Show and tell: How educators are putting agentic AI to work in the classroom

We're entering the show-and-tell phase of agentic AI in education, and it might be the most useful phase yet. After three years of predictions, frameworks, and policy drafts, the most valuable work is now happening in the open. Educators building AI agents, deploying them in actual courses, gathering evidence, and walking peers through what worked and what didn't.

Why you should pay attention to this phase

Show and tell isn't a new idea in education. What's new is the speed at which classroom practice is now ahead of institutional policy. While committees draft guidance documents, individual educators are running pilots that move quickly enough to learn something every term. The bottleneck has shifted. It used to be "what is this technology capable of?" Now, it's "what actually works in front of students?" Closing that gap is going to take more visible, shared practice across institutions, with the kind of detail that lives in classroom stories, not in slide decks.

A few events from the past year show what this can look like at scale.

The Cogniti mini symposiums at the University of Sydney

Cogniti is a platform built at the University of Sydney that lets educators create custom, steerable AI agents for teaching. It now has a community of educators across multiple universities, and once a year that community gathers to share what they've built. The Cogniti mini symposium runs in hybrid format, hosted by Sydney's Educational Innovation team, with the most recent edition held in November 2025.

What stands out is the specificity of what gets presented. It's not "how to think about AI in higher education," but "here is the agent I built to help my pharmacy students practice teamwork." Recent sessions have included AI agents for oral assessment preparation, scaling formative feedback in large cohorts, simulated patient interactions for clinical assessment skills, AI voice chat designed to reveal student thinking, and feedback at scale in radiology education.

The contributors come from University of Sydney, Monash University, Murdoch University, University of Newcastle, University of Saskatchewan, and others. The format is simple: build something, run it with students, gather what you learned, and share it with peers facing similar problems.

I recently sat down with Danny Liu, Professor in Educational Technologies and leader of the Cogniti.ai initiative at University of Sydney, and there are more exciting things to come from that team in the near future.

ASU's Agentic AI and the student experience

Arizona State University (ASU) ran a three-day summit in Tempe in October 2025 called Agentic AI and the Student Experience. It sold out at 500 attendees. The agenda was built around hands-on workshops as much as keynotes. Participants got a live demonstration of ASU's CreateAI toolset and tested adaptive tutoring prototypes and multi-step advising agents. ASU student workers presented an AI-powered chatbot they had designed and built to support prospective students through the admissions process.

The signal in that programming is worth naming. A major research university with the resources to host any kind of AI conference chose to centre the event on what its own students and staff had actually built. The closing message from the organisers was that the community is on its way to navigating the era of agentic AI, but success depends on doing it together. That's a show-and-tell ethic, applied at the level of a flagship event.

Call for submissions has closed for the event to be held on October 20-22, 2026, but registration is open, and the event promises to sell out again. 

What show and tell asks of the rest of us

The pattern across the Cogniti symposiums, the ASU summit, and other events is the same: Show your work. Let other educators see the specifics. Talk with each other about what didn't work as openly as what did, and make practice visible across institutional lines.

That ethic puts a useful kind of pressure on leadership. If the most valuable knowledge about AI in education is being generated in classrooms and shared at events like these, then the leadership question shouldn't be "what is our AI strategy" in the abstract. It's "what are we doing to support, surface, and learn from the practice that's already happening on our own campuses, and what are we doing to connect it to the practice happening on others." For any institution serious about AI, that should include funding educators to attend these events, giving them time to write up what they're learning, a treating cross-institutional sharing as a core capability, not a nice-to-have.

The show-and-tell phase won't last forever. At some point this practice will settle into curricula, into platforms, and into established models. The window we're in right now, where the field is still figuring itself out in public, is the one where the most generous contributions get made and the most learning happens. The educators leading these efforts are already doing their part. Now the rest of the work is showing up, paying attention, and bringing what we learn back to our own institutions.

If you're working on something interesting in this space, the most useful thing you can do is share it. Submit to a symposium and even apply to speak at it. Write up what you're trying and talk with colleagues at other institutions about what they're seeing. Heck, shoot me an email with what you’re doing - rlufkin@instructure.com. The conversation gets better the more voices contribute.