Prompt Party number three: the agent, the TA, and the “Kyle problem.”

Renee Carney

Three sessions in, and the community is just getting warmed up.

Prompt Party number three is in the books, and the pattern is holding: educators showing up, opening their actual Canvas by Instructure courses, and showing the rest of us what they’ve been quietly figuring out. If you’re catching up, the May 21st recap is right here, and it’s worth the read.

This week brought three demos, one bonus workflow from the audience, and a sub-account question that turned into a building plan. Here’s what went down.

Rick built three modules with one prompt

Rick Upchurch , Associate Vice President for eLearning and Educational Innovation at Belhaven University, kicked things off with refreshing honesty about Ignite AI agent by Instructure.

“The first iterations of Ignite were less than impressive, but they have evolved pretty significantly to the point that we are finding value in them now.” — Rick Upchurch, Belhaven University

Then he showed us why. With one prompt, Rick asked the agent to create three modules, add a list of headers inside each, and drop blank headers in between. The agent walked through every step, asked for approval at each addition, and built the scaffolding.

For faculty starting a brand new course, or for anyone who doesn’t want to copy an old shell, this is a real shortcut. As Rick put it, it’s a solid leg up without the manual jinking around and missing little details.

The audience question that landed: could one well-built module become a template the agent duplicates with new names? Fewer approvals, same structure. Emily Koutsoyannis raised it. Rick liked it. We’re all going to go try it.

The honest aside

Daniel Guillot named what every demo viewer was quietly thinking when the agent asked for approval on every single header.

“Bulk approval is one of the most highly requested features right now.”— Daniel Guillot, Instructure

Message received.

Daniel differentiated a calculus lesson on the fly

Daniel Guillot, Solutions Engineer at Instructure, pulled up a calculus lesson (with an apology to anyone who has feelings about Snell’s law) and asked the agent to create a differentiated version with additional scaffolding and support, while keeping the original HTML formatting roughly intact.

Daniel’s been in the early adopter program since October, which gave him a useful long view:

“Early on, it definitely timed out a lot and crashed a lot. But it has gotten so much more reliable and so much faster now.”— Daniel Guillot, Instructure

The new lesson came back with hint call-outs, step-by-step instructions, and worked examples his struggling students didn’t have access to before. Was every detail perfect? No. The agent got a little creative with Snell’s law in a way Daniel hadn’t taught yet. But that’s the point of his favorite framing:

“The agent is a nice TA. It’s a very enthusiastic TA, but it does need some oversight.”— Daniel Guillot, Instructure

Easier to delete than recreate. That’s the whole shape of how this changes the job.

Kyle asked the agent who was struggling, and got a plan

Kyle Cole, Sr. Solutions Engineer at Instructure and an adjunct at a small state college in Florida, brought the prompt he uses consistently every semester. Three asks in one go: check recent assignment grades for students below 70 percent, identify the weak concepts and skills, and draft a message to each group offering concrete next steps.

If you’re an instructor, you already know the manual version of that workflow. Open assignments. Cross-reference the gradebook. Jump back to modules to check the outcomes. Build a spreadsheet. Compose messages. Send. An hour, easy.

Kyle’s prompt came back with grouped students, identified weak spots, and personalized outreach drafts. Then he said the line that stopped the room:

“It’s a great way to call yourself out and give yourself a pulse check. Sometimes it’s not a student problem, it’s a Kyle problem.”— Kyle Cole, Instructure

That’s the part of the demo nobody scripts. The agent flags patterns. The instructor decides whether to retool the lesson, expand office hours, or look harder at how they grouped students in the first place. As Kyle described it, the tool is helping him see blind spots he was missing in his students’ journeys, and in his own teaching.

Bonus round: Emily moved an entire course pacing plan

During the questions, Emily Koutsoyannis dropped a workflow that earned an audible reaction. She’s been demoing in a course with course pacing enabled, where every student is on an individualized schedule. Her three-prompt chain:

First, ask the agent how far off track each student is. Second, shift the due dates and restructure the pacing plan so the whole course resets around where students actually are. Third, send an announcement letting them know.

“That would take me multiple clicks, multiple pace plans, all of that to kind of run through to manually adjust. And it was seconds to do all of that.”— Emily Koutsoyannis

Renee Carney named what Emily had really built:

“What you defined really is your magic TA, isn’t it?”— Renee Carney, Instructure

The sub-account question we’re all going to chase next

Rick raised the question every admin wants answered: blueprint courses set a beautiful navigation panel, but the second you sync to a live course, the order goes wherever it wants. Could the agent reach into a sub-account and fix navigation in bulk?

Kyle’s answer was honest. The agent has access to the API. The admin-level orchestration isn’t fully there yet. But the door isn’t closed.

Daniel ran with the same energy in a different direction. Upload a course outline or syllabus, point the agent at it, and ask it to skeleton out modules and pages from the chapter and section structure of your textbook. The kind of build that used to require a template, a checklist, and a quiet week.

Kyle’s next experiment (steal it)

Before we wrapped, Kyle floated a prompt he hasn’t built yet but wants to. A “what did I miss” recap. How many missing assignments. What needs grading. Which students haven’t logged in. A scoped objective list for the hour and a half he has in the course that day.

“When you teach a course, sometimes you get into your course and you get squirrel brain. And next thing you know, you’re grading, and next thing you know, you’re creating a module.”— Kyle Cole, Instructure

If you build that one, post it to the library before Kyle does.

The Prompt Library is growing. Get in there.

Rick gave us the move of the week. He challenged all 10 instructional designers on his team to post a prompt to the library. Many of them already have. That’s the pattern we want to see spread.

Vote on the ones that helped you. Comment on the ones you tried. Post the ones that are working in your courses. We’re tracking what’s landing, and there are little prizes going out for May’s top contributors. Catch up on the original Prompt Party announcement here.

Prompt Party is going beyond May - and watch for APAC hours!

Three sessions in, and it’s clear this isn’t a one-month thing. Prompt Party rolls on. If you’ve got a prompt you want to demo, reach out. If you just want to watch other people break things and figure out what’s possible, we want you in the room.

See you next week.