The honest accounting: each class day is a roughly three-hour investment built around a single 45-minute session. That’s not a complaint—it’s the reality of teaching a course that doesn’t come from a textbook.

The hour before class is the most important. I’m reviewing what happened last time, checking notes on each student, and building a lesson plan that actually connects to where we left off. Not a template—a living document that reflects the specific six kids sitting in front of me and where they are today.

Then there’s setup time—getting into the room, getting oriented, making sure the technology is ready before the students walk in. Friction here costs real instructional time, and I’ve learned to treat setup as non-negotiable prep rather than an afterthought.

The wind-down afterward matters just as much. That’s when I jot quick notes while the class is still fresh—who said what, what landed, what missed. The longer I wait, the harder it becomes to attribute specific moments to specific students.

The first fifteen or so classes were primarily discussion-based—theology, ethics, how AI actually works, what dignity means in a technological context. Students were present, mostly analog, mostly listening and talking.

The final sessions shifted entirely. Students are now on their computers, building the AI-powered apps they’ve been designing. That transition introduced a category of friction I didn’t fully anticipate: technology setup at the start of class.

Content filtering systems on school networks do exactly what they’re designed to do—which means every subdomain, every authentication endpoint, every tool the students need may require a separate approval. A few minutes of clicking here and there adds up quickly when you only have 45 minutes and six students ready to work.

The lesson: pre-approve every site students will need before the build phase begins. Don’t wait until they’re sitting there ready to go. Map the tools, map the domains, handle the access requests in advance. That friction is entirely avoidable with process.

One practice I didn’t expect to value as much as I do: transcribing and reviewing class audio as quickly as possible after each session.

AI transcription tools are good but imperfect—speaker attribution gets confused, voices blend, off-hand remarks get misassigned. If I review the transcript the same day, I can catch those errors because the conversation is still in my head. Wait a week and it’s largely guesswork.

More than error correction, though, the transcript review does something else: it slows me down long enough to notice things I missed in real time. A student who seemed quiet in the moment turns out to have made three substantive contributions. A student who seemed engaged turns out to have been mostly restating other people’s points. The transcript catches nuance that the live experience doesn’t always surface.

Over time, reviewing multiple transcripts per student reveals arcs—not just snapshots. Who’s growing, who’s plateauing, who needs a different kind of challenge. That longitudinal view is only possible if you do the work of reviewing each session promptly and building on it.

The course is done. Twenty-four sessions, six students, one subject nobody had taught at this level before—and every kid made it to the finish line with something real to show for it.

Final individual summaries reflect not just where students landed, but the full arc of their engagement—how they entered the course, where they struggled, what they figured out along the way, and what they built. The work required to do this well didn’t happen in the classroom. It happened in the hour before, in the notes afterward, and in the slow work of actually paying attention to six individual kids across twenty-four sessions. That’s the job. And it was worth every minute of it.

I’ve thought a lot about which session I’ll remember longest. It’s not the theology discussion. It’s not the Galveston exercise, as good as that was. It’s the moment five of these six kids published something to the internet for the first time in their lives—and realized they’d done something they genuinely didn’t think they could do.

The excitement in that room was real. Not performed, not polite—real. Kids were leaning over to show each other their screens. One student kept refreshing their live URL just to confirm it was actually there. The energy was the kind you can’t manufacture with a lesson plan. You can only create the conditions for it and then get out of the way.

What happened after class was just as telling. Most of them asked if they could keep working on their projects over spring break. Several asked if I’d be teaching another elective. That’s the feedback that matters—not what they say in the room when you’re standing there, but what they reach for when the grade is already in and nobody’s watching.

I am ENORMOUSLY proud of these kids. Every single one of them surprised themselves. That might be the best thing a course can do.

Here’s something I didn’t fully see until the course was nearly over: the most important thing we were teaching wasn’t AI ethics. It was how to function at a higher level as a thinker and a person.

Middle schoolers don’t arrive knowing how to defend an idea under pressure. They don’t naturally know how to take critical feedback without deflating, or how to separate “my idea got challenged” from “I got attacked.” They haven’t had much practice sitting with an argument that’s going sideways and figuring out how to redirect it instead of abandoning it. These aren’t character flaws—they’re developmental gaps that most academic settings never directly address.

This course forced the issue. When you ask a student to defend a product concept against five pillars of evaluation—in real time, in front of peers, with AI also weighing in—you’re asking them to do something genuinely hard. Some handled it gracefully. Some needed help. All of them grew.

The shift from the discussion phase to the build phase made this even more visible. Talking about AI is relatively safe. Building something with it is not. The build phase required commitment, tolerance for uncertainty, and the willingness to keep going when the first version didn’t work. Those are executive functioning skills. They transfer everywhere.

If I teach this course again—and I want to—I’d make this more explicit from the start. The AI ethics content is the vehicle. Learning to debate your ideas, receive hard feedback, handle the friction of real creative work, and see something through to completion: that’s the destination. The two things reinforce each other. You can’t reason well about AI ethics without learning to reason well, period.