for university educators. Every aha moment in that room came from a different starting point. Some educators walked in having never touched a large language model. Others had already built AI-powered simulations for their classrooms. That gap, and what to do about it, is the work that Leszczynski is doing in education. With a field changing this fast, giving 5,000 people at wildly different starting points a way to know where they actually stand matters more than teaching them another tool.
Find your starting point on the AI skills continuum
The tool GAILE built to answer that gap wasn't really a tool at all. Leszczynski and colleague Toni Jones (RMIT's lead education gen AI specialist) built the Gen AI Skills Continuum, published at aiskontinuum.com and on RMIT's own site.
Rather than a single bar educators either clear or don't, the Continuum lays out a progression. Leszczynski describes it in three stages: introducing, adapting, and mastering. RMIT's published version breaks that arc into five stages nested across three broader phases, moving from basic AI awareness through to leading strategic adoption within a department or discipline. It was built as a continuum instead of a fixed benchmark on purpose. In this field, nobody stays right for long. “Everything that’s right today is wrong tomorrow,” he said.
The framework was built for educators first, but something unplanned started happening in classrooms almost immediately. Educators began flipping the Continuum around, using it to decide which AI skills to teach their own students directly, sometimes turning it into an informal assessment: can a student explain why they picked one AI tool over another before they lean on it?
Leszczynski is candid about the limits of a map, though. “The honesty about the Skills Continuum is that we've given people the map, and there's a bit of a sigh of relief that someone has the map,” he said. “And then maybe they put it down.” A map only does its job, he added, once someone actually picks it up and asks what the terrain means for their own classroom.
Match your confidence to your actual practice
There's a cultural backdrop to why RMIT is this comfortable naming what it doesn't know yet. Leszczynski likes to point out that Australia, by his own admission a small country, punches above its weight in behavioral science and educational psychology, and RMIT leans on that local expertise rather than importing someone else's playbook. He also says the most popular phrase in the room lately isn't some AI buzzword. It's “I was wrong, and I don't know.”
That's the temperament he brought to a roundtable GAILE convened in June 2026 with senior leaders and digital experts from every Victorian university, in partnership with Amazon Web Services, to discuss the new framework and where the sector actually stood. Leszczynski ran an exercise using the Continuum: sticky notes, mapped against the stages, showing where each institution believed it sat.
He expected the room to cluster around the middle of the scale. Instead, the sticky notes piled up at the introducing stage, even in cases where the practices people described suggested they were much further along. “Some of the sticky notes that we're seeing appear around the introducing stage, we're actually probably more faced towards that reinforcing or that mastery stage even,” he said. “There's some really good practices coming through. But the reality is that people don't have the confidence yet, because they're kind of looking at others and going, actually, I think this is what we're doing here. Is this good? Do we know this?”
Part of the problem, Leszczynski said, is that no shared benchmark exists yet for what an AI-expert educator even looks like. Anyone claiming that title in public is doing so without much to measure it against, and the ground underneath that claim keeps shifting.
The room was full of chief information officers and other technology leaders with a clear interest in the tools themselves, which made the session's dominant theme notable. Attendees kept returning to pedagogy before technology. “The tail isn't wagging the dog,” Leszczynski said. “The dog is leading here.”
His caveat: the balance won't hold on its own. “Always more work to go on that one, because the pendulum will swing the other way as technology gets more and more exciting,” he conceded.
Give students the consistency they keep asking for
GeneratED also gave students a formal seat in the conversation. A student panel drew a standing-room-only crowd, including RMIT's vice chancellor, who showed up to find people sitting on the floor. GAILE followed that panel by launching an AI student advisory board, made up of students spanning business, design, ethics, and STEM disciplines, from undergraduate through postgraduate.
What Leszczynski heard in that room surprised him. Students raised concerns about their own learning unprompted, specifically around what researchers call metacognitive offloading, the tendency to hand off planning, monitoring, and self-checking to an AI tool rather than doing that reflective work themselves. It's an active area of study: a 2025 framework published in Frontiers in Education examines how AI tools can either erode or support self-regulated learning depending on how they're designed into a course, and a 2026 scale published in the Journal of Educational Computing Research gives the phenomenon a formal name: metacognitive laziness.
Across every discipline represented on the advisory board, one request cut through everything else: not more rules, just consistency, even while the guidance underneath those rules keeps shifting. “There's a desire for clarity around what is right, what is wrong,” Leszczynski said. “That is always going to be a gap. It's going to be difficult to fill. People want the clarity. Everyone wants the understanding.”
One thread students raised: some students are consciously choosing not to use AI tools at all, tied to concerns about unequal access. It's a documented pattern beyond RMIT, too. Researchers writing on AI equity in higher education point to gaps in who can afford premium AI tools and who arrives at university with prior AI exposure at all, disparities that existing access gaps in higher education can easily deepen rather than close.
The focus in higher ed should be purposeful AI use: using a tool deliberately, for a reason he could explain out loud, rather than reaching for it by default. That's the throughline underneath the make-and-take session, the sticky notes, the debate stage, and the advisory board. AI keeps expanding what it can do. Leszczynski's real question is what humans choose to keep doing themselves. What do humans get to keep when AI takes on more and more?
Partway through the recording, his dog started barking at the door. An inherited Samoyed the family calls Aunt Lillian needed letting in, and Leszczynski excused himself for a few seconds to handle it himself. Nobody suggested an AI agent for that job. Some things, purposeful or not, still need a human at the door.
Hear the full conversation on the No Country for Fast Answers: Safeguarding Critical Human Logic in a World of Instant AI Responses episode of the EduCast3000 podcast.
